Closed prabhatnagarajan closed 3 years ago
There are trained Models now, but one is not able to download it. If I run the reproduction example for A3C with pretrained model demo, I got a HTTP Error "403: Forbidden" for the URL https://chainer-assets.preferred.jp/chainerrl/A3C/Breakout\final.zip
I think this is a convenience with joining paths. Are you joining the URL-Path with os.path.join? When it's coming to on-disk-paths python can handle \ or / in whatever OS it is. On windows os.path.join joins with \ and on unix systems it joins with /. When you join a URL on windows it joins a \ instead a /.
When i replace the \ (right before final.zip in the upper URL) with an / I can access the file manually.
Maybe you can fix it in your code?
You're correct that we use join: https://github.com/chainer/chainerrl/blob/master/chainerrl/misc/pretrained_models.py#L124
cache_path = cached_download(os.path.join(url_basepath,file))
Where file
is final.zip
.
Thanks for sharing! I'll try and amend this.