Skip to main content

QuantFlow Research

Batch execution path for feature validation and modeling

Where It Fits

DataInfra → FeatureDAG → Research (Batch Execution)

Overview

Research is the batch execution layer of QuantFlow.

It runs FeatureDAG pipelines on historical data to generate features, labels, and datasets for modeling.

Research does not define features — it evaluates them.

Key Features

  • Batch execution over large historical datasets
  • Numba-fused kernels — multi-computation in single pass
  • Triple barrier labeling for supervised learning
  • Dataset generation for ML pipelines
  • Reproducible research workflows

Workflow

  1. Data Ingestion — Load datasets through DataInfra CDM
  2. Feature Definition — Use FeatureDAG to declare computation
  3. DAG Compilation — Build optimized execution graph
  4. Batch Execution — Compute features across full dataset history
  5. Label Generation — Create supervised learning targets
  6. ML Export — Output features and labels as ML-ready datasets

Scale

  • Designed for large-scale historical datasets
  • Efficient columnar execution via Polars
  • Single-pass computation using fused Numba kernels
Research validates what Trading executes.
← Back to Home