Skip to content

Latest commit

 

History

History
13 lines (7 loc) · 833 Bytes

README.md

File metadata and controls

13 lines (7 loc) · 833 Bytes

kdd-cup-2017

Repo for KDD Cup 2017

URL: https://tianchi.aliyun.com/competition/information.htm?spm=5176.100067.5678.2.JJvKYv&raceId=231597

Participants: Hao Peng, Xiao Zhang, Lianjie Cao

Change log

  1. Training data is parsed and dumped with format: win_start,time_of_win,weekday,routeN_0(tollgateN_Dir_0), ..., routeN_5(tollgateN_Dir_5), y_routeN_0(tollgateN_Dir_0), ... y_routeN_5(tollgateN_Dir_5). Each data point is 1*(3+2*6*n) dimentional vector, where n=5 for volume and n=6 for travel time. A data point is created for each 20-min time window. Values are set to 0 if trajectory information is missing for a certain time window. In the previous version, missing time windows were ignored.
  2. Testing data uses the same format. But there is only one data point for each prediction window (6 20-min windows).