Add N-distill - Githubissues

DeepX-inc / machina

Control section: Deep Reinforcement Learning framework

MIT License

278 stars 45 forks source link

Add N-distill #176

Open pwuethri opened 5 years ago

pwuethri commented 5 years ago

Adding N-distill according to https://arxiv.org/abs/1902.02186

[x] Add next observation to trajectory data structure
[x] Directly compute gradient using the given update rule (This is the difference compared to Teacher distill, on-policy distill and entropy regularised distillation
[x] Update nn parameters accordingly
[ ] Test is policy distillation works using available teacher policy

pwuethri commented 5 years ago

Script exits with terminate called after throwing an instance of 'std::runtime_error what(): invalid argument 13: ldc should be at least max(1, m=0), but have 0 at /pytorch/aten/src/TH/generic/THBlas.cpp:334 [1] 24572 abort (core dumped) python run_n_distill.py -> need to check the computed gradients