f1 metric - micro/macro/weighted

[SSaishruthi]

This PR contains a script for adding f1-macro metric.

Colab notebook: https://colab.research.google.com/drive/1LJ1yb8cUgisNP3ETfMLRYPQEyJivzDv-

[SSaishruthi]

Please let me know what will be my next steps. To do as far as I know will be to add a test for the same.

[Squadrick]

@SSaishruthi Next steps: __init__.py, README.md and tests.

[SSaishruthi]

Thanks @Squadrick
Will update soon

[SSaishruthi]

@Squadrick
Updated with F1 micro, macro and weighted
Test colab notebook: https://colab.research.google.com/drive/1qSq0SsYkPqjdKUgM1RM4kKM67X75ocFj

[Squadrick]

@SSaishruthi The code looks good so far, just a few small changes. Also looking you need to add the test cases, update __init__.py, README.md and BUILD.

[SSaishruthi]

@Squadrick
Updated scripts. Please let me know if there is any change to be made.
Colab: https://colab.research.google.com/drive/1qSq0SsYkPqjdKUgM1RM4kKM67X75ocFj

[Squadrick]

Update the module to be called F1Score to accept micro as well. Like @WindQAQ mentioned here, it's a much cleaner API and reduces the amount of net code.

[SSaishruthi]

Hi @Squadrick @WindQAQ

I have updated code to calculate F1 scores under one class.
Now, we have four options:

1. Micro
2. Macro
3. Weighted
4. None (This will produce F1 scores for each class)

Colab: https://colab.research.google.com/drive/1qSq0SsYkPqjdKUgM1RM4kKM67X75ocFj#scrollTo=hZSxoW4eYqYW

I ran code format to check scripts. I encountered an installation error in clang and removed that step because I noticed that is only for C++.

[Squadrick]

@SSaishruthi Can you pull from master and update this PR?

[SSaishruthi]

@Squadrick

I have updated the branch and resolved conflicts.
Can we trigger the test again?

[Squadrick]

Left a few comments about refactoring the test cases, after which we should be good to merge.

[Squadrick]

Nit: Rename f1_obj -> f1_micro, f1_obj1 -> f1_macro, f1_obj2 -> f1_weighted.

[Squadrick]

Do the renaming through out the file.

[Squadrick]

We can merge once as soon as this is done.

[SSaishruthi]

@Squadrick

Inputs to None condition should be multiclass but not multilabel. That's why I created a separate function for None as they require a different set of inputs. Can we have it like this or if you can suggest me an alternative that would be great.

One solution I thought I can try is to have two sets of input, pass both (actual1, actual2, pred1, pred2) and use only multiclass(actual1, pred1) for none. Will that work?

Thanks

[Squadrick]

@SSaishruthi Oh, my bad. In that case, leave it as is.

[SSaishruthi]

@Squadrick updated the scripts. Can you please trigger the test again?

[Squadrick]

@SSaishruthi LGTM! Thanks for the contribution.