Nanmean is deprecated, imported, unused, and never tested

[thomas-reimonn]

In predict.py nanmean is imported from scipy. nanmean was deprecated and removed from scipy.stats version 0.18.0: https://stackoverflow.com/questions/40944243/module-scipy-stats-has-no-attribute-nanmean and the version is not pinned in requirements.txt

Additionally, nanmean is never used in predict.py so it can just be removed.

There appears to be no testing process that catches this, which prevent the bias pipeline from finishing.

[tjgibson]

I have the same issue. When I run chrombpnet bias pipeline using the most recent docker image, it fails after training and saving the best model. Here is the error:

Traceback (most recent call last):
  File "/opt/conda/bin/chrombpnet", line 33, in <module>
    sys.exit(load_entry_point('chrombpnet', 'console_scripts', 'chrombpnet')())
  File "/scratch/chrombpnet/chrombpnet/CHROMBPNET.py", line 38, in main
    pipelines.train_bias_pipeline(args)
  File "/scratch/chrombpnet/chrombpnet/pipelines.py", line 329, in train_bias_pipeline
    import chrombpnet.training.predict as predict
  File "/scratch/chrombpnet/chrombpnet/training/predict.py", line 14, in <module>
    from scipy import nanmean, nanstd
ImportError: cannot import name 'nanmean' from 'scipy' (/opt/conda/lib/python3.9/site-packages/scipy/__init__.py)