Eigrel

A programming language for Data, Machine Learning and AI.

Write what you want. Let the compiler decide how to run it.

Eigrel is a declarative language for data engineering, machine learning and GenAI. You describe datasets, transformations, features and models in one language; the compiler decides whether each step becomes Python, SQL, Spark or something else.

dataset customers from csv("data/customers.csv")

transform customers {
    filter age >= 18
    select age, income, purchases, churned
}

features customers {
    age
    income
    purchases
}

model churn = random_forest {
    trees = 100
}

train churn {
    target = churned
}

evaluate churn {
    metrics = [accuracy, precision, recall, f1]
}

Status

Eigrel is at v0.2 — Compiler: programs are checked for meaning, lowered into an intermediate representation and compiled to Python (pandas + scikit-learn), so eigrel run trains and evaluates real models. Next up: more data sources (v0.3). See the roadmap.

Getting started

Install from PyPI (Python 3.13+). The python extra adds pandas and scikit-learn, which eigrel run needs:

pip install "eigrel[python]"
eigrel init churn             # creates churn/main.eig and sample data
eigrel run churn/main.eig

churn: random_forest classification, trained on 315 rows, validated on 79
  accuracy   0.7975
  precision  0.7442
  recall     0.8649
  f1         0.8000

CLI

The compiler catches mistakes before anything runs, and points at the exact spot:

error: column 'income' does not exist here; available columns: age, purchases, churned
 --> churn.eig:9:5
  |
9 |     income
  |     ^

How it works

source → lexer → parser → AST → semantic analysis → IR → Python backend → pandas + scikit-learn

eigrel ir shows the graph the backends work from:

%0 = load csv("data/customers.csv")  # customers
%1 = filter %0 (age >= 18)  # customers
%2 = select %1 [age, income, purchases, churned]  # customers
%3 = train %2 random_forest(trees=100, max_depth=8) classification features=[age, income, purchases] target=churned validation=0.2 seed=42  # churn
%4 = evaluate %3 [accuracy, precision, recall, f1]  # churn

Docker

make docker-build
make docker-run      # runs examples/ml.eig inside the container

Language

The full syntax is in docs/LANGUAGE.md. Examples live in examples/.

Contributing

Bug reports, language proposals and pull requests are welcome. Read the contributing guide to get set up, and note that this project follows a Code of Conduct. Security issues go through SECURITY.md.

License

Licensed under the Apache License 2.0.