Collecting
Next snapshots
Stored
Car snapshots per day
Seat model
How likely a free seat is to be taken by departure, in two steps: how many of a car's free seats will sell, learned from how cars actually left, then which of them go first, learned from every pair of snapshots. Retrained each night; its predictions are logged as the timer takes its snapshots, then scored against how the trains actually left.
Step 1: share of a car's free seats sold by departure, by time before departure
For a car none of the factors below apply to (a midday train midweek, car and train not yet filling up).
Step 1: how much each factor changes that
× the odds a free seat in the car is sold by departure (1 = no effect).
Step 2: which seats go first
× the rate a free seat is taken, against one without that factor (1 = no effect). The car's expected bookings are shared out over its free seats in this order.
How right it is
Departures
Every departure we have seat data for. Pick one to see how its cars filled up.