Skip to content

Latest commit

 

History

History
103 lines (75 loc) · 4.06 KB

RELEASE.md

File metadata and controls

103 lines (75 loc) · 4.06 KB
Raw

Release 0.2.3

Bug Fixes and Other Changes

  • Fixed nevopy.callbacks.BestGenomeCheckpoint not working properly with negative fitness values.
  • Fixed nevopy.neat.visualization.columns_graph_layout wrongfully positioning hidden nodes when the network had only one column of hidden nodes.

Breaking Changes

  • Removed the file_prefix parameter from the nevopy.callbacks.BestGenomeCheckpoint.__init__ method.
  • Changed the default values of the parameters out_path and min_improvement_pc of the nevopy.callbacks.BestGenomeCheckpoint.__init__ method.

Release 0.2.2

Bug Fixes and Other Changes

  • Added the gym package to the project's dependencies (version 0.17.3).
  • Fixed missing docs on RTD due to dependencies issues.

NOTE: the version 0.2.1 was released as a quick-fix for PyPI only.

Release 0.2.0

Major Features and Improvements

  • Added a new function (nevopy.neat.visualization.visualize_activations) to visualize the neural topology of NEAT genomes (nevopy.neat.NeatGenome) while they interact with an environment. It's highly customizable and allows, among other things, activated nodes and edges to be drawn with different colors. Check the lunar lander example for a demonstration.
  • Added new utilities to be used with gym environments, including callbacks and custom renderers. They're all contained within the new subpackage nevopy.utils.gym_utils.
  • Added the bipedal walker example.
  • Improved the lunar lander example. Now it shows the neural topology of the evolved genome side by side with the rendering of the environment.

Breaking Changes

  • Removed nevopy.fixed_topology.FixedTopologyPopulation (deprecated since v0.1.0). The nevopy.genetic_algorithm.GeneticPopulation class should be used instead.
  • Removed nevopy.fixed_topology.FixedTopologyConfig (deprecated since v0.1.0). The nevopy.genetic_algorithm.GeneticAlgorithmConfig class should be used instead.
  • Removed the Mario example.
  • Replaced nevopy.utils.GymEnvFitness with nevopy.utils.GymFitnessFunction, which has more features and supports more advanced callbacks.

Release 0.1.1

Bug Fixes and Other Changes

  • Fixed a bug in ne.utils.GymEnvFitness that led to an incorrect interpretation of the agent's chosen action when dealing with TensorFlow tensors.
  • Updated ne.fixed_topology.FixedTopologyGenome.visualize. It now returns the generated PIL.Image.Image object. It's possible now to directly use the method to visualize the genome's topology on a Jupyter Notebook.
  • Added a new XOR example (Jupyter Notebook).
  • Added a wrapper to tf.keras.layers.MaxPool2D in ne.fixed_topology.layers.
  • Fixed a bug that occurred when a string was passed to the layer_type parameter of the constructor of the ne.fixed_topology.TensorFlowLayer class.

Release 0.1.0

Breaking Changes

  • ne.fixed_topology.FixedTopologyPopulation has been deprecated. The new class ne.genetic_algorithm.GeneticPopulation should be used in its place.
  • ne.fixed_topology.FixedTopologyConfig has been deprecated. The new class ne.genetic_algorithm.GeneticAlgorithmConfig should used in its place.

Bug Fixes and Other Changes

  • Added new type of population: ne.genetic_algorithm.GeneticPopulation. It implements a generalizable genetic algorithm that can be used as a base for a wide range of neuroevolutionary algorithms. This resolves #1.
  • Added deprecation.py to utils. It implements the @deprecated decorator, that can be used to mark a function, method or class as being deprecated.
  • Fixed a bug in GymEnvFitness that was causing an incorrect interpretation of the output values of fixed-topology genomes using TensorFlow layers.
  • Made some fixes and additions to the project's docstrings.
  • Added a new example in which NEvoPy is used to create an AI to play the Flappy Bird game.
  • The other examples were reformatted and comments explaining the code were added to them.

Release 0.0.2

Initial public release of the NEvoPy framework.