hippo 0.1.0b2

Hippo is a lightweight machine-learning framework that unifies declarative preprocessing pipelines, automatic hyperparameter optimisation, and local experiment tracking behind a single, friendly CLI.

hippo

Lightweight ML workflow engine: declarative preprocessing ▸ hyper-parameter optimisation ▸ experiment tracking

hippo brings together three everyday ML chores under one friendly roof:

What you need	How hippo helps
Clean, consistent data	Build declarative preprocessing pipelines with Sklearn primitives
Good hyper-parameters without the grind	Run Optuna-powered Bayesian optimisation (TPE, CMA-ES, …)
Remember what actually worked	Log parameters & metrics to a local SQLite DB with zero setup

Why hippo?

• ⚡ Efficient – sensible defaults, parallel Optuna trials
• 🛡️ Reliable – minimal dependencies, 100 % typed codebase
• 🔌 Extensible – register your own models or transformations in a single line
• 🐚 Simple CLI – hippo train config.yml is all you need

Installation

pip install hippo

Quick Start

from sklearn.datasets import load_breast_cancer
from hippo.data.preprocessing import build_pipeline
from hippo.models import get_model
from hippo.tuning import optimise
from optuna import Trial

# 1  Load toy data
data = load_breast_cancer(as_frame=True)
X, y = data.data, data.target

# 2  Preprocessing pipeline
pipe = build_pipeline(num_feats=X.columns.tolist(), cat_feats=[])

# 3  Objective builder for Optuna
def make_objective(trial: Trial):
    n_estimators = trial.suggest_int("n_estimators", 50, 400)
    max_depth = trial.suggest_int("max_depth", 3, 10)

    model = get_model("rf", n_estimators=n_estimators, max_depth=max_depth, random_state=0)

    def objective(trial_: Trial):
        from sklearn.model_selection import cross_val_score
        score = cross_val_score(model, pipe.fit_transform(X), y, cv=3).mean()
        return score
    return objective

study = optimise(make_objective, n_trials=50, direction="maximize")
print("Best accuracy:", study.best_value)

Or the same via CLI:

python -m hippo.cli train config.yml

config.yml example:

run_name: demo_rf
data:
  path: cancer.joblib         # joblib-dumped DataFrame
  target: target
  numerical: [mean radius, mean area, ...]
  categorical: []
model:
  name: rf
  params:
    n_estimators: 200