Recipe

Defined in: ds/src/ml/recipe.js:143

Recipe class for building inspectable preprocessing workflows

A Recipe defines a sequence of data preprocessing steps that can be:

Defined declaratively through chainable methods
Executed with prep() to fit transformers on training data
Applied to new data with bake() using fitted transformers
Inspected at every step to understand transformations

Supported preprocessing operations:

Data Cleaning:

parseNumeric(): Convert string columns to numbers
clean(): Remove rows with invalid categorical values

Missing Value Imputation:

imputeMean(): Impute with mean values
imputeMedian(): Impute with median values
imputeMode(): Impute with mode (most frequent)
imputeKNN(): Impute using K-Nearest Neighbors
imputeIterative(): Impute using iterative MICE algorithm

Outlier Handling:

removeOutliers(): Remove outliers using isolation forest, LOF, or Mahalanobis distance
clipOutliers(): Clip outliers using IQR method

Encoding:

oneHot(): One-hot encode categorical columns

Scaling:

scale(): Scale numeric columns (standard or minmax)

Feature Engineering:

createInteractions(): Create pairwise interaction features
createPolynomial(): Create polynomial features
binContinuous(): Bin continuous variables into discrete categories

Dimensionality Reduction:

pca(): Principal Component Analysis
lda(): Linear Discriminant Analysis (supervised)
rda(): Redundancy Analysis (constrained ordination)

Sampling:

upsample(): Oversample minority class for imbalanced data
downsample(): Undersample majority class for imbalanced data

Feature Selection:

selectByVariance(): Remove low-variance features
selectByCorrelation(): Remove highly correlated features

Data Splitting:

split(): Split into train/test sets

Example

// Complete workflow
const recipe = new Recipe({ data: myData, X: features, y: 'target' })
  .parseNumeric(['age', 'price'])
  .clean({ category: ['A', 'B', 'C'] })
  .oneHot(['category'])
  .scale(['age', 'price'], { method: 'standard' })
  .split({ ratio: 0.7, seed: 42 });

const result = recipe.prep();
// result.train.data - training data
// result.test.data - test data
// result.transformers - fitted transformers (scale, oneHot, etc.)
// result.steps - intermediate outputs for inspection

Constructors

Constructor

new Recipe(__namedParameters): Recipe

Defined in: ds/src/ml/recipe.js:144

Parameters

__namedParameters

data

any

X

any

y

any

Returns

Recipe

Properties

_prepared

_prepared: boolean

Defined in: ds/src/ml/recipe.js:149

_splitResult

_splitResult: any

Defined in: ds/src/ml/recipe.js:1579

_stepOutputs

_stepOutputs: ({ name: any; output: any; transformer: any; } | { name: any; output: any; transformer?: undefined; })[]

Defined in: ds/src/ml/recipe.js:1578

_transformers

_transformers: object

Defined in: ds/src/ml/recipe.js:150

initialData

initialData: any

Defined in: ds/src/ml/recipe.js:145

splitConfig

splitConfig: object

Defined in: ds/src/ml/recipe.js:1496

ratio

ratio: number

seed

seed: number

shuffle

shuffle: boolean

steps

steps: any[]

Defined in: ds/src/ml/recipe.js:148

X

X: any[]

Defined in: ds/src/ml/recipe.js:146

y

y: any

Defined in: ds/src/ml/recipe.js:147

Methods

bake()

bake(data): any

Defined in: ds/src/ml/recipe.js:1613

Apply fitted transformers to new data

Parameters

data

any[]

New data to transform

Returns

any

Transformed data

binContinuous()

binContinuous(column, options?): Recipe

Defined in: ds/src/ml/recipe.js:1047

Bin continuous variables into discrete categories

Parameters

column

string

Column to bin

options?

Binning options

bins

number = 5

Number of bins (default 5)

labels

string[] = null

Custom bin labels

Returns

Recipe

this

Example

recipe.binContinuous('age', { bins: 5, labels: ['child', 'teen', 'adult', 'middle', 'senior'] });

clean()

clean(validCategories): Recipe

Defined in: ds/src/ml/recipe.js:195

Clean categorical columns

Parameters

validCategories

any

Map of column -> valid values

Returns

Recipe

this

clipOutliers()

clipOutliers(columns, options?): Recipe

Defined in: ds/src/ml/recipe.js:844

Clip outliers using IQR method

Parameters

columns

string[]

Columns to clip

options?

