Configuration Guide
DQM-ML pipelines are configured using YAML files. This guide explains how to write configuration files that tell DQM-ML where to find data, which metrics to compute, and where to save results.
Configuration Structure Overview
flowchart TB
subgraph config["config.yaml"]
subgraph feat["features"]
F1["outputs + processors"]
end
subgraph met["metrics"]
M1["outputs + processors"]
end
subgraph g["gap"]
G1["outputs + processors"]
end
subgraph dl["dataloaders"]
DL1["loaders: [parquet, csv, ...]"]
end
end
DL1 --> F1
DL1 --> M1
DL1 --> G1
Configuration has 4 main sections, documented here in order of concern:
| # | Section | What it does | Who configures it |
|---|---|---|---|
| 1 | features, metrics, gap |
Interfaces — what to compute | Data quality team, data scientists |
| 2 | processors (under each interface) |
Processors — how to configure computation | Data scientists |
| 3 | dataloaders |
Data Selection — where data comes from | Data engineers |
| 4 | storage, compute, errors (global) |
Global Configuration — execution environment | DevOps |
The pipeline execution order is different from the authoring order: data loaders feed batches into each interface's processors. The sections below are ordered by what you need to decide first when writing a config.
- Interfaces — Interface configuration and outputs
- Processors — Common processor structure and columns
- Features — Features processors
- Metrics — Metrics processors
- Gap — Gap processors
- Data Selection — Detailed data source configuration
- Global Configuration — Storage, compute, and error settings
Tip: For a complete working example with every section populated, see
examples/config/config_v2.yaml.
Schema Validation
The configuration follows a JSON Schema defined at docs/schema/config.json. Use it in your IDE for autocompletion and validation:
Quick Links
- CLI Reference — Command-line interface
- YAML Basics — Quick reference for YAML syntax
- Formal and Core Concepts for definitions of Sample, Metric, Data Selection, Batch, and related terminology.