Technical Indicators

ML4T Engineer provides 120 registry features across 11 categories, built on Polars with Numba JIT for performance-critical kernels. 60 indicators are validated against TA-Lib at 1e-6 tolerance.

Overview

Category	Count	TA-Lib	Key Indicators
Momentum	31	19	RSI, MACD, Stochastic, CCI, ADX, MFI
Volatility	15	4	ATR, Bollinger, Yang-Zhang, Parkinson, GARCH
Microstructure	15	0	Kyle Lambda, Amihud, Roll Spread, Realized Spread
Trend	10	9	SMA, EMA, WMA, DEMA, TEMA, KAMA
ML Features	14	0	Lag, Entropy, Fourier, Cyclical Encode
Statistics	14	7	STDDEV, Linear Regression, TSF, Variance Ratio
Risk	6	0	Max Drawdown, Downside Deviation, Sharpe, Sortino
Price Transform	5	5	Typical Price, Weighted Close, Average Price
Regime	4	0	Hurst Exponent, Choppiness, Fractal Efficiency
Volume	3	3	OBV, AD, ADOSC
Math	3	3	Maximum, Minimum, Summation

Standalone cross-asset utilities such as beta, rolling correlation, and cointegration are documented below, but they are not part of the 120-feature registry count.

The pinned book notebooks combine direct Engineer calls with manual teaching implementations. Start with The ml4t Library Ecosystem for compute_features(), then use the Book Guide to check each notebook's exact relationship to the API.

Computation API

compute_features() accepts three input formats:

from datetime import date, timedelta

import polars as pl
from ml4t.engineer import compute_features

close = [100.0 + i * 0.1 + (i % 5) * 0.05 for i in range(100)]
df = pl.DataFrame({
    "timestamp": [date(2024, 1, 1) + timedelta(days=i) for i in range(100)],
    "open": close,
    "high": [price + 1.0 for price in close],
    "low": [price - 1.0 for price in close],
    "close": close,
    "volume": [100_000 + i * 100 for i in range(100)],
})

# 1. List of names (default parameters)
result = compute_features(df, ["rsi", "macd", "atr"])

# 2. List of dicts (custom parameters)
result = compute_features(df, [
    {"name": "rsi", "params": {"period": 20}},
    {"name": "sma", "params": {"period": 50}},
    {
        "name": "bollinger_bands",
        "params": {"period": 20, "nbdevup": 2.5, "nbdevdn": 2.5},
    },
])

assert {"rsi", "sma", "bollinger_bands"} <= set(result.columns)

Features are computed in dependency order (topological sort). Circular dependencies raise ValueError. The return type matches the input: DataFrame in, DataFrame out; LazyFrame in, LazyFrame out.

Input is sorted by asset and timestamp before computation. Common asset columns such as asset_id, symbol, and ticker are detected automatically, and every rolling or lagged feature is evaluated independently for each asset. If no timestamp column exists, pass assume_sorted=True to make the row-order assumption explicit.

Use output when requesting more than one configuration of the same feature:

result = compute_features(df, [
    {"name": "sma", "params": {"period": 20}, "output": "sma_20"},
    {"name": "sma", "params": {"period": 50}, "output": "sma_50"},
])

Unknown parameters, repeated output names, and output names that replace input columns raise ValueError before feature execution.

Book: The ml4t Library Ecosystem calls compute_features() with names and parameter dictionaries, and compares the library RSI with a manual implementation.

Category Reference

Momentum (31 indicators)

Price momentum and oscillator indicators. Most produce bounded (normalized) output suitable for direct ML use.

