Maggy: Asynchronous distributed hyperparameter optimization based on Apache Spark Asynchronous algo…

Maggy, an open-source framework for asynchronous distributed hyperparameter optimization, automates model experimentation and data parallelism.

Key takeaways

Maggy is an asynchronous distributed hyperparameter optimization framework based on Apache Spark.
It allows users to define a search space and a model, then train multiple trials in parallel and aggregate the results.
Maggy can be used for ablation studies, where components of a model are left out to see their contribution to the final outcome.
The framework uses a search loop and an inner loop, where the search loop generates trials and the inner loop trains the models.
Maggy works by registering tasks with a driver, which schedules and monitors the tasks, and each task trains a model and returns a metric.
The framework can be used with multiple optimizers, including random search, Bayesian optimization, and ASHA (Asynchronous Successive Halving).
Maggy provides a flexible API for users to define their own optimizers and models.
The framework is designed for distributed training and can scale out to multiple machines.
Maggy provides features such as early stopping, batch sizes, and data parallelism.
It also provides support for GPUs and CPUs.
The framework is designed for big data and can handle large amounts of data.
Maggy is open-source and can be used with Hopsworks, a cloud-based platform for machine learning.
The framework is designed to work with single-host Python environments and can be used locally or on a cloud-based platform.
Maggy provides a convenient interface for users to define their models, optimizers, and hyperparameters, and then run their experiments.
The framework is designed for ease of use and provides features such as automatic hyperparameter tuning, batch sizes, and data parallelism.
Maggy uses a distributed approach to optimize hyperparameters and can handle large amounts of data.
It also provides features such as asynchronous optimization, Bayesian optimization, and ASHA (Asynchronous Successive Halving).
The framework is designed for big data and can handle large amounts of data.
Maggy is a flexible and scalable framework for hyperparameter optimization and can be used for a variety of machine learning tasks.
The framework is designed for ease of use and provides features such as automatic hyperparameter tuning, batch sizes, and data parallelism.
Maggy uses a distributed approach to optimize hyperparameters and can handle large amounts of data.

Maggy: Asynchronous distributed hyperparameter optimization based on Apache Spark Asynchronous algo…

More talks