Skip to main content

A library for hyperparameter optimization of ML models

Project description

This package offers implementations of several black-box optimisation methods to tune hyperparameters of machine learning models. Its purpose is to enable data scientists to use optimization techniques for rapid protyping. Simply import OptML and supply it with a model and the parameters to optimize.

OptML offers a unified interface for models built with Scikit-Learn, Keras, XGBoost (and hopefully soon Statsmodels).

Prerequisites

This package requires scikit-learn with version 0.19.0 or higher. If scikit-learn is not yet install run pip install scikit-learn==0.19.0. If you want to make use of the HyperoptOptimizer then you also need to install hyperopt (e.g. by pip install hyperopt).

In order to run with Keras and XGBoost models these libraries have to be install as well, of course.

Installation

If scikit-learn is version 0.19 or higher simply install mlopt using pip install optml and you’re ready to go.

Usage

Specify your ML model and the parameters you want to optimize over. For the parameters you have to choose the type (such as integer, categorical, boolean, etc.) and the range of values it can take.

model = SomeMLModel()
params = [Parameter(name='param1', param_type='continuous', lower=0.1, upper=5),
          Parameter(name='param2', param_type='integer', lower=1, upper=5),
          Parameter(name='param3', param_type='categorical', possible_values=['val1','val2','val3'])]

Then define the evaluation function. This can be anything from RMSE to crossentropy to custom functions. The first argument of the evaluation function is the array of true labels and the second argument is an array with model predictions.

def clf_score(y_true,y_pred):
    return np.sum(y_true==y_pred)/float(len(y_true))

Import and initialize an optimizer and optimize the model for some training data.

from optml.bayesian_optimizer import BayesianOptimizer
bayesOpt = BayesianOptimizer(model=model,
                             hyperparams=params,
                             eval_func=clf_score)
bayes_best_params, bayes_best_model = bayesOpt.fit(X_train=X_train, y_train=y_train, n_iters=50)

Features

At the moment this library includes: * Random Search * Parallelized Gridsearch * A simple Genetic Algorithm * Bayesian Optimisation * Hyperopt (using hyperopt)

TODOs

  1. algorithms:
  • more options for genetic algorithms
  • meta heuristics/swarm optimisation (Ant Colony Optimization etc.)
  1. functionality
  • early stopping if there is no significant improvement after x iterations
  1. usability
  • add categorical parameters
  • better documenation

Author

License

This project is licensed under the MIT License - see the LICENSE.md file for details

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for optml, version 0.1.0
Filename, size File type Python version Upload date Hashes
Filename, size optml-0.1.0.tar.gz (9.4 kB) File type Source Python version None Upload date Hashes View

Supported by

AWS AWS Cloud computing Datadog Datadog Monitoring Facebook / Instagram Facebook / Instagram PSF Sponsor Fastly Fastly CDN Google Google Object Storage and Download Analytics Huawei Huawei PSF Sponsor Microsoft Microsoft PSF Sponsor NVIDIA NVIDIA PSF Sponsor Pingdom Pingdom Monitoring Salesforce Salesforce PSF Sponsor Sentry Sentry Error logging StatusPage StatusPage Status page