GradientBoostingClassifier
Gradient boosting is a machine learning method which provides predictions by training an ensemble of weak estimatorss. GradientBoostingClassifier is an implementation of gradient boosting for classification task.
Import
import * as DataCook from '@pipcook/datacook';
const { GradientBoostingClassifier } = DataCook.Model;
Constructor
const gb = newGradientBoostingClassifier({ nEstimators: 10 });
Option parameters
| Parameter | Type | Description |
|---|---|---|
| nEstimators | number | number of estimators for fitting. default = 100 |
| criterion | ||
| minSamplesLeaf | number | The minimum number of samples required to be at leaf node, default = 1 |
| minImpurityDecrease | number | A node will be split if this split induces a decrease of the impurity greater than or equal to this value |
| minWeightFractionLeaf | ||
| minSamplesSplit | number | Minimum number of samples required to split an internal node, default = 2va |
| validationFraction | number | The proportion of training data to set aside as validation set for early stopping. Must be between 0 and 1. Only used if nIterNoChange is set to an integer. |
| ccpAlpha | number | Complexity parameter used for Minimal Cost-complexity Pruning. The subtree with the largest cost complexity that is smaller than ccpAlpha will be chosen. By default, no pruning is performed. |
| maxDepth | number | Maximum depth of the individual regression tree, default = 3 |
| maxFeatures | number, or {“auto”, “sqrt”, “log2”} | The number of features to consider when looking for the best split: - If integer value, then consider maxFeatures features at each split.- If not interger value, then maxFeatures is a fraction and Math.floor(maxFeatures * nFeatures) features are considered at each split.- If “auto”, then max_features=sqrt(n_features).- If “sqrt”, then maxFeatures=sqrt(nFeatures).- If “log2”, then maxFeatures=log2(nFeatures).- If none, then maxFeatures=nFatures. |
| tol | number | Tolerance for the early stopping. When the loss is not improving by at least tol for nIterNoChange iterations (if set to a number), the training stops, default = 1e-4 |
| nIterNoChange | number | Used to decide if early stopping will be used to terminate training when validation score is not improving. By default it is set to none to disable early stopping. If set to a number, it will set aside validation_fraction size of the training data as validation and terminate training when validation score is not improving in all of the previous nIterNoChange numbers of iterations. The split is stratified. |
Methods
fit
Fit gradient boosting classifier
Syntax
async fit(xData: number[][] | Tensor2D, yData: number[] | string[] | boolean[] | Tensor1D): Promise<void>
Parameters
| Parameter | type | description |
|---|---|---|
| xData | Tensor2D| number[][] | input data of shape (nSamples,nFeatures) in type of array or tensor |
| yData | Tensor1D| number[] | string[] | boolean[] | input target |
predict
Make predictions using gradient boosting model.
async predict(xData: Tensor|RecursiveArray<number>): Promise<Tensor>
Parameters
| parameter | type | description |
|---|---|---|
| xData | Tensor | RecursiveArray <number> |
Returns
Promise of fitted values
fromJson
Load model paramters from json string object
async fromJson(modelJson: string)
Parameters
| parameter | type | description |
|---|---|---|
| modelJson | string | model json string |
toJson
Export model paramters to json string
async toJson(): Promise<string>
Returns
String output of model json