-
-
Notifications
You must be signed in to change notification settings - Fork 654
Closed
Labels
Description
🚀 Feature
In certain use-cases, for example in RL or multi-data cases, it does not make much sense to call engine.run
with an artificial data that only defines epoch lenght:
trainer.run(data, max_epochs=100)
For example,
timesteps = list(range(10000)) |
trainer.run(timesteps, max_epochs=args.max_episodes) |
The idea is to make data
argument optional. Proposed API
trainer.run(max_epochs=100, epoch_length=10_000)
In this case, training_step(engine, batch)
will receive None
as batch value.
cc @sdesrozis
Ishan-Kumar2 and bibhabasumohapatra