Strange histograms

[jojonki]

Hi, this is really nice project. I started to use this library because I wanted to see histograms of my trainable parameters.

Although I run demo.py, the histograms looks strange. Is it expected behaviors?

• tensorboardX: 0.8
• python 3.6.1 (Anaconda)
• tensorflow: 1.4.0
• tensorflow-tensorboard: 0.4.0rc3
• PyTorch: 0.3.0.post4

$ python demo.py
$ tensorboard --logdir runs

[lanpa]

Please try #42
I also noticed that you have a value 3*10^18, which is unusually big. Did the gradient explode?

[jojonki]

Do you mean to set bins like bins=np.arange(-0.003, 0.003, 0.0001)? Ok, I'll try it.
But I am afraid your official code doesn't work in some environment. Other libraries in my environment may be quite new.

[lanpa]

No, I think your value is too concentrated, therefore default binning method can not plot well. You can check this by printing the max and min of your data in console. And sorry I can't quite understand your second sentence. Is there error message?

[jyzhang-bjtu]

I got the same strange histogram too...

[jyzhang-bjtu]

Got it. The function add_histogram should use bins='auto' as default other than bins='tensorflow'.
The para bins='tensorflow' always cause such strange histogram

[lanpa]

@jyzhang-bjtu 'auto' might freeze the code. see #1 and use with care:)