modep-client

Python client for the modep API. All open-source tabular AutoML frameworks through one unified REST API.

Installation:

pip install modep-client

Available frameworks:

• AutoGluon: Amazons’s version of AutoML with lots of stacking
• auto-sklearn: automatic sklearn pipelines from NIPS 2015
• auto-sklearn 2: new version for 2020 that added portfolios
• Auto-WEKA: AutoML in a JAR if you’re into that
• FLAML: one of Microsoft’s version of AutoML (Fast and Lightweight AutoML)
• GAMA: AutoML project from OpenML, from the authors of the AutoML benchmarking library used here
• H2O AutoML: free version of H2O.ai AutoML
• hyperopt-sklearn: uses hyperopt to search sklearn pipelines
• mljar-supervised: has several modes depending on how aggressively you want to search
• MLNet: command line AutoML by Microsoft
• TPOT: optimizes sklearn pipelines using genetic programming, gives you back the code of the best pipeline

In addition, the following non-AutoML baseline frameworks are available for comparison:

• Constant Predictor: predicts empirical target class probabilities for classification or the target median for regression
• Decision Tree: sklearn Decision Tree with default parameters
• Random Forest: sklearn Random Forest with default parameters except n_estimators = 2000
• Tuned Random Forest: above with tuned max_features parameter

For all frameworks you can:

• Get predictions on new data
• Download anything saved during training like leaderboards, figures, and logs

Helpful links

• For the latest documentation: https://modep-ai.github.io/modep-client
• For the API’s Swagger docs: https://modep.ai/v1/api-docs
• To add new frameworks, make PRs to: https://github.com/openml/automlbenchmark