API Reference Documentation

Top-Level Exports

• Core: Matrix, Vector, Tensor
• Models: LinearRegression, LogisticRegression, NaiveBayes, DiscretePerceptron, PolynomialRegression, KNN, WeightedKNN, KMeans, LinearSVM, PolynomialKernelSVM, SigmoidKernelSVM
• Neural: NeuralNetwork, DenseLayer, SoftmaxCrossEntropy
• API: ModelInferenceEngine
• Loss: MSELoss, BCELoss, CrossEntropy, hinge
• Math: relu, sigmoid, softmax, transpose, reshape, standardize, sum, arange, zeros, ones, gather, scatter, slice, minmax, euclidean_distance

Core Memory Structures

Vector

1D optimized numeric array backed by low-level flat typed buffers.

• new Vector(size) - Initialize clear buffer.
• set(index, value) / get(index) - Explicit buffer manipulation.
• sum() / avg() - Analytical scalar reducers.
• toArray() - Returns JavaScript array instance.
• print(label?) - Logs output buffer to console.
• static from(array) - Instantiates vector out of raw JavaScript arrays.
• static zeros(size) - Generates a zeroed initialization container.
• static random(size) - Fills randomized parameters inside structure.

Matrix

2D structured layout utilizing explicit row-major performance mapping layouts.

• new Matrix(rows, columns) - Allocates 2D structure.
• set(r, c, value) / get(r, c) - Coordinates configuration index hooks.
• getRow(r) / setRow(r, Float64Array) - Direct contiguous byte slice updates.
• print(label?) - Visualizes internal grid layout.
• static zeros(rows, columns) - Allocates flat cleared matrix frame.
• static matrixMulVector(matrix, vector) - High performance standard dot engine multiplication.
• static outerProduct(vectorA, vectorB) - Builds a full 2D output layout.
• static add(matrixA, matrixB) - Native operational sum processor.

Tensor

Flexible N-dimensional numeric container.

• new Tensor(data) - Wraps multidimensional raw array arrays.
• clone() - Deep copies execution graph links and variables.
• static zeros(size) - Instantiates bounded zero tensor space.

Models

• LinearRegression - Linear predictive model with forward and backward methods.
• LogisticRegression - Binary classifier with sigmoid activation.
• NaiveBayes - Probabilistic classification model.
• DiscretePerceptron - Simple binary classifier with step activation.
• PolynomialRegression - Polynomial feature regression model.
• KNN - K-nearest neighbors classification.
• WeightedKNN - Weighted KNN variant.
• KMeans - Unsupervised clustering algorithm.
• LinearSVM - Support Vector Machine with linear kernel.
• PolynomialKernelSVM - SVM with polynomial kernel.
• SigmoidKernelSVM - SVM with sigmoid kernel.

Neural Engine Pipeline

Supports flexible dense multi-layer topological allocations and optimization routines.

• new NeuralNetwork(input, hidden, output, loss) - Constructs network parameters.
• forward(input) - Run the network to produce output.
• backward(target) - Compute gradients from loss.
• ActivationEnum - Activation strategies: Linear, ReLU, Sigmoid, Softmax.
• SoftmaxCrossEntropy - Fused softmax and cross-entropy loss.

Loss Functions

• MSELoss - Mean Squared Error for regression.
• BCELoss - Binary Cross Entropy for binary classification.
• CrossEntropy - Multi-class cross-entropy.
• hinge - Loss used in SVM training.

Math Utilities

• relu(vector) / sigmoid(vector) / softmax(vector) - Array execution activations.
• transpose(matrix) / reshape(array, shape) - Layout dimensional shifting utilities.
• standardize(vector) / minmax(vector) - Feature scaling operations.
• sum(vector) / euclidean_distance(a, b) - Reduction tracking algorithms.
• arange(start, end) / zeros(size) / ones(size) - Linear sequence allocation pools.
• gather(array, indices) / scatter(indices, values, shape) / slice(array, start, end) - Graph data mutations.

Inference Engine

ModelInferenceEngine loads serialized model payloads and performs forward inference.

import { ModelInferenceEngine } from 'tensorforgejs';

const engine = new ModelInferenceEngine(modelPayload);
const output = engine.forward(inputVector);