model: scikit: Data type of feature used for predict not respected

[pdxjohnny]

Predicted values are comming out as floats from regression models. We should
typecast to the feature's dtype.

[pdxjohnny]

Specifically if we have a float we should round to an int if the dtype is int

[pdxjohnny]

Check out model/scikit/dffml_model_scikit/scikit_base.py within the predict method. We should cast the type to self.config.predict.dtype before calling Record.predicted

[aadarshsingh191198]

Hi @pdxjohnny, can you please help me with the tests which are failing? (model/scikit, 3.7), (model/scikit, 3.8) and (model/xgboost, 3.8). The error shows

AssertionError: '0' not found in [-1, 0, 1, 2, 3, 4, 5, 6, 7]

Now, why would someone check for a string in an array of ints. Also, why is the error only in these 3 models?

[pdxjohnny]

It should not be checking for a string. That's is the issue here. Don't worry about model/xgboost that is an intermittent failure.

It is currently checking for a string because of the change you made, with means the change was either not entirely correct, or the tests also need updating somehow.

[aadarshsingh191198]

Okay, I will look into it once again keeping in mind the tests for model/scikit ... Thanks for the guidance.

[B-Anupam]

Hey!
I am Interested in Machine Learning projects and looking openly for a contribution.
Please let me know how can I contribute to this issue?

[pdxjohnny]

@B-Anupam This issue was just fixed (thank you @sidhu1012!). Issue #867 seems like a good target for a first contribution. Please read the contributing documentation first: https://intel.github.io/dffml/master/contributing/index.html