ColumnTransformer applies transformers to columns of an array or matrix.
Constructor
import { ColumnTransformer } from 'scikitjs';
const transformer = new ColumnTransformer(specs, options);
specs
[string, transformer, columns][]
required
List of (name, transformer, columns) tuples:
- name: string identifier
- transformer: transformer object, ‘drop’, or ‘passthrough’
- columns: array of indices or range
options.remainder
'drop' | 'passthrough'
default:"'drop'"
Strategy for columns not specified in specs:
'drop': columns are dropped'passthrough': columns are passed through unchanged
Methods
fit
Fit all transformers.
fit(X: number[][], y?: number[], sampleWeight?: number[]): ColumnTransformer
transform(X: number[][]): number[][]
fitTransform(X: number[][], y?: number[], sampleWeight?: number[]): number[][]
getParams / setParams
getParams(deep?: boolean): Record<string, unknown>
setParams(params: Record<string, unknown>): ColumnTransformer
Properties
transformers_
readonly [string, transformer, number[]][]
The collection of fitted transformers as tuples
namedTransformers_
Record<string, transformer>
Access transformers by name
Examples
Mixed Feature Types
import {
ColumnTransformer,
StandardScaler,
OneHotEncoder
} from 'scikitjs';
// Dataset with numeric and categorical features
const X = [
[0, 'a', 1.5],
[1, 'b', 2.0],
[2, 'a', 1.0],
[3, 'c', 3.5]
];
// Apply different transformers to different columns
const ct = new ColumnTransformer([
['num', new StandardScaler(), [0, 2]], // Numeric columns
['cat', new OneHotEncoder(), [1]] // Categorical column
]);
const XTransformed = ct.fitTransform(X);
console.log(XTransformed);
Column Ranges
import { ColumnTransformer, StandardScaler, MinMaxScaler } from 'scikitjs';
const X = [
[1, 2, 3, 4, 5],
[2, 4, 6, 8, 10],
[3, 6, 9, 12, 15]
];
// Use range notation for columns
const ct = new ColumnTransformer([
['standard', new StandardScaler(), { start: 0, end: 3 }],
['minmax', new MinMaxScaler(), { start: 3, end: 5 }]
]);
ct.fit(X);
const XTransformed = ct.transform(X);
Drop and Passthrough
import { ColumnTransformer, StandardScaler } from 'scikitjs';
const X = [
[1, 2, 3, 4],
[5, 6, 7, 8],
[9, 10, 11, 12]
];
// Scale some columns, drop others
const ct = new ColumnTransformer([
['scaler', new StandardScaler(), [0, 1]],
['drop', 'drop', [2]] // Drop column 2
]);
const XTransformed = ct.fitTransform(X);
console.log(XTransformed.length); // 3 rows
console.log(XTransformed[0].length); // 3 columns (dropped 1)
Passthrough Remainder
import { ColumnTransformer, StandardScaler } from 'scikitjs';
const X = [
[1, 2, 3, 4],
[5, 6, 7, 8]
];
// Transform first 2 columns, keep others unchanged
const ct = new ColumnTransformer(
[['scaler', new StandardScaler(), [0, 1]]],
{ remainder: 'passthrough' }
);
const XTransformed = ct.fitTransform(X);
console.log(XTransformed[0].length); // 4 (scaled + passthrough)
In a Pipeline
import {
Pipeline,
ColumnTransformer,
StandardScaler,
PCA,
LogisticRegression
} from 'scikitjs';
const X = [
[1, 0, 0.5],
[2, 1, 1.5],
[3, 0, 2.5],
[4, 1, 3.5]
];
const y = [0, 0, 1, 1];
// Preprocess different feature groups
const preprocessor = new ColumnTransformer([
['continuous', new StandardScaler(), [0, 2]],
['binary', 'passthrough', [1]]
]);
const pipeline = new Pipeline([
['preprocessor', preprocessor],
['pca', new PCA({ nComponents: 2 })],
['classifier', new LogisticRegression()]
]);
pipeline.fit(X, y);
const predictions = pipeline.predict(X);
Feature Engineering
import {
ColumnTransformer,
StandardScaler,
PolynomialFeatures
} from 'scikitjs';
const X = [
[1, 2],
[3, 4],
[5, 6]
];
// Apply polynomial features to first column, scale second
const ct = new ColumnTransformer([
['poly', new PolynomialFeatures({ degree: 2 }), [0]],
['scale', new StandardScaler(), [1]]
]);
const XTransformed = ct.fitTransform(X);
console.log(XTransformed);
// Each row: [1, x, x^2, scaled_y]
const ct = new ColumnTransformer([
['scaler', new StandardScaler(), [0, 1]],
['keep', 'passthrough', [2]]
]);
ct.fit(X);
// Access fitted transformers
const scaler = ct.namedTransformers_['scaler'];
console.log(scaler.mean_);
console.log(scaler.scale_);
// Inspect transformers
for (const [name, transformer, columns] of ct.transformers_) {
console.log(`${name}: columns ${columns}`);
}