Closed iamDecode closed 3 years ago
Merging #23 (6e698c0) into master (0a42d68) will increase coverage by
0.22%
. The diff coverage is100.00%
.
@@ Coverage Diff @@
## master #23 +/- ##
===========================================
+ Coverage 99.77% 100.00% +0.22%
===========================================
Files 11 12 +1
Lines 440 559 +119
===========================================
+ Hits 439 559 +120
+ Misses 1 0 -1
Impacted Files | Coverage Δ | |
---|---|---|
sklearn_pmml_model/base.py | 100.00% <100.00%> (ø) |
|
sklearn_pmml_model/ensemble/__init__.py | 100.00% <100.00%> (ø) |
|
sklearn_pmml_model/ensemble/forest.py | 100.00% <100.00%> (ø) |
|
sklearn_pmml_model/ensemble/gb.py | 100.00% <100.00%> (ø) |
|
sklearn_pmml_model/linear_model/implementations.py | 100.00% <100.00%> (ø) |
|
sklearn_pmml_model/naive_bayes/implementations.py | 100.00% <100.00%> (ø) |
|
sklearn_pmml_model/tree/__init__.py | 100.00% <100.00%> (ø) |
|
sklearn_pmml_model/tree/tree.py | 100.00% <100.00%> (+1.02%) |
:arrow_up: |
... and 2 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 0a42d68...6e698c0. Read the comment docs.
This pull request introduces 4 alerts when merging 6df5bcbc9f176c0371473aeef13b659fe59c9efa into 0a42d6851c4b7b051b3def8ad9763ec4c66a6b58 - view on LGTM.com
new alerts:
This pull request introduces 1 alert when merging 63d5ca313dc6a26f78d7b9da56fabf03f46d252c into 0a42d6851c4b7b051b3def8ad9763ec4c66a6b58 - view on LGTM.com
new alerts:
This pull request fixes 7 alerts when merging cf9da61f38da9ec5a937742257ac3b6c12b6635d into 0a42d6851c4b7b051b3def8ad9763ec4c66a6b58 - view on LGTM.com
fixed alerts:
This pull request fixes 8 alerts when merging 6e698c0362a9735dce086b25ca46b9c6bf068516 into 0a42d6851c4b7b051b3def8ad9763ec4c66a6b58 - view on LGTM.com
fixed alerts:
The long awaited gradient boosting model support is finally here.
I have found that PMML generation for GBM models is unfortunately not very consistent. I had to account for many different versions which resulted in a rather massive change.
There are some parts I am yet uncertain about:
Target.rescaleFactor
seems related. I've added a FIXME note, and it should be revisited when supporting regression trees outside of gradient boosting.pypmml
. This seems happen at https://github.com/iamDecode/sklearn-pmml-model/blob/27cc103997aa186eb3f42809255eb9658631c1dc/sklearn_pmml_model/tree/tree.py#L224, as switching the - for a + fixes the probabilities for those models. However, that inevitably breaks the predictions of other models, which in my testing is way more significant. I am not sure what causes this discrepancy, but was not able to fix it just yet. However, this is such a slight effect I'm willing to accept it for now. Should reinvestigate later.After this PR, supporting regression should be fairly straightforward.
Closes #20