Way to prune tree after training

[djsegal]

Didn't know where to ask this. What is the best way to prune a CART after it's been trained?

[haggaishachar]

Is there a plan to support categorical features as well? the original CART algorithm supports it.

[aiday-mar]

For the textual features, it is possible to use the spectra-processing library. There are the methods matrixNumericalEncoding, matrixNumericalDecoding, matrixApplyNumericalEncoding. It does label encoding. An example is as follows:

`
let xTrain = [
[73, 'a', 'o', 152],
[93, 'b', 'u', 185],
[89, 'c', 'p', 180],
[96, 'd', 'v', 196],
[73, 'e', 'j', 142],
[53, 'f', 'w', 101],
[69, 'g', 'x', 149],
[47, 'h', 'y', 115],
[87, 'i', 'p', 175],
[79, 'j', 'b', 164],
[69, 'j', 'm', 141],
[70, 'k', 'g', 141],
[93, 'l', 'c', 184],
[79, 'a', 'm', 152],
[70, 'm', 'z', 148],
[93, 'n', 'aa', 192],
];

let xTest = [
[78, 'o', 'ab', 147],
[81, 'p', 'u', 183],
[88, 'q', 's', 177],
[78, 'r', 'x', 159],
[82, 's', 'p', 177],
[86, 't', 'n', 175],
[78, 'r', 'ac', 175],
[76, 'r', 'ad', 149],
[96, 'u', 'l', 192],
];

let yTrain = [....]; // some data

const { dictionnary } = matrixNumericalEncoding(xTrain);
const reg = new DecisionTreeRegression();
reg.train(xTrain, yTrain);
const matrix = matrixApplyNumericalEncoding(xTest,dictionnary);
const estimations = reg.predict(xTest);
`