pandas-ts

pandas for TypeScript & JavaScript, with a C++ core so it can run at native speed.

pandas-ts aims to be a 100% API-compatible port of Python's pandas for the JavaScript ecosystem, with full TypeScript types. Heavy computation runs in a C++ core, exposed as a Node.js native addon (N-API) and compiled to WebAssembly for browsers. The goal is to get as close as possible to the speed of pandas itself.

⚠️ Project status

This is a pre-alpha placeholder release (0.0.x). It reserves the package name and sets up the project. The DataFrame and Series APIs are not available yet. Don't use this in production.

Follow the roadmap to see progress, and watch the repository for releases.

Vision

• Familiar API. If you know pandas, you should already know pandas-ts. Method names, arguments, and behavior follow pandas closely. Where they differ, it's because of JavaScript language limits, and those differences are documented.
• Native performance. Columnar memory, vectorized kernels, and multi-threaded operations are written in modern C++ and run with no copying between JS and native memory where possible.
• Runs anywhere. Node.js, Bun, and Deno use the native addon. Browsers and edge runtimes use the WebAssembly build. A pure-JS fallback covers everything else.
• TypeScript-first. Strong types for dtypes, columns, and method results.
• Interoperable. Planned support for Apache Arrow, CSV, JSON, Parquet, and Excel.

Planned API (preview)

The following shows the intended API. It does not work in the current release.

import * as pd from 'pandas-ts';

const df = pd.DataFrame({
  city: ['Delhi', 'Mumbai', 'Delhi', 'Pune'],
  sales: [120, 340, 90, 210],
  year: [2024, 2024, 2025, 2025],
});

df.head();
df.describe();

const summary = df
  .groupby('city')
  .agg({ sales: ['sum', 'mean'] })
  .sort_values(['sales', 'sum'], { ascending: false });

const csv = await pd.read_csv('data.csv');
csv.to_parquet('data.parquet');

Installation

npm install pandas-ts
# or
yarn add pandas-ts
# or
pnpm add pandas-ts

Prebuilt native binaries are planned for macOS (x64/arm64), Linux (x64/arm64, glibc and musl), and Windows (x64). On other platforms the WebAssembly build will be used automatically.

Current usage (0.0.x)

Only basic build information is exported for now:

import { VERSION, info, isNativeAvailable } from 'pandas-ts';

console.log(VERSION);             // "0.0.1"
console.log(info());              // { version, stage: 'pre-alpha', ... }
console.log(isNativeAvailable()); // false (native core not shipped yet)

CommonJS also works:

const { VERSION } = require('pandas-ts');

Architecture

┌──────────────────────────────────────────────┐
│        TypeScript API (pandas-compatible)    │
│   DataFrame · Series · Index · GroupBy · IO  │
└───────────────┬──────────────────────────────┘
                │ zero-copy buffers (TypedArrays)
     ┌──────────┴───────────┬──────────────────┐
     ▼                      ▼                  ▼
┌──────────────┐   ┌────────────────┐   ┌─────────────┐
│ Native addon │   │  WebAssembly   │   │  Pure JS    │
│  (N-API)     │   │  (browser/edge)│   │  fallback   │
└──────┬───────┘   └───────┬────────┘   └─────────────┘
       └─────────┬─────────┘
                 ▼
     ┌───────────────────────────┐
     │      C++ core engine      │
     │ columnar storage · SIMD   │
     │ hashing · sort · groupby  │
     │ joins · rolling · parsers │
     └───────────────────────────┘

Roadmap

• [x] Reserve package name and set up the project (0.0.1)
• [ ] C++ core: columnar storage, dtypes (int, float, bool, string, datetime, category), null handling
• [ ] N-API bindings and prebuilt binaries for major platforms
• [ ] WebAssembly build and pure-JS fallback
• [ ] Series: construction, indexing, arithmetic, comparisons, reductions
• [ ] DataFrame: construction, loc / iloc, selection, assignment, head / tail / describe
• [ ] Index, RangeIndex, DatetimeIndex, MultiIndex
• [ ] groupby, agg, transform, apply
• [ ] merge, join, concat
• [ ] pivot, pivot_table, melt, stack / unstack
• [ ] rolling, expanding, ewm, resample
• [ ] Missing data: isna, fillna, dropna, interpolate
• [ ] String (.str) and datetime (.dt) accessors
• [ ] IO: CSV, JSON, Parquet, Arrow, Excel
• [ ] Benchmarks against pandas, Polars, and Danfo.js
• [ ] Documentation site and API reference

Contributing

Contributions are welcome. See CONTRIBUTING.md.

License

MIT © Rajan

pandas-ts is an independent project. It is not affiliated with or endorsed by the pandas development team or NumFOCUS.