instadeepai / Mava

🦁 A research-friendly codebase for fast experimentation of multi-agent reinforcement learning in JAX
Apache License 2.0
737 stars 90 forks source link

[BUG] Crash due to floating point instead of int #1009

Closed EdanToledo closed 9 months ago

EdanToledo commented 9 months ago

Describe the bug

The timestep variable $t$ is sometimes a float and other times an int. It depends on the configs.

Super simple fix:

wherever you see :

t = steps_per_rollout * (eval_step + 1)

make it:

t = int(steps_per_rollout * (eval_step + 1))

Expected behavior

A crash not happening.