This release:

• Adds "fast mimic" option to MIMIC algorithm. This allows the algorithm to compute the mutual information in vectorized form, which results in faster speed but requires more memory.
• Corrects an error in the Six Peaks fitness function.
• Updates the fitness curve function in the Hill Climb algorithm to update each time a better state is found, rather than once per restart.

This release:

• Modifies the NeuralNetwork classes to allow the use of these classes in sklearn pipelines and GridSearchCV;
• Fixes several bugs in the algorithms which resulted in the wrong function being used to retrieve fitness values for use in the fitness curves.

This release:

• Adds a parameter to the algorithms to save the fitness scores;
• Adds a random state parameter to set the random seed for the algorithms and optimization objects;
• Adds a restart option to NeuralNetwork();
• Fixes a bug in the genetic algorithm which caused a crash when length = 1;
• Fixes a bug in the softmax function to prevent overflow.

This release corrects a bug in the MIMIC algorithm which can sometimes give rise to an infinite loop.

First stable version of mlrose, released in January 2019.