AutoGluon

1.5.0 · active · verified Sun Apr 12

AutoGluon is an open-source AutoML library developed by AWS AI, designed to automate machine learning tasks for tabular, image, text, and time series data. It enables users to train and deploy high-accuracy machine learning and deep learning models with minimal code, often lauded for its '3 lines of code' capability. The library is actively maintained with frequent updates and is currently at version 1.5.0, offering state-of-the-art predictive performance.

Warnings

breaking Models trained with older versions of AutoGluon are not compatible with newer versions. Users must retrain models after upgrading AutoGluon to a new major or minor release.
Fix: Re-train all AutoGluon models using the new version of the library after upgrading.
deprecated Several `TabularPredictor` methods were deprecated in v1.0.0 and subsequently removed in v1.3.0. Using these old method names will result in errors.
Fix: Update method calls to their new names (e.g., `persist_models` -> `persist`, `get_model_names` -> `model_names`). Refer to the AutoGluon documentation for the full list of changes.
breaking AutoGluon regularly updates its supported Python versions. For instance, Python 3.8 support was dropped in v1.2.0, and newer versions (e.g., v1.5.0) require Python 3.10-3.13. Running AutoGluon on an unsupported Python version will lead to installation or runtime failures.
Fix: Ensure your Python environment matches the requirements for your AutoGluon version. For v1.5.0, use Python 3.10, 3.11, 3.12, or 3.13.
gotcha The `autogluon-common` PyPI package provides helper functionality and is not intended for standalone use to access the full AutoML features. Users seeking the complete AutoGluon experience (e.g., `TabularPredictor`, `TimeSeriesPredictor`) should install the `autogluon` meta-package.
Fix: Install `autogluon` via `pip install autogluon` to get all necessary modules and functionality.
deprecated The `autogluon.eda` module, which provided exploratory data analysis functionality, has been deprecated.
Fix: While no direct replacement in AutoGluon is provided, users should consider external EDA libraries or custom scripts for exploratory data analysis.

Install

pip install autogluon Recommended installation for full AutoGluon functionality
pip install autogluon-common Install only common helper utilities

Imports

TabularPredictor
```
from autogluon.tabular import TabularPredictor
```
Primary class for tabular data prediction tasks.
TabularDataset
```
from autogluon.tabular import TabularDataset
```
Utility for loading tabular data, compatible with pandas DataFrames.
TimeSeriesPredictor
```
from autogluon.timeseries import TimeSeriesPredictor
```
Class for time series forecasting tasks.
MultiModalPredictor
```
from autogluon.multimodal import MultiModalPredictor
```
Class for tasks involving multiple data modalities (image, text, tabular).

Quickstart

This quickstart demonstrates how to use AutoGluon's TabularPredictor to train a model on tabular data and make predictions. It showcases the common workflow for classification or regression tasks with minimal code.

import pandas as pd
from autogluon.tabular import TabularPredictor, TabularDataset

# Create dummy training data
train_data = pd.DataFrame({
    'feature1': [1, 2, 3, 4, 5],
    'feature2': ['A', 'B', 'A', 'C', 'B'],
    'target': [0, 1, 0, 1, 0]
})
# Save to a CSV for TabularDataset
train_data.to_csv('train.csv', index=False)

# Create dummy test data
test_data = pd.DataFrame({
    'feature1': [6, 7],
    'feature2': ['C', 'A']
})
test_data.to_csv('test.csv', index=False)

# Load data using AutoGluon's TabularDataset
train_dataset = TabularDataset('train.csv')
# For demonstration, label is 'target'
label = 'target'

# Initialize and train the TabularPredictor
predictor = TabularPredictor(label=label, path='AutoGluonModels').fit(train_dataset, presets='medium_quality')

# Make predictions on new data
test_dataset = TabularDataset('test.csv')
predictions = predictor.predict(test_dataset)

print("Predictions:\n", predictions)
# Clean up generated files (optional)
import shutil
shutil.rmtree('AutoGluonModels', ignore_errors=True)
import os
os.remove('train.csv')
os.remove('test.csv')

view raw JSON →