grubicy 1.2.0

Config-driven helpers for signac workflows with explicit dependencies and migrations

grubicy

grubicy is a small helper library + CLI that layers lightweight dependency management on top of signac.

It is named after Vittore Grubicy de Dragon, an influential promoter of Italian Divisionism. That movement “divided” light and color into strokes; grubicy does the same for workflows: it divides a signac project into stages, connects them with explicit parent -> child links, and keeps those links consistent even as your schema evolves.

With one TOML/YAML spec you can:

• describe multi-action pipelines in a single file,
• materialize signac jobs with parent pointers stored in state points,
• record full parent state points in docs for traceability (deps_meta),
• render row workflows, and
• migrate existing workspaces with cascading pointer updates without doing it by hand.

Why use it

Signac projects are naturally flat, but real computational work is often staged:

• Prepare -> simulate -> analyze
• Preprocess -> train -> evaluate
• Extract -> transform -> aggregate

grubicy helps when you want those stages to be:

• cached and reusable (shared intermediates across experiments),
• explicitly wired (no hidden coupling via shared parameter keys),
• reviewable and reproducible (the pipeline is a spec file),
• maintainable over time (schema changes do not break downstream links).

What you get:

• Explicit dependencies: parent job ids live in the child state point, so “same params but different parents” never collide.
• One spec for everything: job creation, row workflow rendering, and parameter collection are driven by a single config file.
• Safe migrations: plan/apply state point migrations and automatically cascade dependency-pointer rewrites downstream, with progress logging.

When to use it

• Use grubicy if you have multi-step experiments, pass results downstream between stages, or want row-ready workflows without writing manual include filters.
• If your project is truly single-stage, grubicy will feel like extra structure you do not need.

Quick start

1. Install

pip install git+https://github.com/davide-grheco/grubicy

For local development:

uv sync --extra dev

2. Describe your pipeline (pipeline.toml)

[workspace]
value_file = "signac_statepoint.json"

[[actions]]
name = "s1"
sp_keys = ["p1"]
outputs = ["s1/out.json"]

[[actions]]
name = "s2"
sp_keys = ["p2", "test"]
deps = { action = "s1", sp_key = "parent_action" }
outputs = ["s2/out.json"]

[[actions]]
name = "s3"
sp_keys = ["p3"]
deps = { action = "s2", sp_key = "parent_action" }
outputs = ["s3/out.json"]

[[experiment]]
  [experiment.s1]
  p1 = 1
  [experiment.s2]
  p2 = 10
  test = true
  [experiment.s3]
  p3 = 0.1

Notes:

• Each [[actions]] block defines a stage.
• sp_keys lists the parameters that define identity for that stage.
• deps declares which upstream action this stage depends on. The library writes the upstream job id into the dependent job’s state point using sp_key.
• Experiments use per-action subsections: parameters do not need to be shared across stages.

Defining multiple experiments:

• Repeat the [[experiment]] block to create multiple experiment rows. See a complete multi-experiment spec in examples/library-example/pipeline.toml.

3. Materialize jobs and render a row workflow

grubicy prepare pipeline.toml --output workflow.toml

This will:

• create/open signac jobs in topological order,
• write action and dependency pointers (parent job ids) into each state point,
• store deps_meta in job docs (including full parent state points),
• generate workflow.toml for row.

4. Run jobs (only ready directories)

grubicy submit pipeline.toml

If you want to submit everything to row directly, you can still run row submit.

5. Collect downstream-ready parameters

grubicy collect-params pipeline.toml s3 --format csv > results.csv

This flattens the parameter chain for the s3 stage (and optionally selected doc fields), so you can analyze results without manually walking parents.

Core pieces

Spec

• A spec file contains:
  • actions: list of stages with name, sp_keys, optional deps (parent action + sp_key used to store parent job id), optional outputs, optional runner.
  • experiment: list of experiments with per-action subsections.
  • optional workspace.value_file.
• Supported formats: TOML and YAML.

Materialization

• Creates/opens jobs in topological order and wires dependencies by writing parent job ids into the child state point. Also writes deps_meta into child job docs so parent state points are recorded for traceability and repair.

Row rendering

• Builds workflow.toml with per-action include rules, using either your explicit runner or a default python actions/{name}.py {directory}.

Collection

• collect-params flattens parameters (and optional document fields) across the dependency chain for a target stage.

Migration

• Plan/apply state point migrations with collision detection, cascading parent-pointer rewrites downstream, and restartable progress logs under .pipeline_migrations/.
• Useful when you add defaults (setdefault) or evolve the schema and need downstream pointers updated consistently.

Examples

• examples/sample-project: a plain signac setup with hand-wired parent pointers.
• examples/library-example: the same pipeline expressed with grubicy (pipeline.toml, CLI materialization, row workflow, and helper-based actions).

Documentation

• Hosted docs: https://davide-grheco.github.io/grubicy/
• Source docs: docs/getting-started.md (walkthrough), docs/cli.md (CLI reference), docs/migrations.md (worked migration example)

Development

• Install dev deps: uv sync --extra dev
• Install hooks: uv run pre-commit install
• Run hooks on all files: uv run pre-commit run --all-files