Add rapidsai (cudf and cuml)

[jperez999]

No description provided.

[jperez999]

To run jupyterlabs in the container and have access to the gpu in the notebooks I had to use the following docker run command:

docker run --runtime nvidia --rm -it -e HOME=/tmp -w=/tmp -v /tmp/empty_dir:/usr/local/cuda/lib64/stubs:ro -p 8888:8888 -p 8787:8787 -p 8786:8786 kaggle/python-gpu-build

[Philmod]

The CI tests fail with:

======================================================================
ERROR: test_cudf (unittest.loader._FailedTest)
----------------------------------------------------------------------
ImportError: Failed to import test module: test_cudf
Traceback (most recent call last):
  File "/opt/conda/lib/python3.6/unittest/loader.py", line 428, in _find_test_path
    module = self._get_module_from_name(name)
  File "/opt/conda/lib/python3.6/unittest/loader.py", line 369, in _get_module_from_name
    __import__(name)
  File "/input/tests/test_cudf.py", line 2, in <module>
    import cudf as gd
ModuleNotFoundError: No module named 'cudf'


======================================================================
ERROR: test_cuml (unittest.loader._FailedTest)
----------------------------------------------------------------------
ImportError: Failed to import test module: test_cuml
Traceback (most recent call last):
  File "/opt/conda/lib/python3.6/unittest/loader.py", line 428, in _find_test_path
    module = self._get_module_from_name(name)
  File "/opt/conda/lib/python3.6/unittest/loader.py", line 369, in _get_module_from_name
    __import__(name)
  File "/input/tests/test_cuml.py", line 4, in <module>
    from cuml import LinearRegression as cuLinearRegression
ModuleNotFoundError: No module named 'cuml'

I guess because the import at the beginning of the test files are not available for CPU tests.

[jperez999]

Moved GPU specific imports into the GPU tests themselves to resolve import failures on cpu container. We also cannot incorporate newer version of rapids until pandas is updated >=0.24.2, <0.25.

[Philmod]

You may have forgotten to remove these?

[jperez999]

Your right I will update that.

[Philmod]

We also cannot incorporate newer version of rapids until pandas is updated >=0.24.2, <0.25.

Is this blocking this PR?

[jperez999]

No rapids version 0.8 works fine. However, the most recent released version is 0.9 with 0.10 coming soon. Those versions of rapids will be blocked by that pandas requirement (>=0.24.2, <0.25).

[Philmod]

Thank you for your contribution!

[Philmod]

Unfortunately, we are moving to tensorflow 2 and it looks like the library is not supported. https://github.com/Kaggle/docker-python/pull/573/files