[Numpy] Add sampling method for bernoulli

[xidulu]

Description

Native numpy does not support sampling directly from Bernoulli distribution, in order to do so, users have to use np.random.binomial and set n = 1 manually.
However, I think a separate implementation for Bernoulli is worth consideration for the following reasons:

1. Sampling from a bernoulli, unlike binomial, is very easy to implement.
2. random.bernoulli could be temporarily used as a ''work around'' for binomial sampling, see https://github.com/pytorch/pytorch/blob/071971476d7431a24e527bdc181981678055a95d/torch/distributions/binomial.py#L102 for more details. ( also, some discussions: torch.distributions.Binomial.sample() uses a massive amount of memory pytorch/pytorch#20343 )

This pull request also fixes an existing bug in np.random.uniform, where scalar tensor would cause infer shape failure.

Checklist

Essentials

Please feel free to remove inapplicable items for your PR.

• The PR title starts with [MXNET-$JIRA_ID], where $JIRA_ID refers to the relevant JIRA issue created (except PRs with tiny changes)
• Changes are complete (i.e. I finished coding on this PR)
• All changes have test coverage:
• Unit tests are added for small changes to verify correctness (e.g. adding a new operator)
• Nightly tests are added for complicated/long-running ones (e.g. changing distributed kvstore)
• Build tests will be added for build configuration changes (e.g. adding a new build option with NCCL)
• Code is well-documented:
• For user-facing API changes, API doc string has been updated.
• For new C++ functions in header files, their functionalities and arguments are documented.
• For new examples, README.md is added to explain the what the example does, the source of the dataset, expected performance on test set and reference to the original paper if applicable
• Check the API doc at https://mxnet-ci-doc.s3-accelerate.dualstack.amazonaws.com/PR-$PR_ID/$BUILD_ID/index.html
• To the my best knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change

Changes

• Feature1, tests, (and when applicable, API doc)
• Feature2, tests, (and when applicable, API doc)

Comments

• If this change is a backward incompatible change, why must this change be made.
• Interesting edge cases to note here

[xidulu]

@reminisce
For some reasons, operators under npx.random cannot have access to npi ops, could you help me resolve this issue ?

[xidulu]

Correctness for distribution has been briefly verified by hand.
With the following code:

In [9]: (npx.random.bernoulli(prob=prob, size=(1000000, 10, 10)).mean(0) - prob).mean()
Out[9]: array(-3.0148094e-05, ctx=gpu(0))

In [10]: (npx.random.bernoulli(prob=prob, size=(1000000, 10, 10)).var(0) - prob * (1 - prob)).mean()
Out[10]: array(-9.472111e-06, ctx=gpu(0))

In [11]: logit = np.log(prob) - np.log(1 - prob)

In [12]: (npx.random.bernoulli(logit=logit, size=(1000000, 10, 10)).mean(0) - prob).mean()
Out[12]: array(-3.4588957e-05, ctx=gpu(0))

In [13]: (npx.random.bernoulli(logit=logit, size=(1000000, 10, 10)).var(0) - prob * (1 - prob)).mean()
Out[13]: array(1.1445168e-05, ctx=gpu(0))

[codecov-io]

Codecov Report

Merging #16638 into master will decrease coverage by 0.05%.
The diff coverage is 32.75%.

@@            Coverage Diff             @@
##           master   #16638      +/-   ##
==========================================
- Coverage      67%   66.94%   -0.06%     
==========================================
  Files         264      271       +7     
  Lines       29507    29980     +473     
  Branches     4357     4440      +83     
==========================================
+ Hits        19770    20071     +301     
- Misses       8490     8637     +147     
- Partials     1247     1272      +25

Impacted Files	Coverage Δ
python/mxnet/base.py	69.77% <100%> (-0.16%)	⬇️
python/mxnet/numpy_extension/__init__.py	100% <100%> (ø)	⬆️
python/mxnet/symbol/numpy_extension/__init__.py	100% <100%> (ø)	⬆️
python/mxnet/ndarray/numpy_extension/__init__.py	100% <100%> (ø)	⬆️
python/mxnet/symbol/numpy_extension/random.py	24% <24%> (ø)
python/mxnet/ndarray/numpy_extension/random.py	24% <24%> (ø)
python/mxnet/numpy_extension/random.py	88.88% <75%> (-11.12%)	⬇️
python/mxnet/gluon/contrib/cnn/conv_layers.py	41.6% <0%> (-25.49%)	⬇️
python/mxnet/contrib/quantization.py	61.39% <0%> (-0.49%)	⬇️
...hon/mxnet/gluon/contrib/estimator/event_handler.py	73.56% <0%> (-0.35%)	⬇️
... and 34 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 29e467b...70ae946. Read the comment docs.

[xidulu]

@sxjscience
Could you take a look at my implementation? Thx :)

[haojin2]

LGTM except for 1 minor style problem. @sxjscience Any more comments?