Name	Description	TA-Lib	Normalized	Default Period
rsi	Relative Strength Index	Yes	0-100	14
macd	Moving Average Convergence/Divergence	Yes	No	12/26
stochastic	Stochastic Oscillator (%K, %D)	No	0-100	14/3/3
stochf	Fast Stochastic	Yes	0-100	5/3
stochrsi	Stochastic RSI	Yes	0-100	14
cci	Commodity Channel Index	Yes	~-200 to 200	14
willr	Williams %R	Yes	-100 to 0	14
adx	Average Directional Index	Yes	0-100	14
adxr	ADX Rating	Yes	0-100	14
dx	Directional Movement Index	Yes	0-100	14
plus_di	Plus Directional Indicator	Yes	0-100	14
minus_di	Minus Directional Indicator	Yes	0-100	14
mfi	Money Flow Index	Yes	0-100	14
roc	Rate of Change	Yes	No	10
rocp	Rate of Change (%)	Yes	No	10
mom	Momentum	Yes	No	10
trix	Triple Exponential Average	Yes	No	30
cmo	Chande Momentum Oscillator	Yes	-100 to 100	14
ultosc	Ultimate Oscillator	Yes	0-100	7/14/28
bop	Balance of Power	Yes	-1 to 1	-
imi	Intraday Momentum Index	No	0-100	14
aroon	Aroon (up/down)	Yes	0-100	14
aroonosc	Aroon Oscillator	Yes	-100 to 100	14
apo	Absolute Price Oscillator	Yes	No	12/26
ppo	Percentage Price Oscillator	Yes	No	12/26
sar	Parabolic SAR	Yes	No	0.02/0.2

Book: Price and Volume Feature Families calls Engineer for selected indicators and derives others manually. ETFs: Feature Engineering calls individual Engineer feature functions in a case-study pipeline.

Trend (10 indicators)

Moving averages that produce price-scale outputs. Require preprocessing for ML models.

Name	Description	TA-Lib	Default Period
sma	Simple Moving Average	Yes	20
ema	Exponential Moving Average	Yes	20
wma	Weighted Moving Average	Yes	20
dema	Double Exponential MA	Yes	20
tema	Triple Exponential MA	Yes	20
t3	Triple Exponential T3	Yes	5
kama	Kaufman Adaptive MA	Yes	30
trima	Triangular MA	Yes	20
midpoint	Midpoint over period	Yes	14
donchian_channels	Donchian Channels (highest high/lowest low)	No	20

Volatility (15 indicators)

Volatility estimators ranging from simple (ATR) to advanced (GARCH). Includes range-based estimators that are more efficient than close-to-close.

Name	Description	TA-Lib	Normalized
atr	Average True Range	Yes	No
natr	Normalized ATR (% of price)	Yes	0-100
trange	True Range	Yes	No
bollinger_bands	Bollinger Bands (upper/middle/lower)	Yes	No
yang_zhang_volatility	Yang-Zhang (overnight + intraday)	No	No
parkinson_volatility	Parkinson range-based	No	No
garman_klass_volatility	Garman-Klass OHLC-based	No	No
rogers_satchell_volatility	Rogers-Satchell drift-independent	No	No
realized_volatility	Standard deviation of returns	No	No
ewma_volatility	EWMA of variance	No	No
garch_forecast	GARCH(1,1) conditional volatility	No	No
conditional_volatility_ratio	Up-market vs down-market vol ratio	No	No
volatility_percentile_rank	Current vol vs historical distribution	No	0-100
volatility_of_volatility	Second-order volatility measure	No	No
volatility_regime_probability	High/low vol regime probability	No	No

Efficiency ranking: Yang-Zhang > Garman-Klass ~ Rogers-Satchell > Parkinson > Close-to-Close. See Molnar (2012) for theoretical efficiency ratios.

Book: Price and Volume Feature Families calls Engineer's volatility functions and compares selected estimators on ETF data.

Microstructure (15 indicators)

Market microstructure features from De Prado (2018) and empirical market microstructure literature.

Name	Description
kyle_lambda	Kyle's Lambda (price impact coefficient)
amihud_illiquidity	Amihud illiquidity ratio
roll_spread_estimator	Roll implied bid-ask spread
realized_spread	Realized spread
effective_tick_rule	Effective tick rule classification
order_flow_imbalance	Order flow imbalance
price_impact_ratio	Price impact ratio
volume_weighted_price_momentum	Volume-weighted price momentum
bid_ask_imbalance	Bid-ask imbalance (normalized -1 to 1)
book_depth_ratio	Book depth ratio (normalized 0 to 1)
quote_stuffing_indicator	Quote stuffing detection
trade_intensity	Trade intensity
volume_at_price_ratio	Volume at price ratio
volume_synchronicity	Volume synchronicity
weighted_mid_price	Weighted mid price

Book: Microstructure Features calls Engineer's tick-rule and liquidity functions while teaching the wider workflow. NASDAQ-100 Microstructure: Feature Engineering calls amihud_illiquidity and implements other case-study features separately.

