Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
A METHOD AND AN APPARATUS FOR SEARCHING OR COMPARING SITES USING ROUTES OR ROUTE LENGTHS BETWEEN SITES AND PLACES WITHIN A TRANSPORTATION SYSTEM
Document Type and Number:
WIPO Patent Application WO/2019/164727
Kind Code:
A1
Abstract:
Embodiments relate to searching or comparing sites. One embodiment is a real estate search-or-compare method based on commute durations. The method efficiently processes public transportation and real estate property data to compute the durations of travel between the real estate properties and the vehicle stops. These durations are stored. A request framework is introduced that allows to express a wide range of search-or-compare requests. During request processing, the method identifies parts of the commute paths that depend on any real estate property. Because durations for these parts were precomputed and stored, the method can determine commute durations to every real estate property in a scalable manner. As a result, the method rapidly responds to requests within the real estate market of one of the largest metropolitan areas in existence today. Other embodiments in- elude: searching or comparing based on a monetary cost, transportation using private cars, and sites other than real estate properties. A computer system and a computer service also embody the method.

Inventors:
MALEWICZ, Grzegorz (Alabastrowa 56, 25-753 Kielce, Kielce, PL)
Application Number:
US2019/017909
Publication Date:
August 29, 2019
Filing Date:
February 14, 2019
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
MALEWICZ, Grzegorz (Alabastrowa 56, 25-753 Kielce, Kielce, PL)
International Classes:
G06F16/29; G06Q50/14; G06Q50/16; G06Q50/30
Domestic Patent References:
WO1999014701A11999-03-25
WO2017065431A12017-04-20
Foreign References:
US20130046795A12013-02-21
US20110251789A12011-10-13
US20140107921A12014-04-17
US20130197794A12013-08-01
US8417409B22013-04-09
US8738286B22014-05-27
US8756014B22014-06-17
KR101692501B12017-01-03
KR2016001083W2016-02-01
US20180232824A12018-08-16
KR101905593B12018-10-08
Download PDF:
Claims:
1155 CLAIMS

1156 (Antecedent basis is tracked with boxes: a term is later used as the term], mostly.)

1157 1. A method for searching or comparing at least one site using

1158 at least one route or route length between the 1 at least one site 1 and

1159 at least one place within a transportation system the ]_method_] comprising:

1160 (a) receiving a plurality of sites wherein each site is included in the

1161 transportation system , the ' plurality of sites ' including the ]_at least one site ; ;

1162 (b) receiving at least one representative , wherein each representative is included in

1163 the transportation system ;;

1164 (c) wherein

1165 each hrst precomputed route or route length comprises a description of travel

1166 within the ' transportation system ' between a site included in the ' plurality of sites '

_ _

-

1167 and a representative included in the ' at least one representative

1168 (d) receiving a request comprising the ' at least one place wherein each place is

1169 included in the ' transportation system the ' at least one place ' including

1170

1171 (e) computing the ' at least one route or route length

1172 wherein each route or route length comprises a description of travel within the

1173 t _ransportation system ' between a site included in the 1 - at least one site 1 and a

-

1174 place included in the ' at least one place

1175 wherein the ' computing ' comprises:

1176 i. determining at least one nearby representative , wherein each

1177 nearby representative

1178 A. is included in the ' _ at least one representative and

-

1179 B. wherein a length of travel between the nearby representative · and a

_

-

1180 start/end place included in the \ at least one start/end place \ is within a

_

1181 threshold; and

-

1182 ii. for each · nearby representative · included in the

_

-

1183 ' at least one nearby representative retrieving from the 1184 at least one precomputed route or route length stored in the [_database_] a

1185 second precomputed route or route length between the

1186 ' nearby representative ' and a not nearby site , wherein the ' not nearby site < is

1187 included in the < plurality of sites and

1188 (f) responding to the [ request ! with information comprising a representation of the

-

1189 at least one route or route length >.

1190 2. The ]_method_] of Claim 1, wherein

1191 (a) a site included in the ]_at least one site represents a real estate property, and a

-

1192 place included in the ' _ at least one place ' represents a commute destination; or

1193 (b) a site included in the |_at_ le_a_st_qne sitej represents a commute destination, and

-

1194 a place included in the ' at least one place ' represents a real estate property.

1195 3. The !_method_j of Claim 1, wherein the request further comprises at least one of:

1196 (a) a hrst departure time from a second site included in the ]_at least one sitej;

1197 (b) a hrst arrival deadline at a third site included in the 1 at least one site

1198 (c) a second departure time from a second place included in the

1199 at least one place ; or

(d) a second arrival deadline at a third place included in the ' at least one place '

4. The ]_method_] of Claim 1, wherein the ' at least one place ' includes:

(a) for each i in the range 2 < i < k— 1, for a threshold k that is even, a

1203 place Pi

1204 wherein the at least one start/end place includes:

-

1205 (b) a start place c or an end place for the 1 threshold k

1206 wherein the computing the > route or route length further comprises:

1207 (c) receiving a hrst site or a last site , each included in the 1 at least one site 1

1208 and 1209 (d) using:

- - i. a description of travel from the 1 - hrst site 1 to the ' _ start place Pi ', or a

- - description of travel from the e _nd place L for the 1 - threshold k to the -

1 last site and

1213 ii. for each even i in the range 2 < i < k— 2, for the Threshold k_ j, a

-

1214 description of travel from the ' place Pi ' to the place Pi+ .

- -

1215 5. The ' -method ' of Claim 1, wherein the ' _route or route length ' involves two or more

-

1216 sites included in the 1 _ at least one site '.

-

1217 6. The ]_method_] of Claim 1, wherein

1218 (a) the ' at least one route or route length ' includes a

1219 site-place route or route length ; further comprising:

- -

(b) storing the site-place route or route length in the ]_database_]; wherein the _ computing further comprises:

- -

1223 (c) retrieving from the |_database_] the ' site-place route or route length

1224 7. The method of Claim 1, wherein

1225 (a) the plurality of sites > includes a cluster site , the |_cluster site representing

1226 one or more site included in the ' plurality of sites ', the |_o_ne_ or _mor e _site

1227 clustered based on at least one of:

1228 i. a length of travel; or

1229 ii. a geographical proximity, wherein the |_o_ne_ or _mor e _site further comprises

1230 a geographical location;; or

1231 (b) the ' at least one representative ' includes a cluster representative the

1232 cluster representative ' representing one or more representative included in

1233 the _ at least one representative ', the ' o _ne or more representative ' clustered

1234 based on at least one of: 1235 i. a length of travel; or

1236 ii. a geographical proximity, wherein the ' o _ne or more representative ' further

1237 comprises a geographical location.

- -

1238 8. The 1 method 1 of Claim 1, wherein the ' plurality of sites ' includes

1239 at least one site connector ;;

1240 wherein

1241 (a) the ' not nearby site ' is included in the ]_at_ least _one sit_e_ connector and

1242 (b) the computing ' the ' route or route length ' further comprises:

r - n

1243 i. using a description of travel between the not nearby site and a

1244 zeroth site , wherein the 1 zeroth site 1 is included in the 1 at least one site 1

1245 9. The !_method_j of Claim 1, wherein a compression list comprises one or more

1246 ' _ hrst precomputed route or route length and the ' _ storing ' in the 1 - database 1 the

-

1247 ' _ at least one precomputed route or route length ' further comprises compressing the

-

1248 ' compression list

1249 10. The ]_method_] of Claim 1, wherein

1250 (a) the storing ' in the the

-

1251

_ at least one precomputed route or route length ', or

1252 (b) the _ computing ' the ' _ at least one route or route length

1253 further comprises:

1254 (c) receiving a manipulation function , and

1255 (d) applying the ' manipulation function ' to: the

-

1256

_ at least one precomputed route or route length or the

-

1257 ' at least one route or route length

1258 11. The method of Claim 10, wherein the ' _manipulation function ' comprises

1259 setting to inhnity a hrst route length included in:

1260 (a) the ' at least one precomputed route or route length ', or 1261 (b) the at least one route or route length >;

1262 the ' first route length ' being between a hrst real estate property and a

1263 hrst school , wherein

1264 (c) the ' hrst real estate property ' is included in the ]_at least one site ,

- -

1265 (d) the |_hrst school· is included in the ' at least one place and

- -

1266 (e) the ' hrst real estate property ' is outside of a zone of the |_hrst school·.

1267 12. The [_method_j of Claim 1, further comprising:

1268 (a) receiving data about the transportation system , comprising:

1269 i. a plurality of transportation elements ; and

1270 11. at least one segment length each segment length describing a length of

1271 travel from a hrst transportation element to a

1272 second transportation element , each transportation element included in

1273 the ' plurality of transportation elements ';; and

1274 (b) creating a graph comprising a plurality of graph vertexes including:

1275 i. a plurality of transportation element vertexes each transportation

1276 element vertex representing one of the

-

1277 plurality of transportation elements >;

1278 n. a plurality of site vertexes , each site vertex representing one of the

1279 plurality of sites ; and

1280 111. at least one representative vertex each representative vertex

1281 representing one of the > at least one representative >;

_

1282 the graph > further comprising a plurality of graph edges, including a

1283

1284 the

1285 graph edge > comprising one of:

_

-

1286 iv. the graph edge source vertex comprising a

1287 hrst transportation element vertex included in the 1288 ' plurality of transportation element vertexes the

_

-

1289 ' graph edge target vertex ' comprising a

1290 second transportation element vertex included in the

1291 ' plurality of transportation element vertexes and the graph edge weight

_ _

1292 representing a segment length included in the

-

1293 ' _ at least one segment length the segment length describing a length of

1294 travel from a transportation element being represented by the

-

1295 ' _ hrst transportation element vertex ' to a transportation element being

-

1296 represented by the ' second transportation element vertex

- -

1297 v. the graph edge source vertex and the > graph edge target vertex >

1298 comprising a hrst graph vertex and a second graph vertex m some

1299 order, the ' hrst graph vertex ' being included in the

_

- -

1300 ' plurality of site vertexes ' and the second graph vertex > being included in

_ _

- - -

1301 the plurality of graph vertexes , and the graph edge weight describing a

_ _

-

1302 length of travel between a site being represented by the ' hrst graph vertex '

_

-

1303 and an entity being represented by the > second graph vertex >; or

_

- -

1304 vi. the ' graph edge source vertex ' and the ' graph edge target vertex '

1305 comprising a third graph vertex and a fourth graph vertex m some

1306 order, the third graph vertex > being included in the

_

- -

1307 ' _ at least one representative vertex ' and the ' f _ourth graph vertex ' being

- -

1308 included in the ' plurality of graph vertexes ', and the ' graph edge weight

_ _

1309 describing a length of travel between a representative being represented

-

1310 by the third graph vertex and an entity being represented by the

_

-

1311 ' fourth graph vertex

1312 wherein the ' hrst precomputed route or route length ' comprises a representation of:

1313 (c) a graph path or graph path length between a representative vertex included

1314 in the ' _ at least one representative vertex ' and a - site vertex included in the

-

1315 plurality of site vertexes .

1316 13. The method of Claim 12, wherein the plurality of graph vertexes > includes a

_

1317 graph vertex that represents at least one of: 1318 (a) a transit station of a public transportation vehicle,

1319 (b) a stop of a vehicle,

1320 (c) a turn of a vehicle,

1321 (d) a stop of a walk, or

1322 (e) a turn of a walk; or

1323 wherein the ' _ plurality of transportation elements ' comprises a transportation

1324 element further comprising a time , and a graph vertex included in the

1325 ' plurality of graph vertexes represents the time

-

1326 14. The method of Claim 12, wherein the ' gr _aph edge weight ' represents at least

1327 one of:

1328 (a) a number of transfers among public transportation vehicles,

1329 (b) a duration of travel,

1330 (c) a monetary cost of travel,

1331 (d) a duration of wait,

1332 (e) a monetary cost of wait, or

1333 (f) a distance of travel; or

1334 wherein the ' at least one segment length ' comprises a segment

1335 length further comprising a transportation departure time from the

1336 ' hrst transportation element ' or a transportation arrival time at the

1337 ' _ second transportation element and the ' gr _aph edge weight ' describing a

-

1338 length of travel that starts at the t _ransportation departure time or that ends

-

1339 at the transportation arrival time .

