OneHotEncoder
OneHotEncoder is used to encode categorical features as a one-hot numeric array.
Import
import * as datacook from '@pipcook/datacook';
const { OneHotEncoder } = datacook.Preprocess;
Constructor
const onehotEncoder = new OnehotEncoder({ drop: "first" });
Option in Parameters
| parameter | type | description |
|---|---|---|
drop optional | { ‘none’, ‘binary-only’, ‘first’ } | Specifies a method to drop one of the categories per feature, which is useful to avoid collinear problem. However, dropping one category may introduce a bias term in downstream models. ‘none’: default, return all features ‘first’: drop the first categories in each feature ‘binary-only’: drop the first category in each feature with two categories. default=’none’ |
Properties
categories <Tensor1D>
One dimensional tensor of categories. Onehot result will be consistent with the order of categories appeared in this array.
drop <'first' | 'binary-only' | 'none'>
Drop method for this encoder.
Methods
init()
Initialize one-hot encoder
Syntax
init(x: Tensor1D | number[] | string[]): Promise<void>
| parameter | type | description |
|---|---|---|
| x | Tensor1D | number[] | string[] | data input used to initialize encoder |
Example
const onehotEncoder = new OneHotEncoder();
await onehotEncoder.init([ '1', '2', '3' ]);
encode()
Encode a given feature into one-hot format
Syntax
async encode(x: Tensor | number[] | string[]): Promise<Tensor>
Parameters
| parameter | type | description |
|---|---|---|
| x | Tensor | number[] | string[] | original data need to encode |
Returns
<Tensor> transformed one-hot feature
Example
const onehotEncoder = new OneHotEncoder();
await onehotEncoder.init([ '1', '2', '3' ]);
const encoded = await onehotEncoder.encode(['3', '3', '2']);
/**
* Tensor
* [[0, 0, 1],
* [0, 0, 1],
* [0, 1, 0]]
* /
decode()
Decode one-hot array to original category array
Syntax
async decode(x: Tensor | RecursiveArray<number>): Promise<Tensor>
Parameters
| parameter | type | description |
|---|---|---|
| x | Tensor | RecursiveArray | one-hot format data need to transform |
Returns
<Tensor> transformed category data
Example
const onehotEncoder = new OneHotEncoder();
await onehotEncoder.init([ '1', '2', '3' ]);
const decoded = await onehotEncoder.decode([
[0, 0, 1], [0, 0, 1], [0, 1, 0]
]);
/**
* Tensor
* ['1', '2', '3']
* /