add_histgram : TypeError about tensorboardx HOT 17 CLOSED

I could not locate the corresponding code in your error message in newest (the version on github) tensorboardX . Which version are you on?

from tensorboardx.

I use pip3 install tensorboardX, which version is 0.8. I will update it later.

from tensorboardx.

The pip version is pretty old. XD
Please try pip install git+https://github.com/lanpa/tensorboard-pytorch if you have time~

from tensorboardx.

I upgrade the version from 0.8 to 0.9. But the problem still exist.

from tensorboardx.

So what is the error message in V0.9?

from tensorboardx.

same as V0.8. Given as follows, where fun_wr is my wrapper class to write summary.
File "/tchelpers.py", line 346, in write
fun_wr(mn0,data[mn0], niter)
File "/opt/pytorch/lib/python3.5/site-packages/tensorboardX/writer.py", line 323, in add_histogram
self.file_writer.add_summary(histogram(tag, values, bins), global_step)
File "/opt/pytorch/lib/python3.5/site-packages/tensorboardX/summary.py", line 114, in histogram
hist = make_histogram(values.astype(float), bins)
File "/opt/pytorch/lib/python3.5/site-packages/tensorboardX/summary.py", line 132, in make_histogram
bucket=counts)
TypeError: 0 has type numpy.int64, but expected one of: int, long, float

from tensorboardx.

I will try to debug it later if my training machine is free...

from tensorboardx.

Hi, what is the output of type(mn0), type(data[mn0]).
(a small example to produce the bug will be great help)

from tensorboardx.

I will provide sample data to you later because my machine is not free...

from tensorboardx.

The attached is my data (numpy *.npz file in zip file) to be used in the function "make_histogram" in the summary.py.

The error arises in the last para of "HistogramProto" function, which is bucket=counts.

I hope this will help you to debug the program.

data.zip

from tensorboardx.

no error on my machine :(

import numpy as np
from tensorboardX import SummaryWriter

data = np.load('test.npz')
bins = data['bins']
data = data['values']
print(data.shape)
with SummaryWriter() as w:
    w.add_histogram('data', data, 0, bins=bins)
    w.add_histogram('data', data, 1)
    

from tensorboardx.

I use the code above but the error is still there. The error msg is same as the before.
Is this due to the version of numpy? My numpy version is 1.13.0.

from tensorboardx.

I debug it again and again. I found that error arose in
HistogramProto(min=values.min(),max=values.max(), num=len(values),sum=values.sum(),sum_squares=sum_sq,bucket_limit=limits,bucket=counts).

because the type of counts is np.int64 array where is given by

counts, limits = np.histogram(values, bins=bins) in summary.py.

If I replace counts with a list, such as [1,2], everything is ok.

or use as
HistogramProto(min=values.min(),max=values.max(), num=len(values),sum=values.sum(),sum_squares=sum_sq,bucket_limit=limits,bucket=counts.tolist()) is ok.

It seems that my numpy can not automatically convert an numpy array to list of ints.

from tensorboardx.

Hi, my counts.dtype is also numpy.int64. Based on your description, it looks like the protobuf does not convert ndarray to the datatype it expect. Btw, have you run demo.py before? Did it break the code? Thanks

from tensorboardx.

If use demo.py, got the same error as follows

Traceback (most recent call last):
File "demo.py", line 35, in
writer.add_histogram(name, param, n_iter)
File "/opt/pytorch/lib/python3.5/site-packages/tensorboardX/writer.py", line 323, in add_histogram
self.file_writer.add_summary(histogram(tag, values, bins), global_step)
File "/opt/pytorch/lib/python3.5/site-packages/tensorboardX/summary.py", line 114, in histogram
hist = make_histogram(values.astype(float), bins)
File "/opt/pytorch/lib/python3.5/site-packages/tensorboardX/summary.py", line 132, in make_histogram
bucket=counts)
TypeError: 0 has type numpy.int64, but expected one of: int, long, float
Exception ignored in: <function WeakValueDictionary.init..remove at 0x7f6f077a5378>
Traceback (most recent call last):
File "/opt/pytorch/lib/python3.5/weakref.py", line 117, in remove
TypeError: 'NoneType' object is not callable

from tensorboardx.

After I upgraded the protobuf from 3.0.0 (in debian distribution) to the latest version (3.5.0), everything is ok.

from tensorboardx.

I reproduced it on a VM and I found that the minimum working version is 3.2.0. Thanks!

from tensorboardx.