-

1340 15. The method of Claim 12, wherein the ' gr _aph path or graph path length '

1341 between the ' representative vertex ' and the 1 site vertex 1 is obtained by

1342 applying at least one of:

1343 (a) a Dijkstra’s shortest paths algorithm, or 1344 (b) an A* (A star) search algorithm,

1345 to at least one of:

1346 (c) the ' graph or

1347 (d) the ' graph < with reversed graph edges.

1348 16. The 1 method 1 of Claim 1, wherein the < plurality of sites ' comprises:

1349 (a) for each i in the range 1 < i < m, for a threshold m > 1, a site S

1350 further comprising:

1351 (b) receiving at least one coordinate , each coordinate i in the range 1 < i < m,

-

1352 for the threshold m

1353 wherein the ' at least one precomputed route or route length < comprises:

1354 (c) a vector v comprising, for each coordinate i included in the

1355 at least one coordinate a that comprises the

1356 h _rst precomputed route or route length that comprises a description of travel

- -

1357 from the ' nearby representative ' to the < site S) or

1358 (d) a vector v' comprising, for each coordinate i included in the

-

1359 1 at least one coordinate a that comprises the

1360 h _rst precomputed route or route length that comprises a description of travel

- -

1361 from the < site S) to the ' nearby representative '.

1362 17. The method of Claim 16, wherein

1363 (a) the ]_at least one coordinate^ includes at least two coordinates, and the

- -

1364

_ storing the _ at least one precomputed route or route length further

1365 comprises:

1366 laying out, for a sequence of coordinate i included in the

-

1367

_ at least one coordinate

- -

1368 i. the ! value v[i\ \ or the \ value v'[i\ |,

_ _ -

1369 sequentially in the [_database_]; or

1370 (b) the _ storing the > _ at least one precomputed route or route length > further

1371 comprises:

1372 storing a list or a hash map comprising, for at least one coordinate i

-

1373 included in the 1 _ at least one coordinate

-

1374 i. the ! value v[i\ \ or the \ value v'[i\ |,

_

1375 in the 1 database '.

1376 18. The method of Claim 16, further comprising:

1377 for at least one hrst coordinate i included in the 1 at least one coordinate

1378 (a) receiving a weight w.

1379 wherein, for the 1 hrst coordinate i the < storing further comprises:

1380 (b) determining a length of travel, denoted a GTTP, from the

1381 ' nearby representative < to the < site S; < and

1382 (c) determining a length of travel, denoted a sr; , from the < site S; 1 to the

-

1383 nearby representative

1384 wherein, for the 1 _ hrst coordinate i

1385 (d) the ! value v[i\ \ or the \ value v'[i]

1386 comprises a length of travel equal to W; rs; + (1— W;) sr;, for the > weight W; >,

1387 the frsC, the Ts?y1.

1388 19. The method of Claim 16, further comprising:

1389 for at least one second coordinate i included in the 1 at least one coordinate

1390 (a) receiving a lower bound lb; an upper bound ub; and a

1391 scaling factor

1392 wherein, for the 1 s -econd coor -dinate i the storing further comprises: 1393 (b) determining a length of travel, denoted a bsr~ between the ' site S 1 and

1394 the ' nearby representative

1395 wherein, for the 1 second coordinate i

1396 (c) the | value v[i\ j or the \ value v'[i] j,

_ _

-

1397 comprises a length of travel equal to: the < bsri < multiplied by the

-

1398

_ scaling factor s/j ', rounded to an integer, and capped to within a range from - -

1399 the ' lower bound ¾ 1 to the ' upper bound ubi

1400 20. The method of Claim 16, wherein the ' _ at least one nearby representative '

1401 comprises:

1402 (a) for each j in the range 1 < j < s, for a threshold s > 1, a

1403

1404 further comprising:

1405 (b) receiving a + operation comprising calculating a hrst sum of a number,

1406 for one or more coordinate of a vector, with a value at the coordinate; and

1407 (c) receiving a min operation comprising calculating, for one or more

1408 coordinate, a third minimum of a value at the coordinate across a

1409 plurality of vectors;;

1410 wherein

1411 (d) the | at least one start/end place includes a start place and wherein the

1412

_ computing ' the ' _ at least one route or route length ' further comprises:

1413 i. for each j in the range 1 < j < s, for the Threshold s_j,

1414 determining a that is a length of travel from

1415 the J nearby representative Rj J to the ' start place '

1416 wherein the ' at least one precomputed route or route length ' comprises

1417 for each j in the range 1 < j < s, for the that

1418 is the 1 - vector n' 1 wherein the ' _ nearby representative ' is the

-

1419 I nearby representative R 1420 wherein a length of travel vector v'R comprises, for at least one

1421 third coordinate i included in the 1 at least one coordinate a value that

1422 is a length of travel included in the value v'R\i\ of the j vector vR' \ &t the

3 I _ J J -

1423 1 _ third coordinate i and

1424 ii. hnding a hrst minimum using a hrst mathematical formula

1425 comprising the > min operation > and the > + operation :

1427 ( the | length of travel rp | ) + ( the | length of travel vector vR |

1428

1429 the 1 _ hrst minimum for at least one hfth coordinate i included in the

I - 1

1430 1 _ at least one coordinate

1431 represents a description of travel from the ' site ¾ 1 to the

-

1432 start place >, and

_

-

1433 uses the ' _ second precomputed route or route length wherein the

- -

1434 ' not nearby site is the site S{ or

1435 (e) the | at least one start/end place \ includes an end place and wherein the

1436 computing the at least one route or route length further comprises:

_ _

-

1437 i. for each j in the range 1 < j < s, for the |_threshold s_],

1438 determining a that is a length of travel from

1439 the end place to the | nearby representative R~

- _ _

-

1440 wherein the ' at least one precomputed route or route length ' comprises,

1441 for each j in the range 1 < j < s, for the [_threshold s_], a that

1442 is the 1 vector v 1 wherein the < nearby representative is the

- _

-

1443 | nearby representative Rj J;

_

1444 wherein a length of travel vector vR comprises, for at least one

1445 _ fourth coor _ _dinate i I included in the 1 _ at least one coordinate a value

1446 that is a length of travel included in the value VR \I\ of the | vector VR | at

_

-

1447 the 1 fourth coordinate i and

1448 ii. hnding a second minimum using a second mathematical formula

1449 comprising the > min operation > and the > + operation : 1450

- -

1451 ( the | length of travel pr^ | ) + ( the | length of travel vector VR j

1452

-

1453 the 1 _ second minimum for at least one sixth coordinate i included in

-

1454 the 1 _ at least one coordinate ':

- -

1455 represents a description of travel from the ' end place ' to the ' site ¾

1456 and

-

1457 uses the ' _ second precomputed route or route length wherein the

- -

1458 not nearby site > is the site ¾ .

1459 21. The method of Claim 20, wherein

1460 (a) the calculating included in the + operation < comprises at least one

1461 of:

1462 i. a Saturation Arithmetic operation within a range from a lower

1463 bound to an upper bound,

1464 ii. an arithmetic operation operating on an integer, or

1465 iii. an arithmetic operation operating on a floating point number;

1466 (b) the !_fir_st jmathematical_formul_aj or the

-

1467 [_second mathematical formula_] further comprise at least one Vector

1468 Operation supported by hardware; or

1469 (c) the using of the ]_fir_st nathematical_fo_rmul_a_| or the using of the

-

1470 Lionel a hematical Jprmulaj further comprise: dividing

1471 calculations into a plurality of partitions , and executing the

1472 plurality of partitions > in parallel.

1473 22. The 1 m -ethod 1 of Claim 1, wherein the request comprises:

1474 (a) for each i in the range 1 < i < q, for a threshold q > 1, a

1475 route specihcation Li , wherein the ' route specihcation Li < comprises

1476 one or more place , the < one or more place < including

1477

1478 wherein the request further comprises:

1479 (b) a deriver , wherein the |_deriyer_] comprises an algorithm that computes a

1480 deriver output given a deriver input

1481 wherein the [_information_] further comprises:

1482 (c) the ' _ deriver output wherein the ' _ deriver input ' comprises a sequence

-

1483 Ti, . . . , T, for the ' _ threshold q wherein for each i in the range 1 < i < q, a

-

1484 T- comprises a part of the ^information^, wherein

1485 i. the ' _ at least one place ' is the ' one or more place ' included in the

-

1486 ' _ route specihcation

-

1487 ii. the | at least one start/end place | is the | one or more start/end place |

_ _

-

1488 included in the ' route specihcation *

1489 23. The method of Claim 22, wherein the |_deriyer_] comprises one of:

1490 (a) the ' deriver output < comprising, for a fourth site included in the

-

1491 1 at least one site a minimum route length that is a route length equal

1492 to a fourth minimum, across i in the range 1 < i < q, for the

1493 threshold q of a second route length included in the ' Ί the

1494 second route length ' involving the [ o rth-Sitey;

1495 (b) receiving for the ' threshold q <, as a part of the

1496 deriver input >; and the deriver output > comprising, for a fifth site

1497 included in the 1 _ at least one site a weighted route length that is a

1498 route length equal to a second sum, across i in the range 1 < i < g, of a

1499 third route length included in the ' Tj ', the third route length > involving

1500 the 1 - hfth site ', the < _ third route length weighted by a weight g included

-

1501 in the weights w , ... , wq j; ; or

1502 (c) receiving a site S as a part of the ' deriver input ', the included in

- -

1503 the ]_at_ le_a_st _qne site ; and the ' deriver output ' comprising, for a

1504 sixth site included in the 1 _ at least one site difference route length

1505 that is a route length equal to a difference between a fourth route length 1506 involving the ]_sixth site and a fifth route length involving the |_site S ,

1507 the ' _ fourth route length ' and the ' _ fifth route length ' included in the

-

1508 ' deriver input '.

1509 24. The 1 m -ethod 1 of Claim 1, wherein the request further comprises:

1510 (a) a condition on: the ]_at_ le_a_st _qne sitej, or a part of the ^information^;

1511 wherein the |_mformatiqnj further comprises a filtering of:

1512 (b) the least one sitej, or

1513 (c) the ^information j,

-

1514 that match the 1 condition

1515 25. The method of Claim 24, wherein the |_9 fi itiqn_] comprises at least one of:

1516 (a) a threshold of a maximum commute duration, or a maximum walk

1517 distance;

1518 (b) a real estate property type, or a real estate property price or size range; or

1519 (c) a job type, or a job salary range.

1520 26. The ]_method_] of Claim 1, wherein the request further comprises:

1521 (a) an aggregator , wherein the Taggregatom comprises an algorithm that

1522 computes an aggregator output given an aggregator input

1523 wherein the jiiiforriiatiori further comprises:

1524 (b) the Taggregator output !, wherein the < aggregator input < comprises a sequence

1525 11, . . . , Ir, for a threshold r > 1, wherein for each i in the range 1 < i < r, for

- -

1526 the Threshold r , an J comprises a part of the |_information_].

1527 27. The method of Claim 26, wherein the aggregator output comprises a centrality

1528 of a - seventh site within the t _ransportation system , the 1 - seventh site 1 included

-

1529 in the 1 at least one site '.

1530 28. The ]_method_] of Claim 1, wherein the request further comprises:

1531 (a) a cost function , wherein the [_cost_ function j computes a cost function value

1532 given a cost function input , wherein the ' cost function input ' comprises:

1533 i. an exploration route length being a route length included in the

1534 1 _ information ': and

1535 (b) an exploration algorithm , wherein the exploration algorithm comprises an

1536 algorithm that computes:

- -

1537 i. the l_cos_t_ function _yalue_| for at least one ' cost function input and

1538 n. an exploration output

1539 wherein the Tnformatiom further comprises:

-

1540 (c) the exploration output >.

1541 29. The method of Claim 28, wherein

1542 (a) the [_cost function^ is a differentiable function that depends on the

-

1543 ' e _xploration route length >; and

1544 (b) the ' exploration algorithm ' comprises a gradient descent algorithm that

-

1545 computes a gradient of the l_cos_t_ function^.

1546 30. The [_method_j of Claim 1, wherein the storing or the computing further comprise

1547 applying a multi-objective optimization search based on a multi-dimensional cost

-

1548 31. The method of Claim 30, wherein the ^mujti-dimens na^c stj comprises at

1549 least one of:

1550 (a) a price or a size of a site included in the |_at_ le_a_st_qne site.]; or

1551 (b) a route length, a monetary cost, a wait duration, a number of vehicle

1552 transfers, or a walking distance included in: the

-

1553

_ at least one precomputed route or route length , or the

-

1554 ' _ at least one route or route length '. 1555 A computer system for searching or comparing at least one site using at least one

1556 route or route length between the at least one site and at least one place within a

-

1557 transportation system, the computer system comprising:

1558 (a) one or more processor

1559 (b) a computer-readable storage medium storing one or more program for execution

1560 by the Tone or more processor !; and

1561 (c) the Tone or more program ! comprising instructions to be executed by the

1562 T _one or more processor ! so as to perform the method of one of: Claim 1 to Claim

1563 31.

