PluginRegression

The PluginRegression provides advanced scientific regression capabilities beyond simple line fitting. It supports multiple regression models, automatic model selection based on statistical criteria (AIC, BIC), robust fitting, and real-time prediction.

Installation

import { createChart } from 'velo-plot/scientific';
import { PluginRegression } from 'velo-plot/plugins/regression';
import { PluginAnnotations } from 'velo-plot/plugins/annotations';

const chart = createChart({ container });

// Enable advanced regression and annotations (for visualization)
await chart.use(PluginRegression({
  defaultMethod: 'polynomial',
  modelSelectionCriteria: 'aic' // 'aic' | 'bic' | 'r2' | 'adjusted-r2'
}));
await chart.use(PluginAnnotations());

Core Features

1. Advanced Fitting

Perform multiple types of regression with specific configurations.

const result = await chart.regression.fit('series-1', data, 'polynomial', { 
  degree: 3 
});

console.log(`R²: ${result.statistics.rSquared}`);
console.log(`Parameters: ${result.parameters.parameters}`);

2. Auto-Model Selection

Automatically find the best-fitting model from a list of candidates.

const bestFit = await chart.regression.autoFit('series-1', data, [
  'linear', 'polynomial', 'exponential', 'gaussian'
]);

console.log(`Best model: ${bestFit.method}`);

3. Real-time Fitting

Enable automatic recalculation of the regression whenever the source data updates.

chart.regression.enableRealtimeFitting('series-1', 'linear');
chart.regression.visualizeFit('series-1');

Supported Methods

Method	Equation	Parameters
'linear'	$y = ax + b$	[slope, intercept]
'polynomial'	$y = \sum a_i x^i$	[a0, a1, ..., an]
'exponential'	$y = a \cdot e^{bx} + c$	[amplitude, rate, offset]
'gaussian'	$y = a \cdot e^{-\frac{(x-\mu)^2}{2\sigma^2}} + d$	[amplitude, mean, std, offset]
'logarithmic'	$y = a \cdot \ln(x) + b$	[a, b]
'power'	$y = a \cdot x^b$	[a, b]

API Reference

fit(seriesId, data, method, config)

Performs the regression analysis.

• seriesId: Target series ID.
• data: Object with x, y and optional weights (Float32Array).
• method: One of the supported methods.
• config: Method-specific configuration (e.g., { degree: 3 }).

autoFit(seriesId, data, candidates)

Compares multiple models and selects the best one based on the plugin configuration (modelSelectionCriteria).

predict(seriesId, xValues, resultIndex)

Generates predicted Y values for a given set of X values using a previously computed result.

visualizeFit(seriesId, resultIndex)

Automatically adds a dashed line series and an equation label to the chart.

exportResults(seriesId, format)

Exports the statistical results to 'json', 'csv', or 'matlab' format.

Mathematical Implementation

The plugin uses Gaussian elimination with partial pivoting for linear and polynomial models. For non-linear models like Gaussian, it uses a gradient descent approach with adaptive learning rates to find the optimal parameters.

$R^2$ and Adjusted $R^2$ are provided for all models, allowing for fair comparison between models with different numbers of parameters.

Known limitations

• Realtime fitting debounces updates; rapid series changes may skip intermediate fits.