sermakarevich/jigsaw-toxic-comment-classification-challenge

jigsaw-toxic-comment-classification-challenge's Issues

parallelize_dataframe unable to resolve passed function

Running on Windows 10. I am adding the code from your original Jupyter notebook into my own, and the only other change is to redefine paths on the notebook and config.py for my own situation.

When train = parallelize_dataframe(train, process_comment) is run then the code hangs indefinitely. In the jupyter notebook console I can see that parallelize_dataframe cannot actually resolve the passed function, so nothing is actually happening:

AttributeError: Can't get attribute 'process_comment' on <module 'main' (built-in)>

Could this be a peculiarity of my environment?

UPDATE - Definitely a peculiarity of my environment:
Is it possible to define parallelize_dataframe so it copes with a Windows environment (see below)?

See:
https://stackoverflow.com/questions/37103243/multiprocessing-pool-in-jupyter-notebook-works-on-linux-but-not-windows
https://docs.python.org/2/library/multiprocessing.html#windows

Resolved by wrapping if __name__ == '__main__'

def parallelize_dataframe(df, func):
    if __name__ == '__main__':
        df_split = np.array_split(df, multiprocessing.cpu_count())
        pool = Pool(multiprocessing.cpu_count())
        df = pd.concat(pool.map(func, df_split))
        pool.close()
        pool.join()
    return df 

tensorflow requirement should be tensorflow-gpu

The requirements file lists tensorflow==1.6.0. However pip install only installed the CPU version, I had to install tensorflow-gpu to get the models to work. Consider amending the requirements file - would tensorflow-gpu==1.6.0 work?

GRU not able Build

I run into InvalidArgumentError when my notebook gets to the GRU section.

InvalidArgumentError (see above for traceback): No OpKernel was registered to support Op 'CudnnRNN' with these attrs.  Registered devices: [CPU], Registered kernels:
  <no registered kernels>

	 [[Node: bidirectional_1/CudnnRNN_1 = CudnnRNN[T=DT_FLOAT, direction="unidirectional", dropout=0, input_mode="linear_input", is_training=true, rnn_mode="gru", seed=87654321, seed2=0](bidirectional_1/transpose_2, bidirectional_1/ExpandDims_3, bidirectional_1/Const_1, bidirectional_1/concat_1)]]