ctalearn 0.1.0

A cross-sectional / time-series alpha research toolkit for CTA strategies

ctalearn

A cross-sectional / time-series alpha research toolkit for CTA strategies.

ctalearn is built around a thin DataFrame wrapper over Polars that guarantees a dense, regularly-sampled time grid. On top of it sits a library of vectorized operators (cross-sectional, time-series, arithmetic, strategy), a Numba-accelerated backtester, Glassnode data ingestion, and a small embedded DSL for expressing alphas as plain text.

Key features

• Aligned DataFrame — wraps pl.DataFrame, auto-detects bar frequency, fills gaps to a dense grid, and aligns operands automatically on arithmetic (+ - * /).
• Operator library — cs_* (across assets), ts_* (through time), element-wise transforms, and signal-to-weight strategies. Native Polars where possible; Numba only where Polars has no equivalent.
• Alpha DSL — a Lark grammar for expressions like cs_rank(ts_mean(close, 20) - open), with static type checking before execution. Ships its built-in function registry so consumers only provide their own data loaders.
• Backtesting — Numba-accelerated simulate_trade(weights, prices, fee).
• Data — cached Glassnode REST ingestion, single-asset and parallel cross-sectional fetch.

Installation

Requires Python 3.13+. Add it to your project straight from GitHub (managed with uv):

uv add git+https://github.com/fffelix-huang/ctalearn.git

Quick start

import polars as pl
from ctalearn import DataFrame
from ctalearn.operator import ts_mean, cs_rank

prices = DataFrame(
    pl.DataFrame({
        "ts":  pl.datetime_range(..., interval="1d", eager=True),
        "BTC": [...],
        "ETH": [...],
    }),
    time_col="ts",
)

# 20-bar moving average, then rank across assets each bar.
signal = cs_rank(prices - ts_mean(prices, 20))

Alpha DSL

Express the same idea as text and run it through the interpreter:

from ctalearn.dsl import (
    parser, TypeCheckTransformer, ExecutionTransformer,
    BUILTIN_FUNCTIONS, BUILTIN_FUNCTION_SCHEMA, DslType,
)

code = "return cs_rank(close - ts_mean(close, 20));"

# Caller supplies only the factor side; functions come from the library.
factor_schema = {"close": DslType.DATAFRAME}
data_loaders  = {"close": lambda: prices}

tree   = parser.parse(code)
TypeCheckTransformer(factor_schema, BUILTIN_FUNCTION_SCHEMA).transform(tree)
result = ExecutionTransformer(BUILTIN_FUNCTIONS, data_loaders).transform(tree)

See examples/dsl_interpreter.py for a complete, runnable integration.

Development

git clone https://github.com/fffelix-huang/ctalearn.git
cd ctalearn
uv sync                            # install / sync dependencies
uv run pytest                      # run tests (doctests + coverage; benchmarks skipped)
uv run pytest --benchmark-only     # run only the benchmarks
uv run ruff check --fix            # lint
uv run ruff format                 # format
uv run pre-commit run --all-files  # all hooks

License

MIT