Screamer

A fast C++ library for researching and trading time series: rolling statistics, technical indicators, filters, order-flow measures, costed backtests of single assets, portfolios, and pairs, and significance tests. One C++ core, two languages: a Python API and a JavaScript/WebAssembly build.

Screamer runs in Python and in JavaScript. Both bind the same C++ engine, so results agree to within floating-point rounding, and seeded randomization gives identical draws in both.

Why screamer

• Fast. Every operator is implemented in C++ and routinely outruns equivalent NumPy and pandas code, often by a factor of two or more.
• One API, batch or streaming. The same operator runs on a stored array or a live, event-driven stream, so code tested on historical data deploys to production unchanged.
• Causal by construction. Output depends only on current and past inputs, never future ones, which eliminates look-ahead bias.
• Broad coverage. 250+ operators: rolling and exponentially weighted statistics, technical indicators (MACD, RSI, Bollinger Bands, ATR, and more), OHLC volatility estimators, signal filters, market microstructure, stream operators, and composable pipelines.
• Backtesting and validation. Costed backtests on prices, bars, trades, or quotes, with slippage, fill timing, stops, and funding; portfolios of many assets from one account; pairs and cross-sectional tools; and randomization tests, the deflated Sharpe ratio, and walk-forward evaluation to tell skill from luck.
• Python and JavaScript. The same operators and the same pipeline model in both ecosystems, from one C++ source.

Python

Python 3.12 or newer.

pip install screamer

The wheel is self-contained (a single abi3 wheel per platform, no build step and no runtime dependencies). To build from source, see the Install screamer page.

Fit a quadratic to each sliding window of 50 values, take its slope at the latest value, then the sign of the slope to get the trend direction:

import numpy as np
from screamer import RollingPoly2, Sign

data = np.cumsum(np.random.normal(size=300))

slope = RollingPoly2(window_size=50, derivative_order=1)
sign = Sign()

trend = sign(slope(data))   # the same calls work on a live stream, one value at a time

JavaScript

Node 18 or newer, or Chrome 95, Firefox 100 or Safari 15.2 or newer. The WebAssembly module uses native WebAssembly exceptions, which older browsers lack.

npm install @screamer-labs/screamer

The package is self-contained: the WebAssembly module is embedded, so there is no separate asset to configure. The WASM loads once via await ready(), then you call operators directly:

import { ready, RollingPoly2, Sign } from "@screamer-labs/screamer";
await ready();

const data = Float64Array.from({ length: 300 }, () => Math.random());

const slope = RollingPoly2(50, 1);   // window_size, derivative_order
const sign = Sign();

const trend = sign(slope(data));     // a number, a Float64Array, an iterable, or an async stream

Operators compose into pipelines:

import { ready, Input, Pipeline, RollingMean, Diff } from "@screamer-labs/screamer";
await ready();

const x = Input("x");
const y = Diff(1)(RollingMean(3)(x));      // compose operators into a graph
const p = new Pipeline([x], [y]);
p(data);                                    // bind data at call time; p.live() streams

Documentation

• Start here: the Guide goes from installing screamer to a costed backtest in short task pages.
• Python: the Guide, runnable notebooks, and the full reference grouped by topic live at screamer.readthedocs.io.
• JavaScript: see js/README.md for install and the API. JavaScript API reference and guide: screamer-labs.github.io/screamer.
• AI agents: python -m screamer.docs search <keywords> and show <Name> describe every function of the installed version, and pip install "screamer[mcp]" adds an MCP server. See Using screamer with AI agents.

Contributing

Contributions are welcome. See CONTRIBUTING.md for how to set up a development environment, build the extension, run the tests, and open a pull request. By participating you agree to abide by our Code of Conduct.

License

Screamer is released under the MIT License. See LICENSE.