Pipestat is a Python package for managing pipeline results. It provides a standard API for reporting, storing, and retrieving outputs from any computational pipeline. Results are validated against a JSON Schema and stored in either a YAML file or a PostgreSQL database. A pipeline author defines outputs in a schema, then uses pipestat to report results as the pipeline runs. Downstream tools retrieve those results through the same API.

See Pipestat documentation for complete details.

Quick Start

import pipestat

psm = pipestat.PipestatManager(
    schema_path="output_schema.yaml",
    results_file_path="results.yaml",
    record_identifier="sample1",
)
psm.report(values={"accuracy": 0.95, "processing_time": 12.3})
psm.report(values={"output_file": {"path": "results/output.csv", "title": "Output CSV"}})

Schema

Pipestat requires a schema that defines the results your pipeline reports. Here is a minimal example:

pipeline_name: my_pipeline
samples:
  accuracy:
    type: number
    description: "Model accuracy score"
  processing_time:
    type: number
    description: "Processing time in seconds"
  output_file:
    type: object
    object_type: file
    properties:
      path: {type: string}
      title: {type: string}
    required: ["path", "title"]

Every result needs type and description. See sample_output_schema_generic.yaml in this repo for a fuller example with both sample and project result types.

Generating a schema from existing results

If you already have a pipestat results file, you can auto-generate a schema:

pipestat infer-schema -f results.yaml -o schema.yaml

This inspects your results and produces a schema with the correct types. See pipestat infer-schema --help for options.

Developer tests

Optional Dependencies

Note: to run the pytest suite locally, you will need to install the related requirements:

cd pipestat

pip install -r requirements/requirements-test.txt

Database Backend Configuration for Tests

Many of the tests require a postgres database to be set up otherwise many of the tests will skip.

We recommend using docker:

docker run --rm -it --name pipestat_test_db \
    -e POSTGRES_USER=postgres \
    -e POSTGRES_PASSWORD=pipestat-password \
    -e POSTGRES_DB=pipestat-test \
    -p 5432:5432 \
    postgres