wrong histograms

Hi, I am having problems plotting histograms. I think there is a very good chance that it is not because of a bug in tensorboard-pytorch, but I’m not sure what I could be doing wrong, and I’m not sure where to ask, so if someone could help I would appreciate it.

I am trying to plot histograms of the gradients like this:

loss.backward()
for n, p in filter(lambda np: np[1].grad is not None, spectral_model.named_parameters()):
    print(n, p.grad.data.min(), p.grad.data.max())
    summary_writer.add_histogram(n, p.grad.data.cpu().numpy(), global_step=step)

The mins and maxes show that the values are all between -.15 and .15 (and in fact most values are much closer to zero than that). But the histograms seem to show that all the values are located at one extremely high value, like 3.01e+18:

[greaber]

Thanks. For now I guess I will just stick to ‘doane’ since I don’t want to hand set the binning for each parameter. Actually, I am finding the distributions tab more useful than the histograms tab for my puproses anyway (basically I am interested in monitoring spiking gradients that require clipping and maybe also gradients going to zero). We can close this issue if you want. There is definitely some buggy behavior with the default binning, but if tensorboard is doing the binning by default I guess it is probably a tensorboard issue.

[lanpa]

Using auto will explode the memory (should be numpy bug). see #1 . From you last graph, I think bins=np.arange(-0.003, 0.003, 0.0001) might do the work.