Multi-Dimensional Rail Transit Passenger Flow Prediction

This is the Numpy and Pandas implementation of our solution to the Multi-Dimensional Rail Transit Passenger Flow Prediction competition topic [Link] in the 2nd Guangzhou Pazhou Algorithm Competition [Link].

We came up with this solution from two basic observations of the dataset:

1. Passenger flow on the same day of every week is likely to share the same "pattern".
2. Passenger flow on adjacent days are likely to have similar "amount".

We are proud that our simple but beautiful solution, which is based on "selective" historical average with learned normalization/denormalization (we implemented it with PaddleTS in the competition as required), knocked out various well-designed deep learning based solutions and finally ranked the 3rd on the private leaderboard of this competition topic.

Although this solution was later manually marked down to the 7th by the competition organizers simply because of NOT BEING DEEP LEARNING, we still strongly believe the outstanding performance of it urges us to rethink about the recent trend in deep learning and its application in spatio-temporal data.

With :heart: from GuanyaoLI, zshhans, Sausage.