search

mlpack / benchmarks

Machine Learning Benchmark Scripts
101 stars 49 forks source link

Adding metrics to random forest implementation. #67

Closed Iron-Stark closed 7 years ago

Iron-Stark commented 7 years ago

@zoq @rcurtin According to http://www.shogun-toolbox.org/examples/latest/examples/multiclass_classifier/random_forest.html I need to pass 3 arguments to RandomForest (data labels and number of trees) but it shows an error here. Cant figure out why.

Iron-Stark commented 7 years ago

@zoq @rcurtin

Cannot understand the error this time. Error says "No test reports found." How to rectify it?

zoq commented 7 years ago

Probably because the random forest script throws an exception.

make checks /usr/bin/python3 tests/tests.py terminate called after throwing an instance of 'shogun::ShogunException'

Also I'm not sure I get the idea to replace

randomForest.set_combination_rule(mVote)
    randomForest.set_labels(labels)
    randomForest.train(data)

with randomForest = RandomForest(RealFeatures(data.T), MulticlassLabels(labels), self.numTrees) can you elaborate on that.

And, it looks like that you do RealFeatures twice inside the benchmark script and inside the BuildModel function which probably results in some strange issues.

Also, in the context of this PR we should only incorporate changes for the random forest method. To keep the PR clean let's remove the nbc commits.

Iron-Stark commented 7 years ago

@zoq

I replaced the code because the latest shogun docs implements it in this way. If the previous implementation were also fine I will revert back the changes I made. Please let me know :)

zoq commented 7 years ago

I think we can use the previous implementation, unless you see a problem, if that's the case, let me know and I'll look into the issue.