deephyper.nas#

This sub-package his dedicated to the definition of neural architecture search space and evaluation strategy. The implementation is using Tensorflow 2.X and Keras API. The main concepts are: * deephyper.nas.KSearchSpace: An object to define a search space of neural architectures. * deephyper.nas.run: A sub-package to define the evaluation strategy of a neural architecture (e.g., training procedure). * deephyper.nas.operation: A sub-package to define operations of the neural architecture search space. * deephyper.nas.node: A sub-package to define nodes of the neural architecture search space which is represented as a direct acyclic graph.

Classes

KSearchSpace

A KSearchSpace represents a search space of neural networks.

NxSearchSpace

A NxSearchSpace is an search_space based on a networkx graph.

deephyper.nas.losses

This module provides different loss functions.

deephyper.nas.lr_scheduler

deephyper.nas.metrics

This module provides different metric functions.

deephyper.nas.node

This module provides the available node types to build a KSearchSpace.

deephyper.nas.operation

Operations for neural architecture search space definition.

deephyper.nas.preprocessing

deephyper.nas.run

The deephyper.nas.run sub-package provides a set of functions which can evaluates configurations generated by search algorithms of DeepHyper.

deephyper.nas.spacelib

Library of neural architecture search spaces.

deephyper.nas.trainer