Analysis

In addition to rendering, @torchjsorg/viz provides a suite of GPU-accelerated analysis tools to help you understand your data before it hits the screen.

Visualization of high-dimensional data being compressed into 2D

Dimensionality Reduction

Visualizing high-dimensional embeddings requires compressing them into 2D or 3D space. Viz performs these operations directly on the GPU.

Algorithm	Best For	Complexity
PCA	Fast, linear projections	Low
t-SNE	Revealing local clusters	High
UMAP	Preserving global structure	Medium

import { createPCA } from '@torchjsorg/viz';

const pca = createPCA();
const lowerDim = await pca.fit_transform(highDimTensor, { n_components: 3 });

Statistical Analysis

Quickly compute distributions and metrics across large tensors without stalling the main thread.

import { computeStats, computeHistogram } from '@torchjsorg/viz';

// Fast GPU-based statistics
const stats = await computeStats(myTensor);
console.log(stats.mean, stats.std, stats.median);

// Compute histogram bins on GPU
const histogram = await computeHistogram(myTensor, { bins: 100 });

Next Steps

Renderers - Draw the results of your analysis.
torch.js Performance - Why GPU analysis is faster than CPU.