Fix636

[PGijsbers]

Fixes an issue where a list of lists with simple types could not be serialized (because it was expected to be FeatureUnion, Pipeline or VotingClassifier), as described in #636.

Example flow generated: https://www.openml.org/f/9556
It deserializes correctly with

flow = openml.flows.get_flow(9556)
model = openml.flows.flow_to_sklearn(flow)

[PGijsbers]

The unit test uses OrdinalEncoder which is only available in sklearn >= 0.20, which I forgot about. I will switch in an older scikit-learn component.

[PGijsbers]

I can't find a scikit-learn component in 0.19 or 0.18 which accepts a list of lists. Perhaps I overlooked one, can anyone think of any? If not, would it be fine to only test this against scikit-learn>=0.20?

[mfeurer]

I'm not aware of one either. Unless @amueller or @janvanrijn now of one I think you can simply skip the tests for those sklearn versions.

[janvanrijn]

Neural networks with layers structure param? Op maandag 4 maart 2019 schreef Matthias Feurer <notifications@github.com>:

…

I'm not aware of one either. Unless @amueller <https://github.com/amueller> or @janvanrijn <https://github.com/janvanrijn> now of one I think you can simply skip the tests for those sklearn versions. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#637 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ACL7-ouPMD8UjitntIH1iQHF8Ry-1g7dks5vTVnngaJpZM4bcAq2> .

[PGijsbers]

@janvanrijn thanks for the suggestion, unfortunately it looks like you just specify layer sizes through a tuple of n_layers integers.

[codecov-io]

Codecov Report

Merging #637 into develop will increase coverage by 0.02%.
The diff coverage is 100%.

@@ Coverage Diff @@## develop #637 +/- ##
===========================================
+ Coverage 89.79% 89.81% +0.02% 
===========================================
Files 32 32 Lines 3313 3320 +7 ===========================================
+ Hits 2975 2982 +7 
Misses 338 338

Impacted Files	Coverage Δ
openml/flows/sklearn_converter.py	90.65% <100%> (+0.14%)	⬆️

---

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 0235c51...b98987e. Read the comment docs.