Clipping options

multiplier

number = 1.5

IQR multiplier (default 1.5)

Returns

Recipe

this

Example

recipe.clipOutliers(['price', 'age'], { multiplier: 1.5 });

createInteractions()

createInteractions(columns): Recipe

Defined in: ds/src/ml/recipe.js:933

Create pairwise interaction features

Parameters

columns

string[]

Columns to create interactions from

Returns

Recipe

this

Example

recipe.createInteractions(['feature1', 'feature2', 'feature3']);
// Creates: feature1_x_feature2, feature1_x_feature3, feature2_x_feature3

createPolynomial()

createPolynomial(columns, options?): Recipe

Defined in: ds/src/ml/recipe.js:993

Create polynomial features

Parameters

columns

string[]

Columns to create polynomials from

options?

Polynomial options

degree

number = 2

Polynomial degree (default 2)

Returns

Recipe

this

Example

recipe.createPolynomial(['age', 'income'], { degree: 2 });
// Creates: age^2, income^2

downsample()

downsample(options?): Recipe

Defined in: ds/src/ml/recipe.js:1223

Downsample majority class for imbalanced classification

Parameters

options?

Downsampling options

seed

number = null

Random seed

strategy

string = 'balance'

‘balance’ (equal classes) or ‘ratio’ (custom ratio)

targetRatio

number = 1.0

Target ratio of majority to minority (for ‘ratio’ strategy)

Returns

Recipe

this

Example

recipe.downsample({ strategy: 'balance', seed: 42 });

imputeIterative()

imputeIterative(columns, options?): Recipe

Defined in: ds/src/ml/recipe.js:764

Impute missing values using iterative imputation (MICE)

Parameters

columns

string[]

Columns to impute

options?

Iterative imputer options

maxIter

number = 10

Maximum iterations (default 10)

tol

number = 0.001

Convergence tolerance (default 0.001)

Returns

Recipe

this

Example

recipe.imputeIterative(['age', 'income'], { maxIter: 20 });

imputeKNN()

imputeKNN(columns, options?): Recipe

Defined in: ds/src/ml/recipe.js:725

Impute missing values using KNN

Parameters

columns

string[]

Columns to impute

options?

KNN imputer options

k

number = 5

Number of neighbors (default 5)

Returns

Recipe

this

Example

recipe.imputeKNN(['age', 'income'], { k: 3 });

imputeMean()

imputeMean(columns): Recipe

Defined in: ds/src/ml/recipe.js:618

Impute missing values with mean

Parameters

columns

string[]

Columns to impute

Returns

Recipe

this

Example

recipe.imputeMean(['age', 'income']);

imputeMedian()

imputeMedian(columns): Recipe

Defined in: ds/src/ml/recipe.js:653

Impute missing values with median

Parameters

columns

string[]

Columns to impute

Returns

Recipe

this

Example

recipe.imputeMedian(['age', 'price']);

imputeMode()

imputeMode(columns): Recipe

Defined in: ds/src/ml/recipe.js:688

Impute missing values with mode (most frequent value)

Parameters

columns

string[]

Columns to impute

Returns

Recipe

this

Example

recipe.imputeMode(['category', 'status']);

lda()

lda(options?): Recipe

Defined in: ds/src/ml/recipe.js:439

Apply Linear Discriminant Analysis for supervised dimensionality reduction

Reduces dimensionality while maximizing class separation. Requires a target variable. This is both feature extraction and supervised learning.

Parameters

options?

LDA options

columns

string[]

Columns to include in LDA

nComponents?

number = null

Number of discriminants to keep

scale?

boolean = false

Scale features before LDA

Returns

Recipe

this (for chaining)

Example

recipe.lda({
  columns: ['feature1', 'feature2', 'feature3'],
  nComponents: 2
});

oneHot()

oneHot(columns, options?): Recipe

Defined in: ds/src/ml/recipe.js:223

One-hot encode categorical columns

Parameters

columns

string[]

Columns to encode

options?