1564 An apparatus for searching or comparing at least one site using at least one route or

1565 route length between the at least one site and at least one place within a transportation

1566 system, the [ apparatus ! comprising:

1567 (a) a receiver conhgured to receive from a user the request ,

1568 (b) a transmitter conhgured to respond to the [user] with a part of the [_information_],

1569 and

1570 (c) one or more module conhgured to perform the method of one of: Claim 1 to Claim

1571 31.

1572 34. The [ _apparatus ! of Claim 33, wherein the ' -receiver 1 receives at least one of:

1573 (a) the ' at least one place ;

1574 (b) the [Addition.!;

- -

1575 (c) the ' hrst departure time ', the !_hrst_arr_ival_ deadline^, the

- -

1576 ' _ second departure time or the 1 - second arrival deadline or

-

1577 (d) the |_deriver ; ;

1578 wherein the Transmitter^ further responds with at least one of:

1579 (e) a geographical location of: a place included in the ' at least one place ', or a

-

1580 site included in the ]_at least one site , rendered on a map;; 1581 (f) a departure time or an arrival time for: a place included in the

- -

1582 ' at least one place or a site included in the ]_at_ least sit_e_]; ;

-

1583 (g) a summary of a site included in the |_at_ le_a_st_qne site ; or at least one of: a

1584 name of the site, an address of the site, a price of the site, or a size of the site;;

1585 (0 a summary of a cluster of nearby sites , wherein the ' cluster of nearby sites '

1586 comprises sites included in the ]_at least one sitej clustered based on a length

1587 of travel or a geographical proximity;;

1588 (i) a rendering of the c _luster of nearby sites , a size of the rendering being

-

1589 correlated with a number of sites in the ' cluster of nearby sites >;;

1590 (j) a stacking of sites, each eighth site included in the |_at_ le_a_st_qne sitej, the

1591 stacking in a z-index order of a route length that is included in the

- - -

1592 1 information 1 and involves the eighth site , shorter route length higher up the

1593 stacking, rendered on a map;;

-

1594 (k) a summary of the ' _ route or route length >; or a part of the

-

1595

_ route or route length >, the part including at least one of: i. a route: length,

1596 duration, monetary cost, speed, or wait duration; ii. a name of: a vehicle, a

1597 vehicle road, a walking path, a stop, a turn, or a transit station of a public

1598 transportation vehicle; or iii. a geographical location of: a monetary cost, a

1599 speed, a wait duration, a vehicle, a vehicle road, a walking path, a stop, a

1600 turn, or a transit station of a public transportation vehicle;;

-

1601 (1) a histogram of a route length included in the ^information^, the route length

-

1602 involving a site, across the l_qt_ lqast_ ne sitej; or a heatmap of the route length

-

1603 across the ]_at least one sitej rendered on a map;;

- -

1604 (m) the _ minimum route length >; a histogram of the _ minimum route length > across

- -

1605 the 1 - at least one site ; or a heatmap of the _ minimum route length across the

-

1606 |_at_ le_a_st_qne sitej rendered on a map;;

- -

1607 (n) the ' _ weighted route length >; a histogram of the ' _ weighted route length ' across

- -

1608 the 1 - at least one site or a heatmap of the _ weighted route length across the

-

1609 |_at least one sitej rendered on a map;; 1610 (o) the _ difference route length >; a histogram of the _ difference route length > across

- -

1611 the 1 - at least one site or a heatmap of the ' _ difference route length ' across the

-

1612 |_at_ le_a_st_qne sitej rendered on a map;;

1613 (p) a rendering of the ' deriver output >;;

1614 (q) ninth site that minimizes a sixth route length included in the

1615 [_information_], the sixth route length > involving the [_ninth sitej, minimized

-

1616 across the ]_at least one sitej, constrained by a threshold on a monetary cost of

1617 travel;;

1618 (r) at least one of: the above item (e) to item (q), constrained by the [_cqnditiqn_|; ;

1619 or

1620 (s) a top list comprising sites included in the ]_at least one site , wherein

1621 i. each tenth site included in the top list and

1622 n. a seventh route length that is included in the 1 information 1 and involves

1623 the 1 _ tenth site

- - -

1624 satisfy the 1 co -ndition the ' top list ' sorted by the seventh route length >.

Description:
A Method and an Apparatus for

Searching or Comparing Sites

Using Routes or Route Lengths

Between Sites and Places

Within a Transportation System

CROSS-REFERENCE TO RELATED APPLICATIONS

[001] This application i based upon, and claims the priority dates of, applications:

[Country] [Application Number] [Filing Date]

USA 62/632,419 February 20, 2018

USA 62/758,710 November 12, 2018

USA 62/780,268 December 16, 2018

USA 62/800,428 February 2, 2019

USA 16274242 February 13, 2019, which are incorporated herein by reference. BACKGROUND OF THE INVENTION

[002] The present invention relates to searching or comparing sites. A traditional goal of searching is to hnd a site, from among a range of possible alternatives, that achieves an optimization objective, such as minimize a route length given specihc travel requirements and desired features of the sought site. For example, when searching for real estate proper ties given required destinations of commutes and real estate property features, a goal may be to enumerate real estate properties with matching features that have the shortest com mute durations. Other goal may be to compare any real estate properties using commute durations.

BRIEF SUMMARY OF THE INVENTION

[003] Embodiments include a method for searching or comparing sites, a computer system that implements and executes the method, and a computer service that receives search or compare requests from users and responds with site and route information.

[004] According to an embodiment of the present invention, a method for searching or comparing sites using routes or route lengths is provided. The method uses extensive preprocessing to precompute and store in a database routes or route lengths between each site and representatives within a transportation system. The method introduces a search- or-compare framework for sites. When a request containing a route specihcation is received, precomputed data is retrieved from the database to rapidly compute a route or a route length for each site. Sites may be searched or compared using routes or route lengths.

[005] According to an embodiment of the present invention, a computer system for searching or comparing sites using routes or route lengths is provided. The system is a combination of hardware and software. It obtains data about a transportation system and sites from one or more data providers. The system builds graphs that model travel between the sites and representatives within the transportation system. The system computes graph paths, and stores graph paths or graph path lengths. This enables to quickly compute routes or route lengths for every site when a request is received, and search or compare sites using routes or route lengths.

[006] According to an embodiment of the present invention, a computer service for searching or comparing sites using routes or route lengths is provided. The service allows the user to specify a search or compare request through a User Interface on a device, for example a smartphone. The request contains a route specihcation and a hltering condition. In response, the service presents sites that match the hltering condition along with routes or route lengths for the matched sites, or the service compares sites using routes or route lengths.

[007] It is fundamentally necessary to quickly compute a route length for each site. We sketch a mathematical proof. Consider any search-or-compare method LT We can design an adversarial request that will force L4 to respond with any given ordered list of sites. The adversary has two mechanisms at its disposal: (1) provide a request with a route specihcation that creates an order of sites with respect to the route length, the order selected by the adversary, and (2) provide a request with a hltering condition on sites that matches a subset of sites that is selected by the adversary. Thus the L4 must respond with an ordered list of sites arbitrarily selected by the adversary at request time. Details of the proof sketch are outside of the scope of a patent application.

[008] The method, the computer system, and the computer service each jointly performs tasks that are not simply generic nor well-understood by prior art. Prior art includes: US 8,417,409 B2; continuation US 8,738,286 B2; division US 8,756,014 B2;; KR 10-1692501 Bl; continuation PCT/KR2016/01083; WO 2017/065431 Al; US 2018/0232824 Al;; and KR 10-1905593 Bl.

[009] The embodiments of the invention presented here are for illustrative purpose; they are not intended to be exhaustive. Many modihcations and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the embodiments.

[010] In the presentation, the terms“the first”,“the second”,“the”, and similar, are not used in any limiting sense, but for the purpose of distinguishing, unless otherwise is clear from the context. An expression in a singular form includes the plural form, unless otherwise is clear from the context. The terms“having”,“including”,“comprising”, and similar, indicate an existence of components or features, and do not preclude other components or features from existing or being added. BRIEF DESCRIPTION OF DRAWINGS

[Oil] The drawings included in the invention disclosure exemplify various features and advantages of the embodiments of the invention:

• FIG. 1: depicts an example color rendering of travel durations of commute paths according to an embodiment of the invention, legend:“An example color rendering of travel durations of commute paths for real estate properties with two commute paths: property— y geo.0.0— y geo.0.1— y property, and property— y geo.1.0— y property. We used a map tile engine by Google, any other may be used instead.”;

• FIG. 2: depicts an example process flow of preprocessing and storing of data according to an embodiment of the invention, legend:“An example process flow of preprocessing and storing of data”;

• FIG. 3: depicts an example process flow of responding to a request using preprocessed stored data according to an embodiment of the invention, legend:“An example process flow of responding to a request using preprocessed stored data”;

• FIG. 4: depicts example commute paths according to an embodiment of the invention, legend:“Example commute paths”;

• FIG. 5: depicts example extensions of a public transportation system graph for pre- computing shortest graph paths according to an embodiment of the invention, legend: “Example public transportation system graph extensions for precomputing shortest graph paths”;

• FIG. 6: depicts an example table / vector storage of precomputed shortest travel durations according to an embodiment of the invention, legend: “An example table / vector storage of precomputed shortest travel durations”;

• FIG. 7: depicts an example of a commute path decomposition according to an embod iment of the invention, legend:“An example commute path H— y W >· W 2— > W 3— y W 4 — y H decomposed into a home-dependent part (green) and a home-independent part (black), illustrated on three homes Hi, H , and H”; • FIG. 8: depicts an example of computing the shortest travel duration at the start of a commute path according to an embodiment of the invention, legend: “An example of computing the shortest travel duration at the start of a commute path.”;

• FIG. 9: depicts an example of computing the shortest travel duration at the end of a commute path according to an embodiment of the invention, legend: “An example of computing the shortest travel duration at the end of a commute path.”;

• FIG. 10: depicts an example pseudocode for computing the shortest travel duration at the start of a commute path according to an embodiment of the invention, legend: “An example pseudocode for computing PathFromHomeDurations(— y W ).”;

• FIG. 11: depicts an example flowchart of a computer system according to an em bodiment of the invention, legend: “An example flowchart of a computer system for: precomputing and storing data in a database, and processing requests using data re trieved from the database.”;

• FIG. 12: depicts an example user request and a response by a computer service on a smartphone of the user according to an embodiment of the invention, legend: “An example user request and a response by a computer service on a smartphone of the user.”.

[012] The drawings are for illustrative purpose only. Other drawings can exemplify the invention without departing from the principles of the invention, as will be readily recognized by one of ordinary skill in the art.

DETAILED DESCRIPTION OF THE INVENTION

4 Detailed description

[013] The invention concerns a general case of searching or comparing arbitrary sites us ing an arbitrary optimization objective that uses routes or route lengths to arbitrary places. However, for the sake of ease of presentation, we hrst illustrate the invention through spe- cihc sites that are real estate properties, specihc places that are workplaces, and a specihc optimization objective of minimizing travel durations of commutes between real estate prop erties and workplaces. This illustration is not limiting. In later sections, we explain how the method works in the general case.

4.1 Real estate properties and commute paths

[014] Finding a home is a complex endeavor. People spend a signihcant effort of time and money on a search. Technology has come to help, however. There are several services available online that aggregate real estate property listings, and allow people to search for real estate properties with specihc features, for example a price, a geographical location, a number of bedrooms, and so on, in the convenience of a web browser or a smartphone. This search yields a shortlist that the person then typically inspects in flesh.

[015] Location is arguably the most important feature of any real estate property, as attested by the“location, location, location” slogan of real estate agents. Our invention concerns this feature.

[016] Residents need to commute to work, school, or other places. The travel durations of these commutes bear on the value of the real estate property personalized for its specihc residents. In addition, there is much global economic value in reducing the travel durations for all residents of a metropolitan area, in terms of time that can be allocated to more productive activities, energy that does not need to be spent on moving people around, etc. Our invention helps deliver these personal and global values.

