[methods.partition GridPartitioning]

[FarnazH]

To do:

• [ ] Add test to reach 100% coverage
• [x] Add comments in the code
• [ ] Add formulas to docstring and polish docstring (add reference).

[FanwangM]

The current GridPartition method is missing tests (commented in the corresponding tests). The problem is the usage of compute_diversity in 89f1d5d is outdated where the argument names does not work for hypersphere_overlap_of_subset in compute_diversity function. Once #138 is merged, hope we can solve this problem automatically. But if not, I will provide a quick fix myself.

[FarnazH]

Post PR-#162: Any comments are welcomed:

• [Feature] The same number of bins is used for partitioning each axis/dimension of feature space. It can be desirable to use different values for each axis. This extension should be easy to add.

[FarnazH]

@Ali-Tehrani, when putting together the quick_start.ipynb (see PR #186), I encountered the RuntimeWarning: invalid value encountered in floor_divide bin_index = np.floor_divide(X - axis_minimum, bin_length) from L124 of selector/methods/partition.py. The bin_length ends up being zero. I have copied the code snippet that reproduces this below. I didn't have time to look into it, can you please check what is going on?

from sklearn.datasets import make_blobs
from selector.methods.partition import GridPartition

# generate n_sample data in 2D feature space forming 3 clusters
X, labels = make_blobs(n_samples=500, n_features=2, centers=2, random_state=42)

selector = GridPartition(numb_bins_axis=5, grid_method="equisized_dependent")
selected = selector.select(X, size=50, labels=labels)
print(len(selected))

[FanwangM]

The current code coverage is

  Name                               Stmts   Miss  Cover   Missing
  -----------------------------------------------------------------------------
  selector/methods/partition.py      201      4    98%   375, 407, 520, 619