Keras jupyter notebook outputs blocks during training

Question

When using Keras in a Jupyter Notebook environment, during training my Notebook crashes. While running the fit function, apart from the normal training progress, a lot of squares appear. In fact, there are so many squares the browser crashes...

When running the same code in a normal terminal these squares do not appear.

Here is an example of the output:

Train on 33164 samples, validate on 8292 samples
Epoch 1/3
  960/33164 [..............................] - ETA: 46s - loss: 0.1997 - acc: 0.9394 

Hope anybody can help!

Answer 1

see this issue on github

in short 4 options :

• Use command lines instead of ipython notebooks
• Use verbose=0 or verbose=2 in fit(), the first will not output anything, the second will only output one log line per epoch.
• If you use tensorflow as backend you can monitor everything on tensorboard (slow but better than nothing)
• You can use the RemoteMonitor callback but this seems to require a bit of work.

I hope this helps.