ta-lib Checked

Create sample data

close = np.random.randn(100).cumsum() + 50

high = close + np.abs(np.random.randn(100))

low = close - np.abs(np.random.randn(100))

open_ = close + np.random.randn(100) * 0.5

volume = np.random.randint(1000, 10000, 100).astype(float)

Function API — pass arrays directly

rsi = talib.RSI(close, timeperiod=14)

macd, signal, hist = talib.MACD(close, fastperiod=12, slowperiod=26, signalperiod=9)

upper, middle, lower = talib.BBANDS(close, timeperiod=20, nbdevup=2, nbdevdn=2)

atr = talib.ATR(high, low, close, timeperiod=14)

Candlestick patterns — return +100 (bullish), -100 (bearish), or 0

doji = talib.CDLDOJI(open_, high, low, close)

hammer = talib.CDLHAMMER(open_, high, low, close)

engulfing = talib.CDLENGULFING(open_, high, low, close)

## Function API vs Abstract API

### Function API (Recommended)

Call functions directly with NumPy arrays:

import talib

rsi = talib.RSI(close, timeperiod=14)

sma = talib.SMA(close, timeperiod=20)

upper, mid, lower = talib.BBANDS(close)

### Abstract API

Pass a dictionary of arrays and get results by name:

from talib import abstract

inputs = {"open": open_, "high": high, "low": low, "close": close, "volume": volume}

Call by function name

rsi = abstract.RSI(inputs, timeperiod=14)

macd = abstract.MACD(inputs) # returns (macd, signal, hist)

The abstract API is useful for dynamic indicator selection (e.g., looping over a list of indicator names).

## Function Groups

TA-Lib organizes functions into these groups:

### Overlap Studies
Moving averages and envelope indicators that overlay price charts.

sma = talib.SMA(close, timeperiod=20)

ema = talib.EMA(close, timeperiod=12)

upper, mid, lower = talib.BBANDS(close, timeperiod=20, nbdevup=2, nbdevdn=2)

sar = talib.SAR(high, low, acceleration=0.02, maximum=0.2)

mama, fama = talib.MAMA(close, fastlimit=0.5, slowlimit=0.05)

### Momentum Indicators
Oscillators and trend-strength measures.

rsi = talib.RSI(close, timeperiod=14)

macd, signal, hist = talib.MACD(close, fastperiod=12, slowperiod=26, signalperiod=9)

slowk, slowd = talib.STOCH(high, low, close)

cci = talib.CCI(high, low, close, timeperiod=14)

willr = talib.WILLR(high, low, close, timeperiod=14)

adx = talib.ADX(high, low, close, timeperiod=14)

mfi = talib.MFI(high, low, close, volume, timeperiod=14)

### Volume Indicators
Volume-based analysis functions.

obv = talib.OBV(close, volume)

ad = talib.AD(high, low, close, volume)

adosc = talib.ADOSC(high, low, close, volume, fastperiod=3, slowperiod=10)

### Volatility Indicators
Measures of price variability.

atr = talib.ATR(high, low, close, timeperiod=14)

natr = talib.NATR(high, low, close, timeperiod=14)

trange = talib.TRANGE(high, low, close)

### Pattern Recognition (Candlestick)
61 functions that detect candlestick patterns. All return integer arrays:
- `+100` = bullish pattern detected
- `-100` = bearish pattern detected
- `0` = no pattern

Single patterns

doji = talib.CDLDOJI(open_, high, low, close)

hammer = talib.CDLHAMMER(open_, high, low, close)

engulfing = talib.CDLENGULFING(open_, high, low, close)

Scan all 61 patterns at once

candle_names = talib.get_function_groups()["Pattern Recognition"]

for name in candle_names:

func = getattr(talib, name)

result = func(open_, high, low, close)

hits = np.nonzero(result)[0]

if len(hits) > 0:

print(f"{name}: {len(hits)} detections")

See `references/candlestick_patterns.md` for the full list of 61 patterns with reliability ratings and crypto relevance.

### Math Transform & Math Operators
Mathematical functions (sin, cos, ln, etc.) and operators (add, sub, mult, div) on arrays. Rarely used directly but available.

## Crypto Considerations

### 24/7 Markets
- Candlestick patterns designed for traditional markets with opening/closing gaps may behave differently on crypto's continuous markets
- Gap-based patterns (morning star, evening star) are less reliable without session gaps
- Body-ratio patterns (doji, hammer, engulfing) still work well on any timeframe

### Timeframe Selection
- **1m-5m**: Patterns are noisy; combine with volume confirmation
- **15m-1h**: Good for intraday signals on high-cap tokens
- **4h-1d**: Most reliable for pattern recognition
- **Tip**: Higher timeframes produce fewer but more reliable pattern signals

### NaN Handling
TA-Lib returns `NaN` for the initial lookback period of each indicator. Always account for this:

rsi = talib.RSI(close, timeperiod=14)

First 14 values will be NaN

valid_rsi = rsi[~np.isnan(rsi)]

### Solana Token Data
When using TA-Lib with Solana token OHLCV data:
- Ensure arrays are `float64` dtype — TA-Lib requires this
- Sort by timestamp ascending before passing to TA-Lib
- Handle gaps in low-liquidity token data before computing indicators

Convert to float64 for TA-Lib compatibility

close = df["close"].values.astype(np.float64)

high = df["high"].values.astype(np.float64)

low = df["low"].values.astype(np.float64)

## Integration with Other Skills

### With pandas-ta
pandas-ta can use TA-Lib as a backend when installed, getting C-speed through the pandas-ta API:

import pandas_ta as ta

pandas-ta auto-detects TA-Lib and uses it for supported