ML Features (14 indicators)

Features designed specifically for machine learning pipelines.

Name	Description	Normalized
create_lag_features	Multiple lag columns at once	No
cyclical_encode	Cyclical time encoding (sin/cos)	No
fourier_features	Fourier transform features	No
rolling_entropy	Shannon entropy	0-10
rolling_entropy_lz	Lempel-Ziv entropy	0-10
rolling_entropy_plugin	Plugin entropy estimator	0-10
percentile_rank_features	Rank-based normalization	0-100
interaction_features	Feature interaction terms	No
multi_horizon_returns	Returns at multiple horizons	No
directional_targets	Directional movement targets	No
volatility_adjusted_returns	Returns scaled by volatility	No
regime_conditional_features	Regime-conditional transforms	No
time_decay_weights	Exponential time decay	No
ffdiff	Fractional differencing	No

Book: Slow Features and Context calls Engineer's cyclical_encode and teaches the point-in-time data joins manually.

Risk (6 indicators)

Risk and risk-adjusted return metrics.

Name	Description	Normalized
maximum_drawdown	Maximum drawdown	No
downside_deviation	Downside volatility	0-2
tail_ratio	Right tail / left tail ratio	0-10
higher_moments	Skewness and kurtosis	No
risk_adjusted_returns	Sharpe, Sortino, Calmar, Omega	No
ulcer_index	Ulcer Index (drawdown-based risk)	No

risk_adjusted_returns accepts trading_periods to match the annualization frequency to the data, such as 252 for daily observations or 52 for weekly observations. Its annual risk_free_rate is converted to a per-period threshold using the same value.

Cross-Asset (10 functions)

Multi-asset relationship features. These are standalone functions in ml4t.engineer.features.cross_asset rather than registry entries, since they require two or more price series as input.

Function	Description
rolling_correlation	Rolling Pearson correlation
beta_to_market	Rolling beta vs market index
correlation_regime_indicator	Low/medium/high correlation regimes
lead_lag_correlation	Lead-lag cross-correlation
multi_asset_dispersion	Cross-sectional return dispersion
correlation_matrix_features	Mean/min/max of correlation matrix
relative_strength_index_spread	RSI spread between two assets
volatility_ratio	Volatility ratio between assets
co_integration_score	Rolling cointegration score
cross_asset_momentum	Rank-based cross-asset momentum

These are called directly (not via compute_features) since they require multi-asset DataFrames.

Book: Structural and Cross-Instrument Features calls beta_to_market. Panel Features calls several cross-asset functions and combines them with manual analysis.

Regime (4 indicators)

Market regime detection features. All produce bounded outputs suitable for direct ML use.

Name	Description	Range
hurst_exponent	Hurst exponent (R/S analysis)	0-1
choppiness_index	Market choppiness	0-100
fractal_efficiency	Price path efficiency	0-1
trend_intensity_index	Trend strength	0-100

Statistics (14 indicators)

Statistical features including TA-Lib standard and rolling distribution metrics.

Name	Description	TA-Lib	Normalized
stddev	Standard Deviation	Yes	No
var	Variance	Yes	No
avgdev	Average Deviation	No	No
linearreg	Linear Regression Value	Yes	No
linearreg_slope	Linear Regression Slope	Yes	No
linearreg_angle	Linear Regression Angle	Yes	No
linearreg_intercept	Linear Regression Intercept	Yes	No
tsf	Time Series Forecast	Yes	No
coefficient_of_variation	Rolling coefficient of variation	No	0-10
variance_ratio	Variance ratio test	No	0-5
rolling_cv_zscore	Cross-validated z-score	No	-10 to 10
rolling_drift	Rolling drift estimate	No	-10 to 10
rolling_kl_divergence	KL divergence vs reference	No	0-10
rolling_wasserstein	Wasserstein distance	No	No

Price Transform (5 indicators)

Name	Description	TA-Lib
avgprice	Average Price (O+H+L+C)/4	Yes
typprice	Typical Price (H+L+C)/3	Yes
medprice	Median Price (H+L)/2	Yes
wclprice	Weighted Close (H+L+2C)/4	Yes
midprice	Midpoint Price (H+L)/2	Yes

Volume (3 indicators)

Name	Description	TA-Lib
obv	On Balance Volume	Yes
ad	Accumulation/Distribution	Yes
adosc	A/D Oscillator	Yes

Book: ETFs: Feature Engineering calls Engineer's volume, momentum, trend, volatility, and regime feature functions.

