Address minor comments in #2243

[wschin]

This paper continues addressing some small comments left in #2243.

[shauheen]

As title.

Thanks @wschin , even though this is a small PR, please edit the description and at least mention the other PR here so that they are actually linked.

[wschin]

@shauheen, they are linked. If you go to #2243, there is a footprint. In other words, # followed by a number in commit messages are correctly parsed by github.

[codecov]

Codecov Report

❗ No coverage uploaded for pull request base (master@3eccc93). Click here to learn what that means.
The diff coverage is 100%.

@@            Coverage Diff            @@
##             master    #2369   +/-   ##
=========================================
  Coverage          ?   71.24%           
=========================================
  Files             ?      783           
  Lines             ?   140721           
  Branches          ?    16086           
=========================================
  Hits              ?   100261           
  Misses            ?    36006           
  Partials          ?     4454

Flag	Coverage Δ
#Debug	71.24% <100%> (?)
#production	67.6% <100%> (?)
#test	85.28% <ø> (?)

[sfilipi]

see the documentation about the bitwise complement operator

[sfilipi]

[Ivanidzo4ka]

    public IReadOnlyList<double> TreeWeights { get; }

I think I asked this question in previous PR, but I can't find that any more, so I ask again.
I don't see them been used during prediction, why we expose them?

---

Refers to: src/Microsoft.ML.FastTree/TreeEnsemble.cs:25 in 2eb1ef2. [](commit_id = 2eb1ef2, deletion_comment = False)

[wschin]

    public IReadOnlyList<double> TreeWeights { get; }

I think I asked this question in previous PR, but I can't find that any more, so I ask again.
I don't see them been used during prediction, why we expose them?

Refers to: src/Microsoft.ML.FastTree/TreeEnsemble.cs:25 in 2eb1ef2. [](commit_id = 2eb1ef2, deletion_comment = False)

@TomFinley, do you have any idea about exposing trees' weights to users? They are extracted from InternalRegressionTree and stored in (new) TreeEnsemble.

[TomFinley]

@TomFinley, do you have any idea about exposing trees' weights to users? They are extracted from InternalRegressionTree and stored in (new) TreeEnsemble.

Trees definitely do have weights attached to them. If some code was written that does not account for them, then we ought to address that. Now, functionally I expect this won't have an effect since the weights are under most situations 1, but, that's besides the point.

Here's an example where the weights seem to be taken into account. You can find other usages here and there.

machinelearning/src/Microsoft.ML.FastTree/TreeEnsemble/InternalTreeEnsemble.cs

Line 180 in 40dc35b

sb.Append(_trees[w].Weight);

I think having weights on the trees themselves is useful, and we should continue to have them.

[TomFinley]

Thanks @wschin !