pharo-ai / decision-tree-model Goto Github PK

I'll check to fix it

I believe that DecisionTree can be used not only for machine learning, but also for many other applications (for example, to contain expert knowledge). So it would be nice to have it as a standalone project that lives in a separate repository.

Then there is DecisionTreeModel - a machine learning model that is used for building decision trees. This should be a separate repository can contains an abstract class DecisionTreeModel and several algorithms implemented as subclasses:

• C4.5
• ID3
• etc.

DecisionTreeModel repository should depend on DecisionTree repository.

Finally, I don't remember well how RandomForest works, but if it is the ensembling algorithm that averages the output of several DecisionTreeModels, then you can put it into a separate repository and add a dependency on DecisionTreeModel.

However, if it is the kind of DecisionTreeModel that generates random decision trees and averages their outputs, then I would make it a subclass of the abstract DecisionTreeModel and keep it in the same repository.

This way we can have a clean separation and each module can be used independently.

We have replaced them with trueChild and falseChild, but those methods are still in the repository. They should be removed

This would allow us to ask that object to explain how the decision was made.

Example:

tennisTree := DtmDecisionTree withCondition: [ weather | weather isRaining ].

tennisTree trueChild: (DtmDecisionTreeLeaf withLabel: 'stay home').
tennisTree falseChild: (DtmDecisionTree withCondition: [ :weather |
    (weather temperature < 0) or: (weather temperature > 30) ].

tennisTree falseChild trueChild: (DtmDecisionTreeLeaf withLabel: 'stay home').
tennisThee falseChild falseChild: (DtmDecisionTreeLeaf withLabel: 'play tennis').

weather := Weather new
    raining: false;
    temperature: 42;
    yourself.

decision := tennisTree classify: weather.
decision label. "stay home"

But now we can also explain the decision:

decision why.
"
{
    [ :weather | (weather temperature < 0) or: (weather temperature > 30) ] -> true .
    [ weather | weather isRaining ] -> false
}
"

If you look at the baseline, it depends of dataFrame and Datasets. We should delete those dependencies