From 5d4bf035f638c45c7bf16eb162687ae55a666a1e Mon Sep 17 00:00:00 2001 From: Ethan Caballero Date: Tue, 14 Mar 2017 02:10:00 -0700 Subject: [PATCH] USE V-0.1.9 (or lower) OF PYTORCH, AND NOT V-0.1.10 NEED TO USE V-0.1.9 (or lower) OF PYTORCH, AND NOT V-0.1.10 BECAUSE OF THIS ISSUE: https://discuss.pytorch.org/t/problem-on-variable-grad-data/957/7 Although I'm not positive. Maybe it somehow does still work with V-0.1.10 even though you can't manually reinitialize the gradient. --- README.md | 3 +++ 1 file changed, 3 insertions(+) diff --git a/README.md b/README.md index 5809b56..7566282 100644 --- a/README.md +++ b/README.md @@ -1,5 +1,8 @@ # pytorch-a3c +NEED TO USE V-0.1.9 (or lower) OF PYTORCH, AND NOT V-0.1.10 BECAUSE OF THIS ISSUE: +https://discuss.pytorch.org/t/problem-on-variable-grad-data/957/7 + This is a PyTorch implementation of Asynchronous Advantage Actor Critic (A3C) from ["Asynchronous Methods for Deep Reinforcement Learning"](https://arxiv.org/pdf/1602.01783v1.pdf). This implementation is inspired by [Universe Starter Agent](https://github.com/openai/universe-starter-agent).