[017] As a simple illustration, consider a family of two who live in the capital of South Korea. One person is a government employee at the City Hall, and the other person is a librarian at the Main Library of Seoul National University. They currently live in a two- bedroom one-bathroom apartment of size 69 square meters that requires a deposit payment of 350,000,000 KRW. The apartment is near latitude and longitude (37.5333, 127.0746). Their combined daily roundtrip commute lasts lh 41m (about 25 minutes each way, on average). However, it turns out that they can decrease the travel duration to just lh 16m (25 minutes savings) after moving to an apartment with the same features, but located near (37.5041, 126.8888). The shortest travel duration for any apartment is 50 minutes. However, that apartment has other price and size. (Illustration as of February 19, 2018.)

[018] Why is it hard to deliver these improvements for all? A naive approach is to consider every real estate property available on the market that has the size and other features required by the family, and, given the work places, compute the travel durations by querying any existing online routing engine. This approach does not scale, however. One problem is the high number of real estate properties available on the market in a modern metropolitan area. Other problem is the high number of families/users who may want to seek improvements. Even if we assume that we can query a routing engine cheaply, the quadratic nature of the problem still makes the queries expensive in aggregate.

[019] How can we deliver these improvements for all? Our invention explains. It includes the following components:

1. The invention dehnes a model for a commute path. The model is versatile to cover a wide range of commute paths that occur in practice, for example going to school, then to a piano class, and then returning home. The utility of our model is enhanced by our ability to quickly hnd real estate properties that minimize travel durations.

2. The invention teaches an optimization method that rapidly computes travel durations.

The invention identihes the parts of any commute path that are dependent on any real estate property. Travel durations for these parts are precomputed and stored. As a result, when travel durations need to be found for a commute path, the invention can rapidly assemble time parts to produce a travel duration for every real estate property.

3. An embodiment of the invention is a live computer service. The service enables the 25 million residents of the Seoul Metropolitan Area to search or compare real estate properties using travel durations.

4.2 Method outline

[020] We use the term travel in a broad sense that includes moving objects or data. A description of travel is anything that a person of ordinary skill in the art would name so. Here are some examples of a description of travel: (1)“hey buddy, you need to go one block north, and then turn slightly left”, and (2)”5 dollars”. We may use a term travel path when we mean a description of travel. A length of travel is a numeric value that a person of ordinary skill in the art can associate with travel, for example a distance, a monetary cost, etc. As other example, we may use a term travel duration when we mean a length of travel that represents time. A length of travel is by itself a description of travel. A description of travel: may not include any length of travel, may include only a length of travel, or may also include some other data.

[021] We illustrate some capabilities of the method and introduce terminology subse quently used. The method can compute travel durations of commute paths for every real estate property. FIG. 1 illustrates this. The metropolitan area is colored with squares. The colors represent how long it takes to commute from a real estate property in each specihc area using public transportation: green means short commute, yellow longer, and red longest. A commute path will start at a property , then visit specihc places geo, . . ., and hnally return to the property. In a sense the property is a free variable, while the geo, . . . are hxed. In FIG.l, there are two commute paths: an“open-jaw” commute path property— y geo.0.0— y geo.0.1— y property twice per week, and a“roundtrip” commute path property— y geo.1.0— y property three times per week. We can see that given these two commute paths and their frequencies, the real estate properties with small weighted sum of shortest travel durations form an irregular patch (dark green), which may not come as a surprise, given possibly intricate public transportation routes.

[022] At a high level, the method is composed of two parts. The hrst part computes travel durations between real estate properties and representatives that are vehicle stops of a transportation system. These travel durations are stored in a database, so that they can be readily retrieved when a request is received. See FIG. 2 for an illustration. The second part processes requests. A request contains a commute path and desired features of a real estate property. When a request is received, appropriate travel durations are retrieved from the database, and used together with other data to produce a travel duration for the commute path for every real estate property that has the desired features. See FIG. 3 for an illustration. Details and variants of this outline are described in subsequent sections.

4.3 Commute paths

[023] Our method handles a broad range of commutes that people perform within a metropolitan area. A commute starts at a site H that we call a home. The H is an arbitrary location. It can be any real estate property, for example an apartment, a rented room, a house with a garden, a ranch, a hotel, etc. It can also be a site where a person works, a restaurant, a shop, etc. However, as a naming convention, we use the phrase“home” in most of the disclosure; this convention is not limiting. In one embodiment, a commuter travels to various places, and then returns to the site H.

[024] In one embodiment, a commute lasts within one day, for example a commuter departs from H in the morning and returns back to H in the evening of the same day. In other embodiment, a commute lasts within more than one day, for example if a person works a night shift, or has a work shift of 25 hours. In one embodiment, any travel may start at a specihc time, or may end at a specihc time, for example 8:12 AM. In other embodiment, any travel may begin, or end, within a range of time, for example“in the morning”.

[025] In a simplest form, a commuter travels from H to a place W, that we call work. The W is an arbitrary location. It includes a school, a grandparents home, a weekend golf course, a favorite restaurant, a doctor’s office, a place of worship, etc. It can also be a place where a person lives. However, as a naming convention, we use the phrase“work” in most of the disclosure; this convention is not limiting. Then the commuter returns from W to H. We call these two travels a roundtrip commute path. See FIG. 4 A for an illustration.

[026] An open-jaw commute path is an example of a more complex commute. See FIG. 4B for an illustration. Here the commuter travels from H to a place W . Then the commuter travels from the place W to other place W^- Finally, the commuter travels from the place W2 back to the site H . One example is when a person travels to a school and then to a piano class.

[027] In general, our method allows for arbitrary travels. In FIG. 4C, we see an illustra tion of a commute path with a missing travel from W4 to W2, and a repeated travel from W2 to W3. In FIG. 4D, we see illustrations of “open” commute paths: a commute path that departs from H, but does not return to H ; and a commute path that returns to H, without previously departing form H. In FIG. 4E, we see an illustration of a“disconnected” commute path. A commute path could start at a home, but end at other home. Our method dehnes a commute path as follows.

Definition 1 A commute path is a collection of travels W2— > W3 , W4— > W5 , . . . , W k- 2— t W k 1, for k > 2 that is even, together with Hf irst— W or W k— > Hi ast (so a commute path always contains at least one home and at least one work), that occur at arbitrarily moments of time. [028] A commute path can be viewed as a specification of a route within a transportation system. The various W and H in Definition 1 specify where the commuter wishes to travel.

[029] In one embodiment, we consider a simpler commute path H— y W — y W2 W k — > H, for any k > 1, that has shared endpoints between travels, and the same home at the start and the end. We use this simpler form in most of the disclosure, because it is common in practice and simplifies our presentation. However, it will be obvious to anyone of ordinary skill in the art that our method applies to our (general) definition of a commute path.

[030] Our method finds a shortest travel duration along any commute path. This by itself has been addressed by various prior art. However, our method finds travel durations for all homes in a dramatically reduced amount of time.

4.4 Preprocessing of transportation system

[031] The method preprocesses data about public transportation system, so as to pre compute and store shortest travel for all homes.

4.4.1 Computation of shortest travel

[032] We describe a method for efficiently computing shortest travel between all homes, and all locations of public transportation stops. We call the stops stopstations in most of the disclosure, and they include bus stops, subway stations, or both.

[033] The method starts with an arbitrary public transportation system graph GT that models a public transportation system, and that can be obtained from prior art. The graph may contain vertexes that represent bus stops, subway stations, or both. Other vertexes may exist in the graph, for example representing a stop or a turn of a vehicle, or a stop or a turn of a walk. Vertexes in the graph that correspond to stops of vehicles are denoted by

STOPSTATION s, indexed by s. There are directed weighted edges, each representing a travel duration along a segment from the source vertex of the edge to the target vertex of the edge. Other edges may exists in the graph. An edge may represent a transfer from one vehicle to other vehicle. The graph may contain data about departure times or arrival times of various public transportation vehicles. A Dijkstra’s shortest graph paths algorithm, or an A* (A star) search algorithm, is commonly used to compute a graph path with a shortest graph path length (sum of weights), representing a shortest travel duration, from a

STOPSTATION s' to other

STOPSTATION_s", for any s' and s", also if travel starts at a specihc time, or ends at a specihc time.

[034] Our method extends the public transportation system graph GT. See FIG. 5 for an illustration.

[035] The hrst extension introduces clusters of stopstations. The method clusters stop- stations using any clustering algorithm; in one embodiment the method puts two stopstations into one cluster when the geographical distance between the two stopstations is at most a threshold, for example 5 meters, or when a travel duration between the two stopstations is at most a threshold. We may refer to the location of a cluster, when we mean a geographical location within the cluster, for example the center of the cluster. For each stopstation cluster c, the method adds two vertexes

STOPSTATION _CLUSTER_SOURCE_c and

STOPSTATION _CLU STER TARGET c, and edges that connect the cluster with stopstations of GT

STOP ST AT ION _CLU STER SOU RCE c ® STOPSTATION _s' labeled FirstWaitGetOn, and

STOPSTATION _s' ® STOPSTATION _CLU STER TARGET c labelled Zero, for any s', whenever s' is in cluster c. The edges have zero weight. The resulting graph is denoted by GC, having vertexes VC and edges EC. A subset of the vertexes STOPSTATION CLUSTER SOURCE c, for all c, is denoted by V S.

[036] The seconds extension introduces clusters of homes. The method clusters homes using any clustering algorithm; in one embodiment the method uses the same algorithm as when clustering stopstations. Similarly, we may refer to the location of a home cluster. For each home cluster s, the method adds a vertex

HOME CLUSTER SOURCE s and a vertex

HOME CLU STER TARGET s.

The method connects each home cluster to stopstation clusters using walks. Specihcally, the method adds an edge

HOME CLU STER SOU RCE s ® STOPSTATION _CLUSTER_SOURCE_c labelled Walk, if there is a walk from the location of the home cluster s to the location of the stopstation cluster c, for any s and c, with edge weight set to the duration of the walk; and a“reversed” edge

STOPSTATION _CLUSTER_TARGET_c ® HOME CLU STER TARGET t labelled Walk when there is a walk in the reverse direction from the location of the stopstation cluster c to the location of the home cluster t, for any c and t, with edge weight set to the duration of the walk. In one embodiment, the method limits walks to shortest walks at a specihc speed, for example 4 km/h, lasting at most a threshold, for example 1 hour. The resulting graph is denoted by G. We denote by VH the set of vertexes

HOME CLU STER SOU RCE_s, for all s ; and denote by EH the set of edges

HOME CLU STER SOU RCE s ® STOPSTATION_CLUSTER_SOURCE_c, for all s and c.

[037] We are interested in computing a shortest graph path from each stopstation cluster to each home cluster, and a shortest graph path in the reverse direction from each home cluster to each stopstation cluster.

[038] We remark that clustering allows our method to signihcantly improve the perfor mance of a shortest graph paths computation, by“unifying” locations that are essentially the same with respect to shortest graph paths. For example, in a tall apartment complex, there may be hundreds of homes, and our method will“unify” them into just one home cluster. Thus a shortest graph paths algorithm needs to compute shortest graph paths for just one home cluster, rather than for hundreds of constituent homes.

[039] A standard application of a Dijkstra’s algorithm still yields poor performance, however. That application runs the algorithm from each

STOPSTATION _CLUSTER_SOURCE_c in the G without any

HOME CLUSTER SOURCE s vertexes, and then runs the algorithm from each

HOME CLUSTER SOURCE s in the G without any

HOME CLU STER TARGET t vertexes. The combined asymptotic time complexity is

[040] Our method improves on this standard application. We observe that in a large metropolitan area it is often the case that the number of home clusters is signihcantly larger than the number of stopstation clusters \VH\ (l S'), given the same clustering threshold. Utilizing this observation, our method uses a different algorithm for computing shortest graph paths from home clusters: the method reverses the edges of G, and for each

STOPSTATION _CLUSTER_TARGET_c

in the reversed graph, computes shortest graph paths to all

HOME CLUSTER SOURCE s vertexes using Dijkstra’s algorithm (any

HOME CLU STER TARGET t vertexes can be removed). This yields the desired effect, because when we reverse the edges of any shortest graph path from

STOPSTATION _CLUSTER_TARGET_c to

HOME CLUSTER SOURCE s

in the reversed graph, we obtain a shortest graph path from

HOME CLUSTER SOURCE s to

STOPSTATION _CLUSTER_TARGET_c

in the original (not reversed) graph. Hence, the asymptotic time complexity of our method is just

o(Vs| ((|£C| + \EH\) + (\VC\ + \VH\) log(|V'C| + |IW|))V

In practice, our method offers a signihcant reduction in the total running time on a graph G of the Seoul Metropolitan Area.

