Tensorboardx: Need support for pytorch 0.4.0

[yaodi833]

[GuillaumeLeclerc]

What is not working ? I am pretty sure this library is framework agnostic. I am using it with pytorch 0.4.0

[JindongJiang]

    if isinstance(x, torch.autograd.Variable):
        x = x.data

in x2num.py is no longer needed. It caused an error in my code, so I commented it out, but I forget how to reproduce the error.

[lanpa]

@JindongJiang Before trying to reproduce, did you uncomment then test again?

[JindongJiang]

Hi, before trying to reproduce, I didn't uncomment it but I edited it

if isinstance(x, torch.Tensor):
    x = x.data

and I still don't remember why I did this.

But I found another computer just now trying to reproduce it given that my computer is training a network. And I couldn't reproduce it also, it works well without commented. So I figure it might be just a mistake in my coding.

Thanks.

[jalola]

Hi,

I got the same error when using tensorboardX 0.8.
After upgrading to latest tensorboardX 1.2 I believe, the error is gone.

How to upgrade: pip install tensorboardX --upgrade

Error trace if you still need it:
Traceback (most recent call last): File "train.py", line 309, in <module> train() File "train.py", line 257, in train writer.add_scalar('data/cost', cost, iteration) File "/root/miniconda3/lib/python3.6/site-packages/tensorboardX/writer.py", line 262, in add_scalar self.file_writer.add_summary(scalar(tag, scalar_value), global_step) File "/root/miniconda3/lib/python3.6/site-packages/tensorboardX/summary.py", line 87, in scalar scalar = makenp(scalar) File "/root/miniconda3/lib/python3.6/site-packages/tensorboardX/x2num.py", line 14, in makenp return pytorch_np(x, modality) File "/root/miniconda3/lib/python3.6/site-packages/tensorboardX/x2num.py", line 22, in pytorch_np if isinstance(x, torch.autograd.variable.Variable): AttributeError: 'function' object has no attribute 'Variable'

[JindongJiang]

@jalola Hi,

That's the exact same error I got earlier. I remember it now.

Thank you so much