CLI Reference

After installing epftoolbox2, an epf command is available on your PATH.

pip install epftoolbox2
epf --help

---

Interactive Mode

Running epf with no arguments launches a wizard-style interactive terminal where you can build and run pipelines without writing any YAML.

epf

You’ll see a menu with arrow-key navigation:

  Run experiment from YAML
  Run data pipeline from YAML → CSV
  Run model pipeline from YAML
  ──────────────────────────────────
  Build & run data pipeline interactively
  Build & run model pipeline interactively
  ──────────────────────────────────
  Exit

Build & run data pipeline interactively walks you through:

1. Adding sources (EntsoeSource, OpenMeteoSource, CsvSource, CalendarSource, EpftoolboxBenchmarkSource) with prompted config
2. Adding transformers (optional)
3. Adding validators (optional)
4. Setting date range and caching
5. Choosing an output CSV path
6. Saving a reusable data_pipeline.yaml
7. Running immediately

Build & run model pipeline interactively walks you through:

1. Loading an input CSV (from a previous data run, with a prompt for the time zone if the file doesn’t name it)
2. Adding models: available columns are shown as a checkbox list for predictor selection
3. Adding evaluators and exporters
4. Setting the test period, target column, horizon and resolution (1h or 15min)
5. Saving a reusable model_pipeline.yaml
6. Running immediately

Tip

Both wizards offer to save a YAML at the end. Run that YAML later with epf run or epf data / epf model without going through the wizard again.

---

epf run: Full workflow

Run a complete experiment from a Workflow YAML.

epf run experiment.yaml

Options:

Flag	Description
--model-index N	Run only models[N] (array mode: needs model_pipeline.save_dir, see epf report)
--processes N	Override max_processes
--threads N	Override threads_per_process
--dry-run	Build both pipelines without executing, validates config

Examples:

# Basic run
epf run experiment.yaml

epf run experiment.yaml --model-index 2

# Validate config without running
epf run experiment.yaml --dry-run

---

epf report: Merge array results

After a SLURM array in which each task ran one model (--model-index), merge every model’s JSONL from model_pipeline.save_dir and run the evaluators and exporters once. Prints planned vs produced forecasts per model.

epf report experiment.yaml
epf report experiment.yaml --allow-missing   # report the models that have results

Without --allow-missing it fails if any model has no results file.

---

epf data: Data pipeline only

Run a DataPipeline YAML and save the result as a CSV.

epf data data_pipeline.yaml --start 2022-01-01 --end 2024-01-01 --output data.csv

The CSV keeps the local time zone: the index header names it (e.g. datetime[Europe/Warsaw]), and epf model and the interactive mode read the data back in that zone, so days and hours match an in-process run.

Required flags:

Flag	Description
--start DATE	Fetch start date (YYYY-MM-DD)
--end DATE	Fetch end date (YYYY-MM-DD)
--output FILE	Output CSV path

Optional flags:

Flag	Description
--cache PATH	Cache directory or CSV file for source data

Example:

epf data data_pipeline.yaml \
    --start 2020-01-01 \
    --end 2024-01-01 \
    --output data.csv \
    --cache .cache/

---

epf model: Model pipeline only

Run a ModelPipeline YAML on an existing CSV dataset (e.g. produced by epf data). The model’s days and hours are local, so the file must say which time zone it is in: files from epf data do. For other files with UTC offsets in the timestamps, pass --tz.

epf model model_pipeline.yaml \
    --data data.csv \
    --test-start 2023-01-01 \
    --test-end 2024-01-01

Required flags:

Flag	Description
--data FILE	Input CSV file
--test-start DATE	Test period start
--test-end DATE	Test period end

Optional flags:

Flag	Description
--target COL	Target column name (default: price)
--horizon N	Forecast horizon in days (default: 7)
--freq {1h,15min}	Data resolution: one model per hour or per quarter-hour (default: 1h). Must match the data
--tz ZONE	Local time zone of the data (e.g. Europe/Warsaw). Overrides the zone named in the file
--issue-time HH:MM	Check every predictor against this issue time before fitting (information set). Overrides the YAML’s information_set, or adds one if it has none
--issue-timezone ZONE	Zone of --issue-time (default: the YAML’s, else Europe/Brussels). Without --issue-time, the YAML’s issue time is kept, or 11:45 if the YAML has no information_set
--save-dir DIR	Save per-model prediction JSONL files to this directory
--forecast-only	Skip evaluators and exporters
--allow-manifest-mismatch	Resume even if --save-dir’s manifest disagrees with this run’s configuration (overwrites it)
--require-manifest	Refuse to resume from a --save-dir that has results but no manifest, instead of warning

---

epf validate: Validate any pipeline YAML

Load and parse a YAML file, instantiate all components, and report errors. Without running anything.

epf validate data_pipeline.yaml
epf validate model_pipeline.yaml
epf validate experiment.yaml

The file type is auto-detected from its top-level keys. Exits with code 1 if validation fails.

---

epf init: Scaffold template files

Create experiment.yaml, data_pipeline.yaml, and model_pipeline.yaml with placeholder values in the current directory.

epf init
epf init --force   # overwrite existing files