[041] The method uses a symmetric algorithm in the opposite case when there are more stopstation clusters than home clusters \VH\ < [US'!· In that case, the method reverses the edges when computing shortest graph paths from stopstation clusters to home clusters.

[042] In one embodiment, the method uses departure times. The method computes shortest graph paths from each STOPSTATION CLUSTER SOURCE c to every

HOME CLU STER TARGET t, for a given departure time from the

STOPSTATION CLUSTER SOURCE c.

In that case, the method uses an appropriate prior art graph GT that allows to specify the de parture time from stops of public transportation vehicles; this case is sometimes represented using a vertex that corresponds to a time and a geographical location, and an edge that represents a travel duration starting from the geographical location at the time. Similarly, the method computes in the reversed graph based on a given arrival time at

STOPSTATION _CLU STER TARGET c.

Shortest graph path lengths convert these arrival times to the departure times from each

HOME CLU STER SOU RCE s.

A similar graph is used to compute shortest graph paths given an arrival deadline. An appropriately constructed graph can be used to compute a probability of arrival before a deadline.

[043] In one embodiment, we do not extend GT with any vertexes STOPSTATION CLUSTER SOURCE c nor any vertexes

STOPSTATION CLUSTER TARGET c.

Instead, we connect any vertexes

HOME CLUSTER SOURCE s and any vertexes

HOME CLUSTER TARGET t by direct Walk edges with any vertexes

STOP ST AT ION _s' .

[044] In one embodiment, we do not extend GT with any vertexes

HOME CLUSTER SOURCE s nor any vertexes

HOME CLUSTER TARGET t.

Instead, we add vertexes

HOME s each representing a home, that we connect to any vertexes

STOPSTATION s' directly by Walk edges.

[045] In one embodiment, we do not cluster homes.

[046] In one embodiment, we do not cluster stopstations.

[047] In one embodiment, we compute shortest graph paths given an arrival deadline: from stopstation clusters to home clusters given an arrival deadline at each home cluster, or from home clusters to stopstation clusters given an arrival deadline at each stopstation cluster.

[048] In one embodiment, we use any shortest graph paths algorithm other than a Dijk- stra’s algorithm, for example an A* (A star) search algorithm. In one embodiment, we use an approximation algorithm for shortest graph paths. We may use an algorithm without any of the performance improvements.

[049] In one embodiment, the weights on some graph edges represent monetary costs of travel, instead of travel durations. Then our method searches or compares homes based on the monetary cost of commute paths. Any other semantic of edge weights can be used, for example: a number of transfers among public transportation vehicles, a duration of wait, a monetary cost of wait, or a distance of travel.

[050] In one embodiment, we apply a multi-objective optimization search based on a multi-dimensional cost. For example, we search for a shortest graph path whose length represents a travel duration, such that a monetary cost that represents the graph path is at most a threshold, or when a monetary cost is added as a penalty to the graph path length.

[051] In one embodiment, a travel path is represented by a graph path. We compute various features of a shortest travel path using a shortest graph path, for example: the hrst (i.e. , boarding) or the last (i.e., disembarking) stopstation of the shortest graph path, the main transit stopstation of the shortest graph path, the number of vehicle transfers, the vehicle where most time is spent during travel (for example bus 1234), the total wait time at stopstations, the total walk distance, whether the specihc shortest graph path is typically congested during rush hours, or a sequence of geographical locations along the graph path. These features may be used during request processing to hlter travel paths whose features match a condition specihed in the request.

[052] In one embodiment, we compute two or more graph paths between some graph vertexes. For example, one graph path from a vertex u to a vertex v could be with at most one bus, and other graph path from the vertex u to the vertex v could be without any subway, but no more than 10 minutes longer than a shortest graph path.

[053] In one embodiment, we compute graph paths that match various hltering condi tions. For example, graph paths that have at most one transfer, or at most a specihc walk duration.

[054] In one embodiment, we use a routing engine (for example prior art) to compute shortest travel between homes and public transportation stops. Thus we may sometimes not use the graphs described in the current Section 4.4.1.

[055] The method described in Section 4.4.1 computes a shortest travel from each home to each stopstation, and a shortest travel in the reverse direction. Next we describe a data structure for efficient storage and processing of shortest travel data used by our method. 4.4.2 Storage of shortest travel

[056] In one embodiment, our method stores travel durations in a vector form. The method sequences stopstation clusters as si, . . . , s n , and sequences home clusters as hi, . . . , h m in some orders, for example random orders. In one embodiment, these sequences become fixed. For each s^, the method stores a vector of shortest travel durations from s * to the home clusters using the home cluster sequence so that the value of the vector ¾ at a coordinate j, Vi[j\, is ti equal to the shortest travel duration from the stopstation cluster to the home cluster h j . See FIG. 6 for an illustration.

[057] The home cluster sequence simplifies the calculation of a shortest travel duration to every home cluster. For example, if a commuter needs to travel from a stopstation cluster ,¾ and in addition from a stopstation cluster s i2 , we simply add vectors ¾ and v i2 coordinate- wise, and the resulting sum contains the total travel duration from both stopstation clusters to each home cluster hi, . . . , h m .

[058] In one embodiment, the method stores durations of reversed travel, from homes to stopstations, using the same home cluster sequence. That is, the method stores a vector so that v j '[j\ = tG is the shortest travel duration from the home cluster h j to the stopstation cluster s * (note the transposition; the vector v is for one stopstation cluster, even though travel is in the opposite direction).

[059] In general, the two vectors for the same stopstation cluster are not equal, ¾ f v , because the minimum travel duration to a home may be different than from a home (travel is not symmetric in general). However, in one embodiment the method computes and stores just one of the two vectors, and uses it instead of the other, which can save time and space. In other embodiment, the method stores a coordinate- wise weighted average of the two vectors, which may decrease a worst-case error. The weights may be set to 0.5, or may favor travel from, or may favor travel to a specific home cluster, for example based on request frequency.

[060] In one embodiment, the method stores travel durations rounded to the closest minute, using one byte of computer memory, represented as an unsigned integer from 0 to 254, with 255 denoting an unknown or too large travel duration. This storage enables efficient vector addition using modern computer hardware support for Vector Operations and Saturation Arithmetic (for example the AVX-512 instruction set, or GPU intrinsics), while keeping the error at a practically acceptable level of at most half a minute, and covering common travel durations up to over 4 hours. Any other rounding can be used, for example a duration in seconds can be divided by 120, and rounded to an integer, which will represent durations at a 2-minute granularity.

[061] In one embodiment, the method stores a vector not for clusters. For example, some of the Si, . . . , s n represent stopstations (not stopstation clusters), or some of the hi, . . . , h m represent homes (not home clusters).

[062] In one embodiment, the method uses other form of a vector to store travel dura tions. For example, a hash map, or a (coordinate, value) list. These sparse forms may be advantageous when there are many unknown or too large travel durations.

[063] In one embodiment, the method stores travel paths using vectors that follow the home cluster sequence hi, . . . , h m , or using other form of a vector, for example a hash map or a (coordinate, value) list. These travel paths may be used during request processing to filter travel paths whose features match a condition specified in the request.

4.5 Travel duration of one commute path

[064] We describe how our method efficiently computes the shortest travel duration of any commute path. Consider any commute path from a home cluster H, through k > 1 works, back to the home cluster: H— y Wi— >· W2 W k— > H. We want to find the shortest travel duration of this commute path for every H. This may be expensive, because the number of home clusters may be rather large, for example 500,000 given a clustering radius of a few meters. However, our method introduces a technique that dramatically accelerates the search: the method decomposes any commute path into a home-independent part, and a home-dependent part, as illustrated in FIG. 7.

4.5.1 Middle part: travel

[065] The method finds the shortest travel duration for travel Wi — > W2

W k that excludes the home cluster. For each segment Wi— > W+i, the method queries a routing engine (for example prior art) that computes the shortest travel duration from Wi to Wi+ 1 using the transportation system (for example, involving walks, subway and bus rides, transfers, including a direct walk from Wi to VR+i). Then the method adds up the shortest travel durations across the different i. The resulting sum is denoted by

PathNonHomeDuration

RoutingEngineShortestDuration .

l<i<k l

We note that only k— 1 queries to the routing engine are needed. This number is independent of the number of homes.

[066] In one embodiment, the method uses a departure time, an arrival deadline, or other parts of a request, when querying the routing engine.

[067] Next, the method computes the shortest travel duration for the open-jaw part of the commute path: the two segments H— y W and W k— > H that involve the home cluster. This computation needs to be especially quick, because its result is needed for every home cluster.

4.5.2 Start part: travel H— W

[068] The method considers two ways of getting from H to W . See FIG. 8 for an illustration. The hrst way is a direct walk. When H and W are nearby, a true shortest travel will often be a direct walk. The method thus queries a walk engine (for example prior art) to compute the shortest walk duration walk(H— y Wi). The second way uses the transportation system. The method Ends the stopstation clusters that are within a threshold distance, for example 2000 meters, from W , denoted by a set A. The set A is a subset of {si, . . . , s n } of all stopstation clusters (in FIG. 8 the A {si, s 2 , S 3 }). The method queries a walk engine to retrieve a shortest walk duration walk(si W ), for each S j in A. When H and Wi are not nearby, a true shortest travel will often pass through a stopstation in the set A, and then continue along a corresponding walk. Due to the two-way approach we apply, the resulting travel duration often is a shortest travel duration.

[069] The two-way approach can be applied to each home cluster. We recall that a vector v contains shortest travel durations to stopstation cluster s * from consecutive home 521 clusters. Therefore, the method computes a shortest travel duration from a home cluster h j

522 by retrieving vectors v from the database, and using the following formula

(Equation 1)

523 [070] In one embodiment, the method computes walk(H j — W ) only when the distance

524 from home cluster H j to W is below a threshold, for example 2000 meters, or when a travel

525 duration from home cluster H j to W is below a threshold. Let us denote the set of such

526 home clusters by J.

527 [071] In one embodiment, because of the vector representation of the v', the method

528 jointly computes the travel duration from every home cluster, using vector operations on the

529 v' as follows: for j = 1 to m

Clj = oo

for all Si G A

for all j G J

w = walk(H j -A Wi)

d j = min{¾ ; w }.

530 072 In one embodiment, the method uses a mathematical formula where the“+” operation adds a number to a value at each coordinate of a vector, and the “min” operation computes a minimum value at each coordinate across several vectors.

[073] The vector (oq, . . . , a m ) of travel durations is denoted by

PathFromHomeDurations(— y W ) = (cq, . . . , a m ).

[074] An example pseudocode for computing PathFromHomeDurations is illustrated in FIG. 10.

[075] In one embodiment, a uint8 number format is used to store cq, v[\j], or w. In one embodiment, the unit of a number stored is minutes. In one embodiment, a sum -u'f ] + w, for some j, is performed using Saturation Arithmetic in the range from 0 to a threshold, for example 255. In one embodiment, a summand -u'f ] is hrst converted into a wider number format, for example into an fpl6 format or a uintl6 format, and only then added to w, to avoid an arithmetic overflow during an addition. In one embodiment, some calculations included in the mathematical formula are performed using at least one instruction for Vector Operations (including tensor operations), for example an AVX-512 instruction or a GPU intrinsic supported by hardware. In one embodiment, some calculations included in the mathematical formula are partitioned, and the partitions are executed in parallel.

[076] In one embodiment, the method uses a nearest-neighbor data structure, for example a KD-tree, on stopstation cluster locations to quickly compute the set A during request processing.

[077] In one embodiment, the method limits the set A to at most a certain number of stopstation clusters that are nearest W , for example at most 100.

[078] In one embodiment, the method limits the set A to stopstation clusters within a walk of at most a certain length to W , for example 2000 meters.

[079] In one embodiment, the method precomputes shortest durations of walks between points within a threshold distance from each home cluster location and the home cluster loca tion, or precomputes shortest durations of walks between points within a threshold distance from stopstation cluster locations and the stopstation cluster location. Then, during request processing, the method does not query any walk engine, but instead uses the precomputed walk durations.

[080] In one embodiment, a shortest duration of a walk is estimated using a geodesic line that ignores any obstacles. This may speed up the computation of a walk(si— Wi) at the expense of accuracy.

