Closed ablaom closed 2 years ago
@bkamins Am responding to your comment here.
There has already been some discussion about what features are accepted by DecisionTree.jl: See https://github.com/JuliaAI/MLJDecisionTreeInterface.jl/issues/10 and https://github.com/bensadeghi/DecisionTree.jl/issues/92 . I believe the statement in DecisionTree.jl readme does have a bug - which I just reported here. What I believe is accurate is that DecisionTree accepts any feature for which <
is defined. For the MLJ wrapper, this translates into any Count
, OrderedFactor
or Continuous
feature, but not Multiclass
features, which is what we have in the current PR.
I couldn't find any mention of missing value support in the DecisionTree.jl readme - could you point this out to me? I doubt features can have missing values, as this breaks ordering (the BetaML tree models do support missing however). I imagine missing values in the target are possibly allowed, but I can't find this documented.
I couldn't find any mention of missing value support in the DecisionTree.jl readme - could you point this out to me?
I just manually checked that they were allowed in target - i.e. the model was built without error (but I do not know if then they are dropped or considered as a separate class)
Merging #12 (9a3f54c) into master (ab1e98e) will increase coverage by
0.19%
. The diff coverage is100.00%
.
@@ Coverage Diff @@
## master #12 +/- ##
==========================================
+ Coverage 83.52% 83.72% +0.19%
==========================================
Files 1 1
Lines 85 86 +1
==========================================
+ Hits 71 72 +1
Misses 14 14
Impacted Files | Coverage Δ | |
---|---|---|
src/MLJDecisionTreeInterface.jl | 83.72% <100.00%> (+0.19%) |
:arrow_up: |
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 ab1e98e...9a3f54c. Read the comment docs.
This PR needs:
This PR is simultaneously a genuine draft PR to revise the DecisionTree document strings, and a live proposal of what the [new standard]() for doc-strings should look like. General discussion of the standard should go [here](). However, specific line-by-line comments can also be made in this PR in the usual way.