[Review] Precision recall curve using cupy

[daxiongshu]

In this PR, I will refactor _ranking.py so that basic functions can be reused among existing and upcoming metrics.

[GPUtester]

Please update the changelog in order to start CI tests.

View the gpuCI docs here.

[dantegd]

Just a few comments on a first look

[dantegd]

why float32?

[daxiongshu]

I just wanted to convert it to float since later on it is used in the RawKernel which expects float data type. float64 should also be fine.

[dantegd]

I think float then is probably fine. Technically we could use

cuml/python/cuml/common/kernel_utils.py

Line 55 in 489a7d8

def cuda_kernel_factory(nvrtc_kernel_str, dtypes, kernel_name=None):

but probably not worth it here, float shouldn't pose any problems, correct?

[daxiongshu]

I just tried several variations and it is really surprising:

# trial 1
ones = y_true[ids]

# trial 2
ones = y_true[ids].astype('float')

#trial 3
ones = y_true[ids].astype('float64')

All of them have assertion error due to mismatched results against sklearn.
Only ones = y_true[ids].astype('float32') works. Please note that in each trial I not only changed this line but every line with astype('float32').

[daxiongshu]

@dantegd Please let me know if this is acceptable. And if there are other changes to be made. Thank you!

[dantegd]

rerun tests

[dantegd]

@daxiongshu could you solve the conflicts? The PR is very close, I think it looks good so far!

[daxiongshu]

@dantegd merge conflicts fixed. Thank you!