anyml

AutoML for tabular data — classify, regress, and forecast in 3 lines of code.

anyml is a dead-simple AutoML library for tabular data. Point it at a CSV or a DataFrame and a target column, and it will auto-detect column types, build a preprocessing pipeline, train and compare several models via cross-validation, and return the best one. It uses scikit-learn under the hood for the core models and optionally integrates XGBoost and LightGBM for stronger gradient-boosted baselines.

Built by Viet-Anh Nguyen at NRL.ai.

Why anyml?

• One-liner API — anyml.classify(df, target="y") trains and benchmarks models automatically
• Plugin architecture — Register custom estimators or preprocessing steps
• Local-first — Runs entirely on CPU, no API calls
• Minimal core deps — scikit-learn, pandas, numpy; XGBoost/LightGBM are optional
• Production-ready — Serializable pipelines, prediction intervals, feature importance

Installation

pip install anyml

For optional boosting libraries:

pip install anyml[xgboost]     # XGBoost estimator
pip install anyml[lightgbm]    # LightGBM estimator
pip install anyml[all]         # everything

Python 3.8+ supported (tested on 3.8, 3.9, 3.10, 3.11, 3.12, 3.13)

Quick Start

import anyml
import pandas as pd

df = pd.read_csv("titanic.csv")

# 1. AutoML classification (auto-preprocesses, trains 3-5 models, returns the best)
model = anyml.classify(df, target="Survived")
print(model.best_name, model.best_score)    # e.g. "RandomForestClassifier" 0.83

# 2. Predict on new data
preds = model.predict(df.drop(columns=["Survived"]))

# 3. Regression is the same one-liner
price_model = anyml.regress(pd.read_csv("houses.csv"), target="price")
print(price_model.metrics)                  # rmse, mae, r2

# 4. Time-series forecasting
sales = pd.read_csv("sales.csv", parse_dates=["date"])
forecast = anyml.forecast(sales, time_col="date", target="sales", horizon=30)

Models & Methods

anyml trains and compares multiple models, then picks the best via cross-validation.

Classification models:

• LogisticRegression (sklearn) — fast linear baseline
• RandomForestClassifier (sklearn) — robust tree ensemble
• XGBoostClassifier (optional via [xgboost]) — gradient boosting
• LGBMClassifier (optional via [lightgbm]) — fast gradient boosting

Regression models:

• LinearRegression (sklearn)
• RandomForestRegressor (sklearn)
• XGBoostRegressor (optional)
• LGBMRegressor (optional)

Auto-preprocessing pipeline:

1. Type detection — numeric, categorical, datetime columns auto-identified
2. Missing value imputation — median for numeric, mode for categorical
3. Encoding — OneHotEncoder for low-cardinality categoricals, OrdinalEncoder for high-cardinality
4. Scaling — StandardScaler for numeric features
5. Datetime features — extracts year, month, day, dayofweek, hour from datetime columns

Model selection — Stratified 5-fold cross-validation, picks best by F1 (classification) or R² (regression).

Feature importance — Native (tree models), coefficient-based (linear), or permutation importance fallback.

Time series forecasting — moving average, exponential smoothing, or linear trend (auto-selected by in-sample RMSE).

API Reference

CLI Usage

anyml classify titanic.csv --target Survived --out titanic.pkl
anyml regress houses.csv --target price
anyml forecast sales.csv --time-col date --target sales --horizon 30
anyml predict titanic.pkl new_passengers.csv --out preds.csv

Examples

Classification with feature importance

import anyml
import pandas as pd

df = pd.read_csv("churn.csv")
model = anyml.classify(df, target="churn", cv=5)

print(f"Best model: {model.best_name} ({model.best_score:.3f})")
for feat, imp in model.feature_importance().head(10).items():
    print(f"  {feat}: {imp:.3f}")

model.save("churn.pkl")

Constrain which models are tried

import anyml

# Only try XGBoost and LightGBM (requires anyml[all])
model = anyml.classify(df, target="y", models=["xgboost", "lightgbm"])

Forecast next 90 days

import anyml, pandas as pd

sales = pd.read_csv("sales.csv", parse_dates=["date"])
fc = anyml.forecast(sales, time_col="date", target="revenue", horizon=90)

fc.plot()            # matplotlib plot of history + forecast
fc.to_csv("forecast.csv")

License

MIT (c) Viet-Anh Nguyen