Encoding options

dropFirst

boolean = true

Drop first category

prefix

boolean = true

Use column name prefix

Returns

Recipe

this

parseNumeric()

parseNumeric(columns): Recipe

Defined in: ds/src/ml/recipe.js:165

Parse string columns as numeric

Converts string representations of numbers to actual numeric values. Useful when CSV parsers incorrectly infer column types.

Parameters

columns

string[]

Column names to parse

Returns

Recipe

this (for chaining)

Example

recipe.parseNumeric(['age', 'price', 'quantity']);

pca()

pca(options?): Recipe

Defined in: ds/src/ml/recipe.js:342

Apply Principal Component Analysis for dimensionality reduction

Reduces the dimensionality of numeric features by projecting them onto principal components. This is a feature extraction/transformation step.

Parameters

options?

PCA options

center?

boolean = true

Center features before PCA

columns

string[]

Columns to include in PCA

nComponents?

number = null

Number of components to keep (default: all)

scale?

boolean = true

Scale features before PCA

Returns

Recipe

this (for chaining)

Example

recipe.pca({
  columns: ['feature1', 'feature2', 'feature3'],
  nComponents: 2,
  scale: true
});

prep()

prep(): any

Defined in: ds/src/ml/recipe.js:1505

Execute the recipe on the initial data Returns train/test data and all fitted transformers

Returns

any

Prepared data with train, test, transformers

rda()

rda(options?): Recipe

Defined in: ds/src/ml/recipe.js:539

Apply Redundancy Analysis for constrained ordination

RDA combines regression and PCA to find patterns in response variables that are explained by predictor variables. Useful for ecological data.

Parameters

options?

RDA options

nComponents?

number = null

Number of RDA axes to keep

predictors

string[]

Predictor variable columns

response

string[]

Response variable columns

scale?

boolean = false

Scale variables before RDA

Returns

Recipe

this (for chaining)

Example

recipe.rda({
  response: ['species1', 'species2', 'species3'],
  predictors: ['temperature', 'rainfall'],
  nComponents: 2
});

removeOutliers()

removeOutliers(columns, options?): Recipe

Defined in: ds/src/ml/recipe.js:804

Remove outliers from the dataset

Parameters

columns

string[]

Columns to check for outliers

options?

Outlier detection options

contamination

number = 0.1

Expected proportion of outliers (default 0.1)

method

string = 'isolation_forest'

Detection method: ‘isolation_forest’, ‘lof’, or ‘mahalanobis’

Returns

Recipe

this

Example

recipe.removeOutliers(['price', 'quantity'], { method: 'isolation_forest', contamination: 0.05 });

scale()

scale(columns, options?): Recipe

Defined in: ds/src/ml/recipe.js:293

Scale numeric columns

Parameters

columns

string[]

Columns to scale

options?

Scaling options

method

string = 'standard'

‘standard’ or ‘minmax’

Returns

Recipe

this

selectByCorrelation()

selectByCorrelation(options?): Recipe

Defined in: ds/src/ml/recipe.js:1384

Remove highly correlated features

Parameters

options?

Feature selection options

threshold

number = 0.95

Correlation threshold (default 0.95)

Returns

Recipe

this

Example

recipe.selectByCorrelation({ threshold: 0.9 });

selectByVariance()

selectByVariance(options?): Recipe

Defined in: ds/src/ml/recipe.js:1299

Remove low-variance features

Parameters

options?

Feature selection options

threshold

number = 0.0

Variance threshold (default 0.0)

Returns

Recipe

this

Example

recipe.selectByVariance({ threshold: 0.01 });

split()

split(options?): Recipe

Defined in: ds/src/ml/recipe.js:1495

Split data into train/test sets

Parameters

options?

Split options

ratio

number = 0.7

Training ratio (default 0.7)

seed

number = null

Random seed

shuffle

boolean = true

Shuffle before split

Returns

Recipe

this

summary()

summary(): string

Defined in: ds/src/ml/recipe.js:1661

Get a summary of the recipe

Returns

string

Recipe summary

upsample()

upsample(options?): Recipe

Defined in: ds/src/ml/recipe.js:1160

Upsample minority class for imbalanced classification

Parameters

options?

Upsampling options

seed

number = null

Random seed

targetRatio

number = 1.0

Target ratio of minority to majority (default 1.0 for balanced)

Returns

Recipe

this

Example

recipe.upsample({ targetRatio: 1.0, seed: 42 });