[TEST] Add statistical inference for Truncated Gumbel distribution

[AetherPrior]

Description

Attempts to fix #1299 using statistical inference

Checklist

Essentials

• [x] PR's title starts with a category (e.g. [BUGFIX], [MODEL], [TUTORIAL], [FEATURE], [DOC], etc)
• [x] Changes are complete (i.e. I finished coding on this PR)
• [x] All changes have test coverage
• [x] Code is well-documented

Changes

• [x] statistical inference test for truncated_gumbel

Comments

Brief description of the test

• The test involves checking for truncation and also checking if the distribution was initially sampled from a gumbel distribution
• Using the ks-test, it is possible to compare 2 different samples and check if they arise from the same distribution using p-values (statistical inference).
  • if p_value > 0.05 for over 90% of the time the test is performed, we can accept the null hypothesis (i.e. the sample we have generated is indeed from gumbel).

Method

1. We first attempt to verify that all the samples have values under the threshold
2. Then, we undo the truncation, with the help of the implementation of trunc_gumbel in src/gluonnlp/op.py and then sample another gumbel distribution

Note

• Since this attempts to remove truncation from the gumbel distribution, I'm not sure if the approach is "correct". Requesting comments/approval on the same.

cc @dmlc/gluon-nlp-team

[github-actions[bot]]

The documentation website for preview: http://gluon-nlp-staging.s3-accelerate.dualstack.amazonaws.com/PR1578/bd0cb75943156f5a5369e4e7ab7896e87962dbc0/index.html

[AetherPrior]

Tagging @szha and @zheyuye

[github-actions[bot]]

The documentation website for preview: http://gluon-nlp-staging.s3-accelerate.dualstack.amazonaws.com/PR1578/5bd7e3c8f69b6ea0fb81631fd0d2d44c0fc3172f/index.html

[AetherPrior]

Thank you for your approval of my PR. The build checks have passed. I request for the pull request to be merged and #1299 to be closed.