taquant 0.1.0

High-performance technical indicator library

TAQuant

TAQuant is a high-performance technical indicator library for quantitative research and trading systems. It uses a layered design:

• pandas for user-facing API
• numpy arrays for heavy computation
• numba kernels for hot loops

TAQuant targets:

• large batch backtests (100k to 1M rows)
• multi-symbol processing
• realtime incremental updates via stateful classes

Installation

pip install -e .

Optional extras:

pip install -e .[dev]
pip install -e .[talib]
pip install -e .[benchmark]

Quickstart

import pandas as pd
import taquant

# df contains: Timestamp, Open, High, Low, Close, Volume
out = taquant.trend.ema(df, window=20)
out = taquant.momentum.rsi(out, window=14)
out = taquant.volatility.bollinger_bands(out, window=20, num_std=2.0)

# Warm-path mode: skip runtime validation overhead
out = taquant.trend.ema(df, window=20, fast_mode=True)

API Examples

from taquant.trend import sma, ema, wma, macd
from taquant.momentum import rsi, stochastic, cci
from taquant.volatility import stddev, atr, bollinger_bands
from taquant.volume import volume_ma, obv, vwap
from taquant.price import typical_price, median_price, weighted_close

sma(df, window=20)
ema(df, window=20, symbol_col="Symbol")
macd(df, fast_window=12, slow_window=26, signal_window=9)

Array-native API (ndarray -> ndarray, no DataFrame overhead):

from taquant import array_api as fa

ema = fa.ema_array(close, window=20)
rsi = fa.rsi_array(close, window=14)
line, signal, hist = fa.macd_array(close, 12, 26, 9)

Stateful realtime usage:

from taquant.stateful import EMAState, RSIState

ema_state = EMAState(window=20)
rsi_state = RSIState(window=14)

ema_value = ema_state.update(price)
rsi_value = rsi_state.update(price)

Benchmark Highlights

From benchmark/results/speedup_summary.csv (warm mode, single_1m, speedup vs pandas):

• SMA: 2.04x
• EMA: 3.45x
• MACD: 6.39x
• RSI: 4.61x
• BBANDS: 5.72x

Artifacts:

• benchmark/results/benchmark_summary.csv
• benchmark/results/speedup_summary.csv
• benchmark/results/runtime_speedup.png
• benchmark/results/metadata.json

Design Philosophy

• Performance-first compute paths on ndarray(float64)
• Deterministic, explicit validation and error messages
• Reusable base abstraction (apply_indicator) for consistent APIs
• Dependency-aware caching to avoid redundant recompute
• Maintainable architecture over over-engineering