Fix divide-by-zero, deprecation, and futurewarnings #236
Conversation
`np.float` and `np.int` have been deprecated. Replaced with the built in `float` and `int` respectively.
`sklearn` added random state as input to `AffinityPropogation` in 0.23. To preserve the old behavior `randomstate = 0` was added as this is the previous hard coded value according to the docs here `https://scikit-learn.org/stable/modules/generated/sklearn.cluster.AffinityPropagation.html`
Division by 0 in the clustering and visualization code seems to happen in the normal course of operation and should result in `inf` thus silencing the warnings at those points.
Codecov Report
@@ Coverage Diff @@
## master #236 +/- ##
==========================================
+ Coverage 80.27% 80.32% +0.04%
==========================================
Files 11 11
Lines 862 864 +2
Branches 189 189
==========================================
+ Hits 692 694 +2
Misses 138 138
Partials 32 32
Continue to review full report at Codecov.
|
|
Cool, thanks. I wonder why plotlyviz had not-unix line endings... Can you make any future PRs off of not-master branches? It's okay for this one. |
Sure, is there a particular branch you would like me to use? |
|
Oh, you know what, I totally misread it. I thought this was coming from your fork's master branch. Long day, sorry. PRs against this repo's master is preferred, as you've been doing. |
Currently the test suite generates a handful of warnings. These include:
np.floatandnp.intAffinityPropogationapi changesThis pull request aims to fix these warnings. It changes all occurences of
np.intandnp.floattointandfloatrespectively, addsrandomstate=0toAffinityPropogationto keep previous behavior, and silences division by 0 errors as the behaviorx / 0 = infseems to be the expect behavior where they arose. It does not fix the invalid division error as I don't understand the function which raises that warning well enough to tell ifNaNis expected behavior or if the data has to be cleansed somehow.