Python API¶

QuantLab exposes its documented entry points through package __all__ lists and the quantlab.config module. Import from the location shown below rather than from private modules or names beginning with an underscore.

Package	Main public objects
`quantlab.config`	`ExperimentConfig` and its validated configuration models
`quantlab.data`	`DataLoader`, `DataValidator`, `DataCleaner`, `ParquetStorage`, `Universe`, schema and resampling helpers
`quantlab.features`	Return, momentum, volatility, mean-reversion, cross-sectional and technical features; stationarity/cointegration (`adf_test`, `cointegration_test`, `hurst_exponent`), correlation (`correlation_matrix`) and pairs-trading diagnostics (`compute_pair_diagnostics`); optional `FeaturePipeline`
`quantlab.strategies`	Built-in strategies, registry helpers and `BaseStrategy`
`quantlab.portfolio`	Allocators, constraints, rebalancing and volatility targeting
`quantlab.execution`	Commission, spread, slippage and aggregate execution models
`quantlab.backtesting`	`BacktestEngine`, `BacktestResult`, accounting, benchmark and trade-log helpers
`quantlab.risk`	Performance, drawdown, exposure, VaR/CVaR and stress helpers
`quantlab.validation`	Holdout, walk-forward (`FoldResult`, `WalkForwardResult`, `WalkForwardValidator`), sensitivity, bootstrap and stress validation
`quantlab.reporting`	Tables, charts and HTML-report generation

Recommended entry point¶

For a configuration-driven experiment, use the runner so the same construction path is shared by Python, the CLI and the dashboard:

from quantlab.backtesting import run_backtest_from_config
from quantlab.config import ExperimentConfig
from quantlab.data import DataLoader

config = ExperimentConfig.from_yaml("configs/demo_offline.yaml")
data, quality_report = DataLoader().load(config)
result = run_backtest_from_config(
    data,
    config,
    data_quality_report=quality_report,
)

print(result.summary())

Use BacktestEngine directly when supplying custom strategy, allocator or execution-model instances. Its data argument must be a pandas DataFrame in QuantLab's canonical long OHLCV schema: one row per (timestamp, symbol) and the columns listed in quantlab.constants.OHLCV_COLUMNS. The engine rejects a missing timestamp/symbol axis and any configured tradable symbol absent from the frame before running the strategy. Deliberate engine failures raise BacktestError; deeper schema, strategy and configuration checks use the more specific exceptions in quantlab.exceptions.

WalkForwardValidator.run() has the same canonical-data expectation and requires every symbol in config.data.symbols to be present, even when the history is too short to form a fold. It accepts a caller-provided parameter grid and returns a WalkForwardResult containing public FoldResult records.

Extension points¶

Subclass BaseStrategy, implement generate_signals(), validate constructor parameters, then register the strategy with register_strategy().
Subclass PortfolioAllocator and register it with register_allocator().
Implement SlippageModel for a custom slippage assumption.
Use FeaturePipeline only when an explicit reusable feature transformation pipeline is useful; built-in strategies call feature functions directly.

See Strategies, Backtesting and Validation for the contracts and methodological assumptions.