[081] In one embodiment, travel paths are used in Equation 1. For example, if a user request specihes a condition that limits the number of vehicle transfers, we filter s * and j in the equation: we look up the number of transfers stored in a travel path, and when that number exceeds the limit, we ignore the specific s * and j in the equation.

[082] In one embodiment, grouping travel durations by the stopstation s * , just like we did in v), improves data access performance. Indeed, although the set A depends on a user request that is not known in advance, for each s * in the set A, predictably a large number of travel durations need to be accessed.

[083] In one embodiment, the vectors v', for a collection of i, are distributed randomly across different processing units. This may decrease the latency of computing the vector (ai, . . . , a m ), because the stopstations included in the set A will often be evenly divided across the processing units.

[084] In one embodiment, the vectors v', for a collection of i, are grouped geographi cally inside a processing unit. This may increase the throughput of computing the vector (a i, . . . , a m ), because of a reduced need for data transfer, due to the fact that the set A often consists of stopstations located near one another.

4.5.3 End part: travel Wk — > H

[085] The computation is analogous, but uses vectors v, not v' . See FIG. 9 for an illustration. The method computes a set B of stopstation clusters within a threshold distance from W k (in FIG. 9 the B = {s 4 , s 5 }). Then the method computes a shortest travel duration to the home cluster h j as

(Equation 2)

[086] Similar as before, in one embodiment, because of the vector representation of the v, the method jointly computes a travel duration to every home cluster using vector operations on the v according to Equation 2. In one embodiment, the method uses a mathematical formula

[087] The vector (¾, . . . , b m ) of travel durations is denoted by

PathToHomeDurations(W — /) = (¾, . . . , b m ).

[088] The method uses embodiments similar as in Section 4.5.2. For example, in one embodiment, the method uses a vector operation that together computes a part of the vector (a l , . . . , a m ) and a part of the vector (¾, . . . , b m ).

4.5.4 Combining start, middle, and end parts

[089] Finally, the method computes the shortest travel duration of the commute path for every home cluster. The method simply adds the two vectors and shifts values at the coordinates. We denote the resulting vector by

PathDurations

PathFromHomeDurations(— y W )

+ PathToHomeDurations(P t)

+ PathNonHomeDuration W k ), (Equation 3) where the hrst“+” is a coordinate- wise addition of vectors, and the second“+” is an addition of a number to a value at every coordinate of a vector.

[090] In one embodiment, the method computes travel durations given specihc departure times from, or arrival deadlines to, any geographical location along any commute path.

[091] In one embodiment, the method precomputes PathDurations for specihc commute paths. For example, the method computes and stores

PathFromHomeDurations(— y W) and

PathToHomeDurations(hP— /) for every W that is a school. During request processing, the method retrieves precomputed vectors from the database, instead of computing them with Equation 1, Equation 2, and Equation 3.

[092] In one embodiment, the value of PathDurations may be changed. Consider a school W that has a zoning requirement that specihes that only certain homes may send children to the school. This can be simply achieved by setting the values at coordinates of a vector PathFromHomeDurations(— y W) to inhnity for these home clusters that are outside of the zone of the school, and similarly setting values at coordinates of a vector PathToHomeDurations . A similar change can be made when W is a restaurant with a limited delivery area, or a government office with a limited jurisdiction. Thus changed values may be stored, and retrieved during request processing.

[093] In one embodiment, our method applies an arbitrary“manipulation” function to the values at coordinates of vectors ¾ , v , or PathDurations. For example, this can help implement a policy of a government that reduces the transportation charge for commuting from or to certain parts of a metropolitan area in a special economic zone affected by a disaster. An application can be performed before request processing (for example a global policy), or during request processing (for example a user-personalized policy).

[094] In one embodiment, the method computes PathDurations for a subset of home clusters. For example, the subset is determined from a condition on home features specihed by a user request. In one embodiment, we select the sequence hi, . . . , h m of home clusters so that these subsets often reside in a short segment of the sequence, which reduces data transfer.

4.6 Search for one home

[095] Our method teaches how to efficiently search for a home using commutes. We present a few example search requests first, before we introduce a general search request for one home.

4.6.1 Weighted sum request

[096] Consider a family with one parent going to work located at geol five times per week, and the other parent going to work at geo2 three times per week. This family wants to hnd a home with a short weekly travel duration. We can hnd the weekly travel duration for the family for every home cluster as a weighted sum of two vectors, as follows:

5 · PathDurations(— y geol— /)

+ 3 · PathDurations(— y geo2— /).

4.6.2 Minimum request

[097] Consider a single mother who works from home, and is sending a child to school. The mother wants to hnd a home that will be near any school from among schools E. We can hnd the daily travel duration for the child for every home cluster as a coordinate-wise minimum of vectors, as follows: min PathDurations(— y e— /).

4.6.3 General search for one home

[098] In one embodiment, our method dehnes a request as any sequence of commute paths pathi , . . . ,path q , for any q, and a function Deriver : BP— y M that maps any vector of q numbers into a number. The method computes the travel duration vectors (from scratch, by retrieving precomputed vectors, or both)

PatliDurations(pat/q) =(ci 1,1 , · · · , di ,m ) >

PathDurations(pat g ) = {d Q i , . . . , d q^m ), and applies the function Deriver coordinate- wise to the vectors, to produce a vector

RequestDurations^at x, ., path q , Deriver) =

Deriver (dpi, . . . , d q i ),

Deriver (di 2 , . . . , d q 2 ),

Deriver

with a“derived” travel duration for each home cluster.

[099] In one embodiment, the function Deriver is a weighted sum of numbers

Deriver(ay, . . . , x q ) = w x + . . . + w q x q , for weights w , . . . , w q . Weights can be positive or negative. If a travel duration is interpreted as a monetary cost of travel, then a negative weight can be interpreted as a monetary gain, for example if the commuter is expected to beneht from performing the specihc commute (e.g., a delivery of a parcel). Weights can represent relative importance of commute paths, for example a commute path of a Chief Executive Officer may have a higher weight than a commute path of a hrst-line manager.

[100] In one embodiment, the function Deriver is a minimum of numbers

Deriver (ay, . . . , x q ) = min ay.

l<i<¾r

[101] In one embodiment, the function Deriver is a weighted sum and a minimum Deriver( i, . . . , x r , x r +i, . . . , x q ) = w ay + . . . + w r x r + min wy · ay.

r+l<i<g

[102] In one embodiment, the function Deriver is a conditional, for example

Deriver(oy, ay) = [

if Xi < 30 then

return ay + ay

else

return oo] , or any algorithm.

4.7 Compare two or more homes

[103] Our method teaches how to efficiently compare multiple homes using commutes. We illustrate the method on a few examples, before we introduce a general comparison request. 4.7.1 Current home

[104] Consider a family that currently lives in a home. The family members have specific commute paths to work places, schools, and other places. The family considers a move to other home, and wants to compare the total travel duration for their current home with the total travel duration for other prospective homes. Our method makes such a comparison rather simple.

[105] In one embodiment, the method computes the travel duration for every home cluster, including the home cluster h j of the current home S of the family

RequestDurations = (¾, . . . , q m ).

The method then responds with a“difference” vector: the method subtracts the value at the -th coordinate from the value at each coordinate

A negative value at a coordinate of the“difference” vector indicates that the home cor responding to the coordinate has a shorter travel duration compared to the current home

S.

[106] In one embodiment, the method uses some commute paths for the current home, but other commute paths for other homes. Thus the method can help a family evaluate a what-if scenario:“Suppose we change workplaces, and move to some other home. How will the new commute duration compare to our current commute duration?”

4.7.2 General comparison for two or more homes

[107] Consider a family where the mother and the father must continue commuting to the same work places (no job change), but the children can change schools. Consider other example of two families: parents and their maternal grandparents. They want to find two homes within around 30 minutes from each other, so that one home is near any hospital, and the other home is near specific schools and work places. Our method makes comparisons of such homes with other homes rather simple.

[108] In one embodiment, the method computes travel durations PathDurations for home clusters for a range of commute paths PathDurations(pat/ii) = (dy i, . . . , di ?n ),

PathDurations(pat 9 ) = (d 9 l , . . . , d q^m ).

ese Then the method applies a“generalized” function Deriver that does not operate coordinate- 687 wise as in Section 4.6.3, but instead operates jointly on all q-m travel durations, and produces ess a vector of one or more numbers. In other words, the generalized function is

Deriver : R q m ® R y ess for some y.

690 4.8 Search-or-compare

691 [109] In one embodiment, the function Deriver is any algorithm, for example a ran

692 domized algorithm, that takes any input (for example: the commute paths, shortest travel

693 paths, shortest travel durations, the homes, or any condition specihed by a user request),

694 and produces any output (for example: a“top list” containing homes that satisfy the user

695 condition, and that have shortest travels that also satisfy the user condition, sorted by the

696 shortest travel durations).

697 4.9 Variants

698 [110] Many modihcations and variations will be apparent to those of ordinary skill in

699 the art without departing from the scope and spirit of the embodiments. We present of few

700 variants for illustration.

701 4.9.1 Extending travels

702 [111] In one embodiment, the method hrst computes incomplete shortest travels, and

703 then extends them to selected home clusters. This trades off additional processing and

704 storage for a decrease in storage and processing due to fewer home clusters in vectors ¾ and [112] The method determines home connectors which are some elements of the trans portation system. In one embodiment, the home connectors are clusters of homes or clusters of stopstations, which may differ from the clusters discussed in Section 4.4.1. Then the method precomputes and stores in a database shortest travels between the stopstation clus ters and the home connectors, using embodiments similar to these of Section 4.4.

[113] When a request is received, the method computes shortest travels between a work W contained in the request, and the selected home clusters. For this purpose, the method determines stopstation clusters nearby the work W, and retrieves precomputed shortest trav els between the nearby stopstation clusters and the home connectors, to compute shortest travels between the work W and the home connectors, similar to Section 4.5. In one em bodiment, this computation uses vector operations similar to these of Sections 4.5.2 and 4.5.3. Then these shortest travels are extended beyond the home connectors, to form short est travels between the work W and the selected home clusters. This can be achieved by simply hnding, for each home cluster, a minimum extended travel (which may be not just a walk) that passes through any home connector near the home cluster, for example within 2000 meters, and sometimes hnding a direct shortest travel between the work W and the home cluster (similar to Sections 4.5.2 and 4.5.3). In one embodiment, in an overlap area the method uses both the home connectors and the vectors ¾ and v'.

[114] The performance of extending travels is proportional to the total number of home connectors, and also proportional to the number of home connectors near any selected home cluster. In one embodiment, the extending is used in sparse parts of a metropolitan area, where the two quantities are likely to be low. In one embodiment, the method uses a performance cost function to: (1) determine home connectors, (2) select home clusters, and (3) determine a subset of nearby home connectors for each of the selected home clusters.

4.9.2 Commute by car

[115] In one embodiment, we compute travel durations of commute paths by car, rather than by public transportation. This can be achieved by simple modihcations to earlier sections.

[116] When preprocessing a transportation system for homes, Section 4.4, instead of taking a GT to be a graph of a public transportation system, we take a GT to be a graph of a car drive system. This car drive graph can be obtained, for example, from prior art. That graph may have vertexes representing geographical locations on roads, and edges representing driving a car or turning a car; edges may contain data about drive durations at various times of the day, for example during rush hours.

[117] We extend GT. For each home cluster s, we add vertexes

HOME CLUSTER SOURCE s and

HOME CLU STER TARGET s.

We connect each home cluster with some roads that are within a threshold distance, for example 100 meters, by adding a vertex

CONNECTOR_r, for at least one r. In one embodiment, the vertex denotes a shortest-distance projection of a home cluster location onto a road that is within a threshold distance; the road may be inside a car parking assigned to the home cluster. The vertex is connected to the two home cluster vertexes by edges labeled Zero with weight 0. The vertex is also connected to vertexes representing endpoints of a segment of the road. Instead of adding clusters of stopstations, which we should not because now GT does not model any subway station nor bus stop, we add vertexes

REPRESENTATIVE SOURCE s' and

REPRESENT ATIV E_TARGET_t’ for a collection of s' and t'; these vertexes represent locations that frequently occur in shortest travels; the locations can be obtained, for example, using prior art. These vertexes may represent clusters. These vertexes get appropriately connected with other vertexes using edges.

