Pytorch 0.2 doesn't work with newest Visdom
Maosef opened this issue · 4 comments
Maosef commented
When getting the training started, I get something along the lines of
File "/scratch1/anaconda2/envs/road/lib/python3.6/site-packages/visdom/__init__.py", line 1171, in line
assert Y.ndim == 1 or Y.ndim == 2, 'Y should have 1 or 2 dim'
AttributeError: 'torch.DoubleTensor' object has no attribute 'ndim'
Looking into it (https://github.com/facebookresearch/visdom/blob/master/py/visdom/__init__.py), their support for PyTorch < 0.4 is deprecated. What version worked here?
gurkirt commented
Like mentioned in the description, the code is compatible with pytorch 0.3. Please follow the instructions
Maosef commented
The pytorch version mentioned in the beginning of the readme is 0.2. It
then says it would be compatible with "0.03" (typo?) with a "few fixes" but
those aren't described in detail later.
…On Wed, Jan 2, 2019 at 12:09 PM Gurkirt Singh ***@***.***> wrote:
Like mentioned in the description, the code is compatible with pytorch
0.3. Please follow the instructions
—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
<#30 (comment)>,
or mute the thread
<https://github.com/notifications/unsubscribe-auth/ATDtUZep7f-i4Zhrgv19MRue6JNBjqOhks5u_OfegaJpZM4ZlJXV>
.
gurkirt commented
The code was developed within at that time visdom and pytorch 0.2 and 0.3 is not much different. I think you are getting problem form newer visdom. The error you are getting is a typecasting error. I would suggest you try some typecasting. Look at the type of current variables required in newer visdom.
Maosef commented
Typecasting is what I ended up doing, unfortunately the visdom display
remained blank. I tested their demo to be sure, which did display. The rest
of the code proceeded normally.
…On Thu, Jan 3, 2019 at 6:44 AM Gurkirt Singh ***@***.***> wrote:
The code was developed within at that time visdom and pytorch 0.2 and 0.3
is not much different. I think you are getting problem form newer visdom.
The error you are getting is a typecasting error. I would suggest you try
some typecasting. Look at the type of current variables required in newer
visdom.
—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
<#30 (comment)>,
or mute the thread
<https://github.com/notifications/unsubscribe-auth/ATDtUY4KPOyGMuAc0lMOqAcRV2IG3sDsks5u_e0WgaJpZM4ZlJXV>
.