TA Indicators

Vectorized technical analysis library that implements classic indicators from scratch with NumPy — no TA-Lib, pandas-ta, or third-party dependencies. Fourth module of the quant-core-infra ecosystem.

Repository: github.com/juanmmm21/ta-indicators-from-scratch

What it is and what problem it solves

Indicator libraries (TA-Lib, etc.) are black boxes: they return a number without the developer understanding the exact math. In a quantitative portfolio that's a problem: you can't audit, optimize, or explain the signals.

This project implements the formulas explicitly using vectorized NumPy operations:

• You know exactly what happens in each warmup window
• You can compose indicators into your own pipelines
• Performance stays at C level thanks to vectorization

Consumes OHLCV candles from market-data-lakehouse and feeds alpha-signal-generator.

---

Role in quant-core-infra

market-data-lakehouse ──► OHLCV candles ──► ta-indicators-from-scratch ──► indicators
                                                    │
                                          alpha-signal-generator

Translates historical price into statistical features that strategies evaluate bar by bar.

---

Supported indicators

All return NaN during the warmup window until enough candles have accumulated.

---

How it works

1. Input: float64 arrays or JSONL with OHLCV fields.
2. Container: OhlcvSeries validates lengths and finite values.
3. Calculation: each indicator is a pure ndarray → ndarray function.
4. Engine: TechnicalAnalysisEngine orchestrates all indicators over the same series.
5. Output: IndicatorResult with arrays aligned by candle index.
6. CLI: serializes to JSON with null wherever there's a NaN.

---

Architecture

OHLCV candles (JSONL / arrays)
        │
        ▼
OhlcvSeries
        │
        ▼
TechnicalAnalysisEngine
   ├─ sma / ema / rsi
   ├─ macd (line, signal, histogram)
   └─ bollinger_bands (mean, upper, lower)
        │
        ▼
IndicatorResult

Components

Technical decisions

• float64 for statistical math (acceptable for pure TA per ecosystem rules)
• No pandas / TA-Lib — minimal dependency (NumPy only)
• Explicit NaN — downstream strategies can filter warmup
• MACD: the signal is calculated only over the finite portion of the MACD line

---

Requirements

• Python 3.11+

---

Installation

cd ta-indicators-from-scratch
python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"

---

CLI usage

# All indicators
ta-indicators-from-scratch compute \
  --input samples/btcusdt_1m_candles.jsonl \
  --indicator all

# RSI only with custom period
ta-indicators-from-scratch compute \
  --input samples/btcusdt_1m_candles.jsonl \
  --indicator rsi \
  --rsi-period 14 \
  --output indicators.json

Configurable parameters

---

Candle JSONL format

{
  "open": 100.5,
  "high": 101.0,
  "low": 99.5,
  "close": 100.8,
  "volume": 12.0
}

---

Programmatic usage

import numpy as np

from ta_indicators_from_scratch import OhlcvSeries, TechnicalAnalysisEngine, rsi, sma

close = np.linspace(100.0, 130.0, num=60, dtype=np.float64)
series = OhlcvSeries.from_arrays(
    open_=close - 0.5,
    high=close + 1.0,
    low=close - 1.0,
    close=close,
    volume=np.full(60, 10.0),
)

engine = TechnicalAnalysisEngine()
results = engine.compute_all(series)

# Direct access to individual functions
rsi_values = rsi(series.close, period=14)
sma_values = sma(series.close, period=20)

---

Development

pytest -q
ruff check src tests
mypy src

---

Roadmap

• Additional indicators (ATR, Stochastic, VWAP)
• Parquet export of indicator series
• Direct pipeline from the lakehouse's DuckDB

---

License

MIT