[118] We use the extended GT to compute shortest durations of graph paths from each REPRESENTATIVE SOURCE s' to every

HOME CLU STER TARGET s, and from each

HOME CLUSTER SOURCE s to every

REPRESENTATIVE TARGET f possibly using embodiments similar to these described in Section 4.4.1. These travel dura tions will be stored in a vector form ¾ and v as described in Section 4.4.2.

[119] When computing a travel duration of a commute path, we use car rides, instead of using walks as in Section 4.5.

[120] For a start part of a travel, we make the following modihcations to Section 4.5.2. Instead of walk(H j— y W ), we use a duration of a car ride from a home cluster H j to W \ , for example obtained from prior art; in one embodiment we instead set this duration to inhnity when the distance between H j and W is above a threshold. The set A is a set of vertexes

REPRESENT ATIV E_TARGET_t’ that are within a threshold distance from W . For each vertex G A, we compute a car drive duration from s * to W \ , for example using prior art. Then we apply these modihcations to Equation 1.

[121] For an end part of a travel, we make similar modihcations, but now to Section 4.5.3. Instead of walkiW k— > H j ), we use a duration of a car ride from W k to a home cluster H j , for example obtained from prior art; in one embodiment we instead set this duration to inhnity when the distance between W k and H j is above a threshold. The set B is a set of vertexes

REPRESENTATIVE SOURCE s' that are within a threshold distance from W k - For each vertex s * G B, we compute a car drive duration from W k to s * , for example using prior art. Then we apply these modihcations to Equation 2.

[122] We use other embodiments stated in Section 4.5. For example, we may estimate a travel duration of a car ride using a geodesic.

[123] The search framework of Section 4.6 can be simply extended to allow to restrict what kind of vehicles should be used for each commute path. As a result, for example, we can determine travel durations of commute paths for every home for a family with one parent going to work located at geo 1 by car hve times per week, and the other parent going to work at geo2 by public transportation three times per week.

[124] The comparison framework of Section 4.7 can similarily be extended. As a result, for example, we can search for a new home from which a person will commute by car, and compare with the current travel duration by public transportation. A result of our method may be used as an incentive to purchase a car due to a move.

4.9.3 Commute by other means

[125] In one embodiment, the method uses commute paths by other means, for example: walk only; bicycle only; express buses and walks only; subway and walks only; express buses, subway and walks only; shared vans and walks only; boats; airplanes; and so on. We simply use modihcations similar to these described in Section 4.9.2. In one embodiment, the method identihes other home and other means of travel, and recommends these to the user, explaining gains compared with the current home and the current means of travel.

[126] Shortest graph paths in a given graph can be computed without knowing geograph ical locations of the various vertexes of the graph. Hence, in one embodiment, the method uses a transportation system whose various elements lack geographical location.

[127] The transportation system need not physically move objects. The method merely needs to be able to determine route or route lengths between the elements of the transporta tion system. Thus a computer network, that moves data, is an example of a transportation system, comprising these transportation elements: wires/lines (analogous to roads), and hubs/switches (analogous to stops/turns). Many other examples of a transportation system will be apparent to anyone of ordinary skill in the art.

4.9.4 Conditions on travel paths

[128] We can simply realize various filtering conditions on travel paths. For example, we can build a graph G that has no transfer between any subway lines but has walks between home clusters and station clusters. Along with modihcations similar to these described in Section 4.9.2, the method will then search-or- compare homes when a commuter may sit all the way during a ride between home and work (both the home and the work will need to be within a walking distance from a subway stop of one subway line). Similarly, we can build a G and modify shortest graph path algorithms so as to restrict graph paths to at most one transfer, or a subway-bus transfer, or a bus-subway transfer, or a transfer that occurs within a time window, or a type of travel path commonly used by commuters. Other variations for hltering travel paths fall within the scope of our method, as will be apparent to anyone of ordinary skill in the art.

4.9.5 Conditions on homes

[129] In one embodiment, the method receives a hltering condition on various features of homes, and searches or compares homes whose features match the condition. The features may be: a type (e.g., a detached home or a high-rise apartment), a transaction type (e.g., sale or rent), a price, a real estate agent commission fee, taxes, a maximum bank loan amount, a number of bedrooms or bathrooms, an area / size of a home, geographical directions of windows, a floor number of a home, the number of floors in the building, or a typical monthly management fee. Simply the method maintains a feature list for each home, and, given a condition, determines the homes whose features match the condition. The travel durations for these matching homes can be obtained from travel durations for home clusters. Other variations for hltering homes fall within the scope of our method, as will be apparent to anyone of ordinary skill in the art.

4.9.6 Meta search-or-compare

[130] In one embodiment, the method searches or compares homes using an ensemble of prior search-or-compare requests and responses. This can be viewed as a meta method (a method that uses itself). It is useful, for example, for estimating the value of a proposed new real estate development.

[131] We describe an embodiment of the meta method. We receive some number u of commute paths pat hi, . . . , path u . In one embodiment, the commute paths come from a usage log of a computer service for searching or comparing homes, each commute path may be provided by a different user of the service. For each commute path path k , we compute the travel durations for all home clusters PathDurations(pat f c) according to Section 4.5. Then, we apply an aggregator that processes the travel durations. In one embodiment, for each commute path path k , we receive a vector of weights In one embodiment, each weight is a probability of a user clicking on a home in the corresponding home cluster, which may be affected by a condition for a real estate property or for a travel path specihed in the user request. Then we compute aggregates coordinate-wise, as in the following formula

The j-th aggregate is an average weighted travel duration for home cluster Hj . In one embod iment, the -th aggregate denotes a normalized contribution of the home cluster to the total travel duration of the entire metropolitan area; in a sense, it is a centrality of the home cluster within the metropolitan area. Because our method rapidly computes PathDurations(pat/ ), we can compute the aggregates quickly. This enables a rapid computation of a desirability of each real estate property with respect to commuting.

[132] Many other embodiments of the meta method will be apparent to those of ordinary skill in the art. In other embodiment, the commute paths are generated from geographical locations of homes and works. In other embodiment, the weights are set to non-zero for some number of home clusters with the lowest travel durations, others are set to zero. In other embodiment, a data scientist evaluates what-if scenarios; because of advantages of our method, these scenarios can be evaluated rapidly. In other embodiment, the aggregator is an arbitrary algorithm, for example one that computes a variance, quantiles, a cumulative distribution function, or a probability of exceeding a threshold. 4.9.7 Commute path involving two or more homes

[133] In one embodiment, a commute path involves any two or more homes. This is useful, for example, when searching or comparing homes jointly for two or more families.

[134] In one embodiment, Hi and H 2 are two home clusters. We can compute a travel duration of any commute path Hi— y path— y H 2 that starts at Hi, but ends at H 2 that may be different from H . We simply use a j corresponding to Hi in Equation 1, but use a j corresponding to H 2 in Equation 2.

[135] In one embodiment, we precompute both vectors (a 1 , . . . , a m ) and . . . , b m ), and then we find a travel duration for arbitrary two home clusters Hi and H j by simply adding a value at a coordinate of (cq, . . . , a m ) corresponding to Hi to a value at a coordinate of (&x, . . . , b m ) corresponding to H j , plus the travel duration of the middle part of the path (which is independent of Hi and H j ). We remark that a travel duration for any pair of home clusters, from among the m 2 pairs, can be found in constant 0(1) time using only linear O(m) space. This is because of an additive structure of the travel duration and the fact that the path acts as a separator between any two home clusters.

[136] In one embodiment, the set of home clusters allowed for Hi may be different from the set of home clusters allowed for H 2 . For example, a user condition can restrict Hi to be in the east part of a city, while H 2 to be in the west part of the city. In this case we may use a vector (bi, , b m ) of length m' that is different from m.

[137] More generally, the method computes a travel duration for any sequence of k > 2 home clusters Hi— y pat hi— y H 2 — y path 2 path^ i— y H Using precomputing, this can be done in O(k) time and 0(m k) space.

[138] In one embodiment, a commute path includes a direct travel between home clusters, H x— H y , as in this example commute path: Hi— y pathi— H x— y H y— y path 2— H 2 . In this example, our method partitions the computation into three parts: (1) compute a travel duration of Hi— y pathi— > H x , (2) compute a travel duration of H x— H y , and (3) compute a travel duration of H y — path 2 — H 2 . The parts (1) and (3) can be performed using the method described earlier in the current Section 4.9.7. The part (2) can be performed similarly, conceptually treating either H x or H y as a work. For example, using the vectors v at the coordinate corresponding to H x , we obtain travel durations from the home cluster H x to every stopstation cluster; these can be retrieved as a column of the precomputed vectors n ί ', 1 < i < n. Then we extend from each stopstation cluster within a threshold distance from H u to form a complete travel to H y , similar to Section 4.5.2, see for example FIG. 8 and Equation 1. Because of precomputing due to our method, the (3) can be performed in time proportional to the number of the stopstation clusters within the threshold distance. In other embodiment, because the home clusters are known in advance, we precompute a travel duration between every pair of home clusters, and then simply retrieve a travel duration in constant 0(1) time. Naturally, a similar partitioning applies to any commute path that contains one or more direct travel between home clusters.

4.9.8 Space exploration

[139] Our method introduces high-performance space exploration algorithms. Consider a case when a commuter travels along a specihc commute path Hi— y path— y H 2 which has a specihc cost; the cost function depends on the two homes Hi and H 2 . The commuter seeks homes Hi and H 2 that minimize the cost. This is useful, for example, when a family is willing to change work locations (Hi and H 2 ) of the two parents at the same time, while minimizing the total travel duration from the current home location (inside path ) of the family, aka the two-body problem.

[140] One embodiment of space exploration is a gradient descent algorithm. The cost function can be any differentiable function; for example, given a hxed path, the function takes two homes Hi and H 2 as an input, and returns as an output a distance between the two homes Hi and H 2 multiplied by a travel duration of a commute path Hi— y path— y H 2 (this function can be appropriately extended outside of the discrete domain of home pairs, for example via geographical location of homes and extrapolation). At a given step of the gradient descent algorithm, the algorithm computes a gradient of the cost function given a pair of homes, and then picks a pair of homes along the direction of the gradient for the next step of the algorithm. The gradient can be computed using values of the cost function, for example using the two-point formula that needs two values. In one embodiment, the output of the gradient descent algorithm is a pair of homes where the gradient has a small enough norm.

[141] In one embodiment, we precompute the vectors (oq, . . . , a m ) and . . . , b m ) con suming linear O(m) space, and at each step our method computes the gradient in constant 0(1) time. Thus the gradient descent algorithm can often make rapid progress.

[142] In one embodiment, we do not precompute (oq, . . . , a m ) and (¾, . . . , b m ), but in stead produce the needed values per Equation 1 and Equation 2 on-demand. Such ap proach may be used when the number of homes is so large that precomputing (a 1 , , a m ) or (pi, . . . , b m ) is infeasible.

[143] In one embodiment, a space exploration algorithm may have constraints on homes. For example, if we require a round-trip commute to one home, a constraint will be Hi = H 2 . We may restrict a geographical region allowed for Hi, or allowed for H 2 .

[144] In one embodiment, a space exploration algorithm uses a commute path involving only one home: Hi — y path or path— y Hi. In other embodiment, a space exploration algorithm uses a commute path involving two or more homes, as in Section 4.9.7.

[145] In one embodiment, a cost function depends on homes and works. For example, such cost function can be used when a person seeks to minimize a blend of expended time and money. In one embodiment, the cost function is: the travel duration between H and W, plus the monetary cost of renting H, minus the salary paid by W, possibly with weights that represent relative importance of the constituents. In one embodiment, because of advantages of our method, a gradient descent can efficiently be used to search for homes and works under such cost function.

4.10 General case

[146] In one embodiment, the words“home” and“work” have an arbitrary semantic meaning. For example, consider a case where a person is looking for work that is located near the current home of the person. The person does not want to move to a different house, but merely wants to End a job closer to the current home. In this case, our method can be simply applied. Given a range of sites where people work throughout a metropolitan area (for example various offices, factories, etc.), the method computes travel durations between every work site cluster, and each stopstation cluster. Thus the method can be viewed as a “work search-or-compare using commute” method. In one embodiment, the method searches or compares work sites based on a user-specified job type, or a salary range, and based on a travel duration from the current home of the user. As other example, consider a corporation that wants to move its headquarters to a new location. Our method can be used to compute the total“corporate travel duration” for every new location of the headquarters throughout the metropolitan area. Thus, the corporation may determine how each new location will affect commutes of the employees. A new location may be selected, for example, so that: (1) the worst-case commute duration is limited, and (2) the average commute duration is low; thus satisfying individual and social objectives.

