search

Kaixhin / Rainbow

Rainbow: Combining Improvements in Deep Reinforcement Learning
MIT License
1.56k stars 282 forks source link

Use preallocated tensor in replay memory #5

Closed Kaixhin closed 6 years ago

Kaixhin commented 6 years ago

Using a preallocated tensor (the size of which is known) in the replay memory instead of doing CUDA casts and hence allocating new memory should provide a speed boost (maybe)?

Kaixhin commented 6 years ago

Closed as the batch size changes due to invalid samples from the prioritised replay memory.