Skip to main content
Uber logo

Schedule rides in advance

Reserve a rideReserve a ride

Schedule rides in advance

Reserve a rideReserve a ride
Data / ML

Gleaning Insights from Uber’s Partner Activity Matrix with Genomic Biclustering and Machine Learning

December 7, 2017 / Global
Featured image for Gleaning Insights from Uber’s Partner Activity Matrix with Genomic Biclustering and Machine Learning
Figure 1: In an example Partner Activity Matrix, each row represents one partner, and each column represents one minute of the week ordered by day of week (Monday-Sunday). The top X-axis shows day of week and the bottom hour of day, with white indicating the partner was online and black offline.
Figure 2: In the matrix above, times with more driver activity are highlighted in dark grey. For example, the first pattern shows drivers driving during weekday morning hours, the second pattern in the middle of the day throughout the week, and the third pattern in evenings Thursday through Saturday.
Figure 3: In matrix U, the hours driven by partner j combine patterns 2 and 3, and so include the middle of the day on all days of week and the evenings on Thursday, Friday, and Saturday. The hours driven by partner i are those from driving pattern 1: weekday mornings.
Image
Figure 4: We use SVD to calculate the decomposition and approximation of the partner activity matrix.
Figure 5: We can visualize the reshuffling the pattern weights to place partners in the same cluster adjacent to each other. In Figure 5, above, there are five clusters, each of which follows a distinct collection of driving patterns.
Figure 6: Our daytime drivers’ partner activity matrix, which includes only those rows from the partner activity matrix corresponding to drivers in the daytime driver cluster, shows that online time, colored as white, occurs mainly from 6:00 a.m. to 6:00 p.m. on both weekdays and weekends, with fewer partners going online during weekends.
Figure 7: Partners in the above daytime driver cluster typically drive between 6:00 a.m. and 6:00 p.m., with more drivers online during weekdays, especially morning and evening rush hour, and fewer during weekends.
Figure 8: On our nighttime driver partner activity matrix, online time (in white) occurs mainly from 6:00 p.m. to midnight on weekdays and extends to roughly 3:00 a.m. on Friday and Saturday nights.
Figure 9: The ratio of nighttime drivers online per hour of the day indicate that some partners drive between 6:00 p.m. and 1:00 a.m. on weekdays, extending to 2:00 a.m. on Friday and Saturday nights.
Peter Frazier

Peter Frazier

Peter Frazier is a staff data scientist on Uber's Marketplace Pricing team.

Posted by Qing Feng, Peter Frazier

Category: