Add support for `sklearn.mixture.BayesianGaussianMixture`

[mshr-h]

633

[codecov-commenter]

Codecov Report

Merging #677 (f863b9c) into main (36ebab4) will decrease coverage by 0.12%. The diff coverage is 83.75%.

@@            Coverage Diff             @@
##             main     #677      +/-   ##
==========================================
- Coverage   90.25%   90.14%   -0.12%     
==========================================
  Files          78       80       +2     
  Lines        4545     4625      +80     
  Branches      840      848       +8     
==========================================
+ Hits         4102     4169      +67     
- Misses        252      259       +7     
- Partials      191      197       +6     

Flag	Coverage Δ
unittests	90.14% <83.75%> (-0.12%)	:arrow_down:

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
...ml/operator_converters/_mixture_implementations.py	79.68% <79.68%> (ø)
hummingbird/ml/operator_converters/__init__.py	100.00% <100.00%> (ø)
...mingbird/ml/operator_converters/sklearn/mixture.py	100.00% <100.00%> (ø)
hummingbird/ml/supported.py	93.04% <100.00%> (+0.06%)	:arrow_up:

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

[mshr-h]

@ksaur @interesaaat Can you review it? Thanks!

[interesaaat]

LGTM. @mshr-h thanks for adding this! I am curios, how fast is this on GPU compared to sklearn?

[mshr-h]

@interesaaat We haven't benchmarked on the GPU since we are using HB to convert SKL models for microcontrollers through the microTVM. The BGMM->microTVM conversion is not currently available because there's some missing operator support in the PyTorch frontend of the TVM.

But on the CPU (Core i9-10885H), HB is 3x faster than SKL.

[interesaaat]

Did you reach out to the TVM folks regarding the missing ops? I am surprised that on CPU we are already 3x faster than SKL!

[mshr-h]

Did you reach out to the TVM folks regarding the missing ops?

Not yet.

[interesaaat]

Tagging @masahi, maybe he can help.