[147] Our description so far mainly talked about a travel duration as the search or com pare objective. However, the method can use any other objective, for example: a monetary cost of travel; a metric distance; specihc features or attributes of travel paths, for example: the number of transfers, or a walking distance; or features of homes, for example: a price, a size, or a type. In one embodiment, this can be simply achieved by building graphs and set ting edge weights appropriately. Various objectives may be combined into a multi-objective optimization search based on a multi-dimensional cost, for example to search for a home that minimizes a travel duration that is penalized by the monetary cost of travel.

[148] In general, the method uses arbitrary sites S , , S m (a site was called a home in earlier sections) and arbitrary places P 1 ; . . . , *. (a place was called a work in earlier sections), and the method searches or compares the sites S , , S m using routes or route lengths (in earlier sections: a route was called a description of travel, and a route length was called a length of travel) that originate or end at some of the sites and visit some of the places, as specihed by arbitrary route specihcations (a route specihcation was called a commute path in earlier sections) involving the sites and the places. The searching or comparing may use any variant described in Section 4.9. Thus the method may respond with routes or route lengths, or their representation. Information computed by one embodiment of the method can be used, recursively, as an input to any embodiment of the method.

4.11 Computer system

[149] One of the embodiments of the invention is a computer system that searches or compares real estate properties using commutes. We illustrate an embodiment of the com puter system in FIG. 11.

[150] We use the term“module” in our description. It is known in the art that the term means a computer (sub)system that provides some specihc functionality. Our choice of partitioning the computer system into the specihc modules is exemplary, not mandatory. 990 Those of ordinary skill in the art will notice that the system can be organized into modules

991 in other manner without departing from the scope of the invention.

992 [151] In one embodiment, each travel along any commute path has a specihc departure

993 time.

994 [152] One module (1101) of the system reads data about the transportation system from

995 a data source (1102), and constructs the graph G. During the construction, the module

996 retrieves data about homes from a real estate data source (1103), and retrieves from other

997 data source (1104) shortest walks between and near home clusters and stopstation clusters.

998 The graph G contains timing data about vehicles. The module outputs the graph without

999 any

1000 HOME CLUSTER SOURCE s vertexes (1105), and also outputs the graph without any

HOME CLU STER TARGET s

1003 vertexes but with reversed edges (1106). The module also builds a nearest neighbor data

1004 structure (1107) that can hnd stopstations within a threshold distance from any given ge

1005 ographical location, and precomputes shortest walks near home clusters and stopstation

1006 clusters (1108).

1007 [153] In the meantime, other module (1109) of the system reads the two graphs, and

1008 computes shortest graph paths. The module considers a range of times during the day, in one

1009 embodiment every 5 minutes. For each departure time, the module generates one table (1110) with shortest travel durations from stopstation clusters to home clusters using (1105), and the other table (1111) with shortest travel durations from home clusters to stopstation clusters using (1106). In one embodiment, each travel duration, rounded to the nearest minute, is

1013 stored as the uint8_t type of the C++ programming language, with the maximum value of

1014 255 reserved to denote an unknown or too large travel duration. In one embodiment, the

1015 tables are laid out on HDD disks in the row-major order. In one embodiment, the system

1016 uses a cache hierarchy involving HDD disks, SSD disks, and the main memory. In one

1017 embodiment, the tables or their parts are compressed using any compression algorithm, for

1018 example delta compression. We observe that for any pair comprising a home cluster and a stopstation cluster, travel durations are often similar during a period of time. This similarity often also holds in a neighborhood of the pair. In one embodiment, we select the sequence hi, . . . , h m of home clusters so that any hi and /q +i that are adjacent in the sequence often are nearby home clusters, or select the sequence Si, . . . , s n of stopstation clusters so that any

1023 Si that are adjacent in the sequence often are nearby stopstation clusters.

1024 [154] The modules (1101) and (1109) operate continuously. As a result, the system

1025 maintains fresh data about travel durations given departure times.

1026 [155] Concurrently, a path durations module (1112) computes PathDurations. Given a

1027 commute path with departure times, the module queries (1113) for any relevant PathDura

1028 tions that has already been precomputed. Any missing one is computed from scratch: the

1029 module queries a navigation data source (1114) to compute the PathNonHomeDuration of

1030 the part of the commute path that does not involve any home. The module also computes the

1031 travel durations PathFromHomeDurations and PathToHomeDurations that involve homes,

1032 by querying the nearest stopstations (1107), walks (1108), and home travel duration vectors

1033 (1110 and 1111) at the departure times.

1034 [156] Concurrently, the request processing module (1115) searches or compares real estate

1035 properties. Any request (1116) contains commute paths including geographical locations

1036 along the commute paths and departure times, and a Deriver. When a request is received

1037 from a user, the module retrieves PathDurations from the path durations module (1112),

1038 applies the Deriver, and responds to the user with information representing an output of the

1039 Deriver (1117).

1040 [157] Aspects of the invention may take form of a hardware embodiment, a software

1041 embodiment, or a combination of the two. Steps of the invention, for example blocks of

1042 any flowchart, may be executed out of order, partially concurrently or served from a cache,

1043 depending on functionality or optimization. Aspects may take form of a sequential system,

1044 or parallel/distributed system, where each component embodies some aspect, possibly re

1045 dundantly with other components, and components may communicate, for example using a

1046 network of any kind. The invention is not described with reference to any specihc program

1047 ming language. A computer program carrying out operations for aspects of the invention

1048 may be written in any programming language, for example C++, Java, or JavaScript. Any

1049 program may execute on an arbitrary hardware platform, for example a Central Processing Unit (CPU) or a Graphics Processing Unit (GPU), and associated memory or storage de- 1051 vices. A program may execute aspects of the invention on one or more software platforms,

1052 including, but not limited to: a smartphone running Android or iOS operating systems, or

1053 a web browser, for example Firefox, Chrome, Internet Explorer, or Safari.

1054 4.12 Computer service

1055 [158] One of the embodiments of the invention is a computer service for searching or

1056 comparing real estate properties using commutes. The service is available to users through a

1057 user-accessed device, for example a smartphone application or a webpage. It will be obvious

1058 to anyone of ordinary skill in the art that the invention is not limited to these devices. It

1059 will also be obvious that the presentation of the service in our drawings can be modihed (for

1060 example by rearranging, resizing, changing colors, shape, adding or removing components)

1061 without departing from the scope of the invention.

1062 [159] In one embodiment, the service is accessed through a smartphone application. See

1063 FIG. 12 for an illustration. The user inputs a request. In one embodiment, the request

1064 includes:

1065 • the desired features of a real estate property (1201), for example“3 bedroom, tall

1066 building, high floor”;

1067 • the commute paths (1202, 1203,’’schools Towsend or Jericho High” 1204), departure

1068 times (1205, 1206), and the frequency of each commute path (1207, 1208); and

1069 • the geographical location of the current home (1209) of the user.

1070 [160] In response, the service returns information representing travel durations. For

1071 example, the service renders geographical locations of real estate properties that match the

1072 user request (1210). The service renders how the travel durations from these real estate

1073 properties compare to the travel duration from the current home. The service renders a

1074 summary about each matching real estate property, for example its price. The real estate

1075 properties may be stacked on a 2D map, so that a real estate property with a lower travel

1076 duration appears above these real estate properties with higher travel durations. When

1077 there is clutter on a map, the service may instead render a cluster of real estate properties, the size of which is correlated with the number of real estate properties in the cluster. The 1079 cluster may display a summary, for example the number, or typical features of the real estate

1080 properties in the cluster.

1081 [161] In one embodiment, the service renders real estate properties with the shortest

1082 travel durations. Summaries of the real estate properties are also rendered (1211). The real

1083 estate properties may be sorted by the travel duration.

1084 [162] In one embodiment, the service renders a“heatmap” that uses a color to indicate

1085 a travel duration from each region of a metropolitan area, for example a minimum for the

1086 region, see FIG. 1 for an illustration. A heatmap may depict a difference between a travel

1087 duration for any home and the travel duration for the current home of the user. In one

1088 embodiment, a heatmap renders only the real estate properties that match the user-desired

1089 features of a real estate property.

1090 [163] In one embodiment, the service renders a histogram of travel durations (1212).

1091 The histogram has the travel duration on one axis, and on the other axis the fraction of real

1092 estate properties that yield this travel duration. In one embodiment, a histogram renders

1093 only the real estate properties that match the user-desired features of a real estate property.

1094 In one embodiment, the user can scroll (1213) through a histogram to any section of the

1095 histogram, and the service renders results for travel durations of this specihc section. In one

1096 embodiment, the method uses other form of a histogram, for example a pie chart.

1097 [164] In one embodiment, the user may restrict the travel paths, for example by limiting

1098 a total walk duration, a number of transfers, etc., for example using range sliders (1214).

1099 [165] In one embodiment, the service renders a summary of a travel path for a real estate property.

[166] In one embodiment, the service responds to the user with at least one of:

(a) a geographical location of: a place, or a site, rendered on a map;;

1103 (b) a departure time or an arrival time for: a place, or a site;;

1104 (c) a summary of a site; a summary may include at least one of: a name or an address of

1105 the site, a price of the site, or a size of the site;;

(d) a summary of sites that form a cluster of nearby sites;; 1107 (e) a rendering of the cluster of nearby sites, a size of the rendering being correlated with

1108 a number of sites in the cluster of nearby sites;;

1109 (f) a stacking of sites in a z-index order of a route length, shorter route length higher up the stacking, rendered on a map;;

(g) information about a route or route length; information may include at least one of: (i) a route: length, duration, monetary cost, speed, or wait duration; (ii) a name of: a

1113 vehicle, a vehicle road, a walking path, a stop, a turn, or a transit station of a public

1114 transportation vehicle; or (iii) a geographical location of: a monetary cost, a speed, a

1115 wait duration, a vehicle, a vehicle road, a walking path, a stop, a turn, or a transit

1116 station of a public transportation vehicle;;

1117 (h) a histogram of a route length across the sites; or a heatmap of a route length across

1118 the sites rendered on a map;;

1119 (i) a minimum route length; a histogram of a minimum route length across the sites; or a heatmap of a minimum route length across the sites rendered on a map;;

(j) a weighted route length; a histogram of a weighted route length across the sites; or a heatmap of a weighted route length across the sites rendered on a map;;

1123 (k) a difference route length; a histogram of a difference route length across the sites; or a

1124 heatmap of a difference route length across the sites rendered on a map;;

1125 (l) a rendering of an output returned by the Deriver;;

1126 (m) a site that minimizes a route length across the sites, under a limit on a monetary cost

1127 of travel;;

1128 (n) one of the above items (a) to (m) constrained by a condition specihed by a user;; or

1129 (o) a top list of sites, from among these sites, routes, and route lengths that satisfy the condition, sorted by a route length. 1131 4.13 Claims

1132 [167] Those of ordinary skill in the art shall notice that various modihcations may be

1133 made, and substitutions may be made with essentially equivalents, without departing from

1134 the scope of the present invention. Besides, a specihc situation may be adapted to the

1135 teachings of the invention without departing from its scope. Therefore, despite the fact

1136 that the invention has been described with reference to the disclosed embodiments, the

1137 invention shall not be restricted to these embodiments. Rather, the invention will include

1138 all embodiments that fall within the scope of the appended claims.

1139 4.14 Glossary

1140 [168] We include a glossary of selected phrases that occur in the claims, and example

1141 references to the specihcation. These references are not intended to be exhaustive; other

1142 references exist. The selected phrases follow the order in which the phrases hrst appear in

1143 the claims.

1144 [169] The claims also use the following phrases, whose meaning we explain:

1145 1. a phrase“at least one A” is equivalent > 1, wherein FA is a number of A”;

1146 2. a phrase“one or more A” is equivalent > 1, wherein F is a number of A”;

1147 3. a phrase“plurality of As” is equivalent to“#A > 2, wherein F is a number of A”;

1148 4. a phrase“at least two As” is equivalent to“#A > 2, wherein F is a number of A”;

1149 5. a phrase“one of: A, or B” is equivalent to“#A + yB = 1, wherein F is a number of

1150 A, and B is a number of B”;

1151 6. a phrase“at least one of: A, or B” is equivalent to “#A + B > 1, wherein F is a

1152 number of A, and B is a number of B”; and

1153 7. a phrase“at least one B or C” is equivalent to“at least one A, wherein each A is (B or C)”.