Celebi logo Celebi

Celebi / Guide

Your first workflow

This guide walks through the canonical Celebi example: generate some samples with a Gaussian distribution and fit them (see also demo-basic-01).

0. Start a project

celebi init .            # in an empty directory
celebi use .             # in an existing Celebi project folder
celebi                   # enter the Celebi shell
celebi projects          # list projects;  celebi workon [name]  to switch

1. Construct the workflow

Each workflow node is a folder on disk. Create two algorithms and two tasks, then link them:

create-algorithm AlgGen
create-algorithm AlgFit
create-task Gen
create-task Fit

cd @/Gen    add-algorithm ../AlgGen
cd @/Fit    add-algorithm ../AlgFit
cd @/Fit    add-input ../Gen gen

add-input ../Gen gen makes the Gen task visible inside Fit as a sub-folder named gen — that is the alias system at work.

2. Write the code and configure

In the algorithm, write the command template with a parameter placeholder:

# AlgGen/celebi.yaml
environment: script
commands:
  - root -l code/gen.C('${events}')

Write the actual code with your preferred editor:

cd @/AlgGen
edit-script gen.C

In the task, set the environment and the parameters:

cd @/Gen
set-environment env:root6
add-parameters events 1000

At run time the template is translated into root -l code/gen.C('1000').

3. Develop interactively

Navigate to a task and run:

workaround

You enter a shell inside a created folder: the task’s algorithm is copied into code/, and its dependencies are copied as aliases — it simulates the real running environment, so you can debug exactly what will run.

4. Submit

cd @/Fit
submit                 # or: submit [runner_name]

The runner resolves the dependencies, runs Gen first, then Fit, and the outputs come back through Yuki. Running submit in a directory executes it in all child nodes — one action submits the whole workflow.

5. Everyday operations

  • Copy sub-workflows — cp TasksGroup TaskGroupB duplicates part of the workflow with connections automatically rewired. Very handy for systematic-uncertainty studies.
  • Rename / reorganize — mv freely; the alias system keeps the underlying workflow intact.
  • Document — every object has its own README.md, which never affects results (impressions don’t seal it), so polish it any time.