Error when importing nimbusml 1.0.0

[pieths]

Create a fresh python 3.7.3 virtual environment (python -m venv [destDir]) and pip install:

• nimbusml 1.0.0 (from local build)
• jupyterlab
• matplotlib

Then start jupyter lab. Enter the following in the first cell of a new notebook,

import numpy as np
import pandas as pd
import nimbusml as nml

Running the cell prints out the following error message.

---------------------------------------------------------------------------
ImportError                               Traceback (most recent call last)
<ipython-input-3-27d67527a163> in <module>
      1 import numpy as np
      2 import pandas as pd
----> 3 import nimbusml as nml

c:\nimbusml\venv\lib\site-packages\nimbusml\__init__.py in <module>
     23 from .internal.utils.data_stream import FileDataStream
     24 from .internal.utils.utils import run_tests
---> 25 from .pipeline import Pipeline
     26 
     27 if sys.platform.lower() == "linux":

c:\nimbusml\venv\lib\site-packages\nimbusml\pipeline.py in <module>
     19 from sklearn.utils.validation import check_X_y, check_array
     20 
---> 21 from .internal.core.base_pipeline_item import BasePipelineItem
     22 from .internal.entrypoints.data_customtextloader import \
     23     data_customtextloader

c:\nimbusml\venv\lib\site-packages\nimbusml\internal\core\base_pipeline_item.py in <module>
     19 
     20 import six
---> 21 from sklearn.utils.fixes import signature
     22 
     23 from ..utils.data_roles import DataRoles, Role

ImportError: cannot import name 'signature' from 'sklearn.utils.fixes' (c:\nimbusml\venv\lib\site-packages\sklearn\utils\fixes.py)

[pieths]

It looks like this might be a problem with scikit-learn 0.21.x. Downgrading to version 0.20.x gets rid of the error.

[pieths]

Note: the same error will also occur when executing the Component.get_params(...) method in internal\utils\entrypoints.py.

The signature method in sklearn.utils.fixes was removed in version 0.21.rc2.
See this issue for more details.

The signature method exists in fixes.py because of the following code.

try:
    from inspect import signature
except ImportError:
    from ..externals.funcsigs import signature

Though the inspect module exists in both Python 2.x and 3.x, the inspect.signature method is new in Python 3.3.

The externals.funcsigs module is a

Back port of Python 3.3's function signature tools from the inspect
module, modified to be compatible with Python 2.6, 2.7 and 3.2+.

funcsigs.py and its related files in sklearn.externals were copied in to the scikit-learn source tree from the funcsigs module.

Here are a few options which would address this issue:

1. Remove support for Python < 3.3. The code could then just use the inspect module from the standard library.

2. Limit the version of the scikit-learn dependency in NimbusML to [0.19.0, 0.20.3].

3. Add the funcsigs package as a dependency of NimbusML.

4. Add the following code to both files that require it. This has the drawback that it relies on a non-public api in scikit-learn.

   try:
       from inspect import signature
   except ImportError:
       from sklearn.externals.funcsigs import signature

5. Import the funcsigs code in to NimbusML (like scikit-learn does).

[ganik]

Thank you pieths for detailed investigation. I suggest to go with option 4 and rely on scikit-learn to deal with signature. Option 1 is no good, we don't want to cut python 2.7 bcs of this minor thing. Same goes with scikit learn version, this would force users to downgrade from latest scikit-learn version. Adding another dependency or copying funcsigs are inferior to option 4.