Update Dockerfile to add Catalyst

[Scitator]

Hi, just want to add Catalyst - high-level PyTorch framework for reproducible and fast R&D.

[rosbo]

Hi @Scitator,

Can you add tests to prevent regression. Examples can be found under the /tests folder in this repo.

Thank you

[Scitator]

Hi @rosbo

So minimal test case is done ;)
All other test are done during catalyst CI.

[rosbo]

Hi @Scitator,

If you can add a simple smoke test that exercise a common path of catalyst that would be great.

Because we are installing 1000s of packages in this image, some dependencies may get reinstalled to a different version which may cause you package to fail inside the Kaggle image even though it works in your own CI (given the dependency may differ).

Thank you

[Scitator]

@rosbo Would mnist test be enough?

[Scitator]

@rosbo Could you please check example now?

[Scitator]

Hmm...why do you think tests have failed?

[rosbo]

@Scitator

======================================================================

ERROR: test_version (test_catalyst.TestCatalyst)

----------------------------------------------------------------------

Traceback (most recent call last):

  File "/input/tests/test_catalyst.py", line 85, in test_version

    self.assertIsNotNone(catalyst.__version__)

NameError: name 'catalyst' is not defined

[rosbo]

It seems like the installation probably failed. I will look at the build logs.

[rosbo]

The logs seems to suggest the package was successfully installed:
Sep 26 22:03:20 Successfully installed catalyst-19.9.5 crc32c-1.7 safitty-1.2.5: Successfully installed catalyst-19.9.5.

The image it ran the tests against is gcr.io/kaggle-images/python:PR-616-pretest. I will pull it and inspect it to see if the catalyst package is present.

[rosbo]

Do you need to call some sort of Close method?

The tests shows a ResourceWarning:

unclosed file <_io.TextIOWrapper name='/working/logs/log.txt' mode='a' encoding='UTF-8'>

[rosbo]

Ah. I found the issue.

The package is properly installed on the image:

$ pip show catalyst
Name: catalyst
Version: 19.9.5
Summary: Catalyst. High-level utils for PyTorch DL & RL research.
Home-page: https://github.com/catalyst-team/catalyst
Author: Sergey Kolesnikov
Author-email: [email protected]
License: Apache License 2.0
Location: /opt/conda/lib/python3.6/site-packages
Requires: opencv-python, tensorboardX, matplotlib, scikit-image, crc32c, pandas, PyYAML, torchvision, numpy, tqdm, scikit-learn, safitty, torch, imageio, plotly, seaborn

$ python
>>> import catalyst
>>> print(catalyst.__version__)
19.09.5

You only need to add the missing import statement (see my inline code comment).

[Scitator]

@rosbo Looks like all checks have passed :)
unclosed file issue will be resolved in Catalyst 19.10 release