Python classifier.py crash when running classification.ipynb

[dgmp88]

I was getting the following error running the classification.ipynb notebook on a fresh installation on OSX:

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-4-7167c0e21af1> in <module>()
----> 1 prediction = net.predict([input_image])  # predict takes any number of images, and formats them for the Caffe net automatically
      2 print 'prediction shape:', prediction[0].shape
      3 plt.plot(prediction[0])
      4 print 'predicted class:', prediction[0].argmax()

/Users/george/Documents/caffe/python/caffe/classifier.py in predict(self, inputs, oversample)
     84             caffe_in[ix] = self.transformer.preprocess(self.inputs[0], in_)
     85         out = self.forward_all(**{self.inputs[0]: caffe_in})
---> 86         predictions = out[self.outputs[0]].squeeze(axis=(2,3))
     87 
     88         # For oversampling, average predictions across crops.

ValueError: 'axis' entry 2 is out of bounds [-2, 2)

which I was able to fix by deleting the arguments for squeeze so that 86 looks like this:

predictions = out[self.outputs[0]].squeeze()

[dgmp88]

Hmm, though doing this means that in the next section of classification.ipynb, when oversampling is switched off, the results don't make sense:

print 'prediction shape:', prediction[0].shape
plt.plot(prediction[0])
print 'predicted class:', prediction[0].argmax()

prediction shape: ()
predicted class: 0

unless I remove the indexing:

print 'prediction shape:', prediction.shape
plt.plot(prediction)
print 'predicted class:', prediction.argmax()

prediction shape: (1000,)
predicted class: 281

[seanbell]

I think that the problem is that some of the Python code hasn't yet been updated to match #1970 which was recently merged into master.

Perhaps there should be tests for python/caffe/classifier.py and all the steps in examples/classification.ipynb. Ideally, these kinds of bugs would be identified by Travis before major changes are merged.

[shelhamer]

@seanbell is right about the cause. We haven't come up with a good method for auto-checking the notebooks but these would be valuable integration checks that catch these kinds of issues. I suppose we could at least check they not crash.

There hasn't been a chance to fix these yet but this is on the list. If anyone has fixed these already, PRs are welcome.

[dgmp88]

Done. I don't know why squeeze was specifying the axis though, so hope I'm not breaking anything.

[latharam]

I am facing the same problem but with detector.py

Am getting the same error:
predictions = out[self.outputs[0]].squeeze(axis=(2,3))
ValueError: 'axis' entry 2 is out of bounds [-2, 2)

removing the arguments from
predictions = out[self.outputs[0]].squeeze(axis=(2,3))
and making it
predictions = out[self.outputs[0]].squeeze()
fixes this problem but later

when I execute the following piece of code from http://nbviewer.ipython.org/github/bvlc/caffe/blob/master/examples/detection.ipynb

---->predictions_df = pd.DataFrame(np.vstack(df.prediction.values), columns=labels_df['name'])
print(predictions_df.iloc[0])

I got the following error
Traceback (most recent call last):
File "ipynotebuk.py", line 22, in
predictions_df = pd.DataFrame(np.vstack(df.prediction.values), columns=labels_df['name'])
File "/home/lpemula/LPIR_Caffe/env/lib/python2.7/site-packages/pandas/core/frame.py", line 239, in init
copy=copy)
File "/home/lpemula/LPIR_Caffe/env/lib/python2.7/site-packages/pandas/core/frame.py", line 399, in _init_ndarray
return create_block_manager_from_blocks([values.T], [columns, index])
File "/home/lpemula/LPIR_Caffe/env/lib/python2.7/site-packages/pandas/core/internals.py", line 3534, in create_block_manager_from_blocks
construction_error(tot_items, blocks[0].shape[1:], axes, e)
File "/home/lpemula/LPIR_Caffe/env/lib/python2.7/site-packages/pandas/core/internals.py", line 3515, in construction_error
passed,implied))
ValueError: Shape of passed values is (1000, 1570), indices imply (200, 1570)

If I modify it as below,
predictions_df = pd.DataFrame(np.vstack(df.prediction.values))

It runs fine but I am not sure if what I did is right. Please let me know the right fix for this.