Imitation learning cart

Game

Game 1

Your data

Games recorded 0
You caught –
Train / validation 0 / 0

Play a few games to collect data.

Training

not trained
Train accuracy –
Validation accuracy –
Network catch rate –
train accuracy validation accuracy network catch rate

Decisions

Network

positive weight negative weight
data: (s, a) every 0.05 s · a = your key
p(a | s) = softmax(network(s))
loss = −log p(your action | s)
every 5th game is held out for validation