Math (3 indicators)

O(n) rolling operations using monotonic deque.

Name	Description	TA-Lib
maximum	Rolling maximum	Yes
minimum	Rolling minimum	Yes
summation	Rolling sum	Yes

Fractional Differencing (4 functions)

See the dedicated Fractional Differencing guide for the full workflow.

Feature Discovery

The FeatureCatalog provides filtering and full-text search over all 120 features:

from ml4t.engineer import feature_catalog

# List by category
momentum = feature_catalog.list(category="momentum")

# Filter by multiple criteria
ml_ready = feature_catalog.list(normalized=True, ta_lib_compatible=True)

# Full-text search
results = feature_catalog.search("volatility estimator")
# Returns: [("parkinson_volatility", 0.65), ("garman_klass_volatility", 0.45), ...]

# Detailed feature info
info = feature_catalog.describe("yang_zhang_volatility")
# {'name': 'yang_zhang_volatility', 'category': 'volatility', ...}

# List all categories
print(feature_catalog.categories())
# ['cross_asset', 'math', 'microstructure', 'ml', 'momentum', ...]

# List all tags
print(feature_catalog.tags())

See the dedicated Feature Discovery guide for complete examples.

Book: The ml4t Library Ecosystem inspects registry metadata for RSI, ATR, and Garman-Klass before computing features.

YAML Configuration

For reproducible feature pipelines, store configurations in YAML:

# features.yaml
features:
  - name: rsi
    params:
      period: 14

  - name: macd
    params:
      fast_period: 12
      slow_period: 26

  - name: bollinger_bands
    params:
      period: 20
      nbdevup: 2.0
      nbdevdn: 2.0

  - name: yang_zhang_volatility

Load with compute_features(df, "features.yaml"). The YAML format supports version comments and parameter documentation inline.

Input Requirements

OHLCV DataFrame

Most features expect a DataFrame with standardized column names (lowercase):

Column	Type	Required By
timestamp	datetime	Temporal ordering, unless assume_sorted=True
asset_id	string	Optional panel grouping, auto-detected when present
open	float	OHLCV, OHLC features
high	float	OHLCV, OHLC, HLC, HL features
low	float	OHLCV, OHLC, HLC, HL features
close	float	All features
volume	float	OHLCV, volume features
returns	float	Return-based features (auto-computed if missing)

Features declare their input_type metadata (e.g., "OHLCV", "close", "returns"), and compute_features validates that required columns are present.

Missing Columns

If a feature requires a column that's missing, compute_features raises a clear error:

ValueError: Feature 'mfi' requires column 'volume' (input_type='OHLCV') but it was not found.

Custom Parameters

Override default parameters per feature:

# Check defaults
from ml4t.engineer.core.registry import get_registry
meta = get_registry().get("rsi")
print(meta.parameters)  # {'period': 14}

# Override
result = compute_features(df, [{"name": "rsi", "params": {"period": 20}}])

Invalid parameters raise ValueError with the valid parameter names.

Performance

• Polars-native: All computations use Polars expressions for automatic parallelism
• Numba JIT: Numerical kernels (volatility estimators, microstructure) are Numba-accelerated
• compute_features resolves feature dependencies before execution.
• LazyFrame input remains lazy until you collect the result.
• Measure throughput with your feature set, row count, grouping columns, and hardware.

See It In The Book

• The ml4t Library Ecosystem calls the registry and compute_features().
• Price and Volume Feature Families combines Engineer calls with manual teaching implementations.
• ETFs: Feature Engineering calls individual Engineer functions in a case-study pipeline.
• Book Guide records the pinned revision and relationship for every selected notebook.

Next Steps

• Read Feature Discovery to choose features through metadata and search instead of hardcoding.
• Read ML Readiness to separate normalized from non-normalized outputs.
• Read Dataset Builder when features are ready to move into training workflows.
• Use the API Reference for exact function and module locations.

References

• Lopez de Prado, M. (2018). Advances in Financial Machine Learning. Wiley.
• Molnar, P. (2012). Properties of range-based volatility estimators. International Review of Financial Analysis.