Gradient Streaming Mixture Model

Final model for dissertation's results.

Gradient Streaming Mixture Model

• fixed bug in weightWrapper initialization
• changed default parameters for accelerated methods

Gradient Streaming Mixture Model

• Added ADAMAX gradient ascent
• improved method's names for ConjugatePrior
• simplified factory object for main model
• added nTries as a parameter for K-Means initialization to allow selecting the best model from many tries
• corrected errors in methods' description

Gradient Streaming Mixture Model

• fixed formula error in ADAM
• fixed formula error in conjugate prior
• fixed numerical stability bugs in weight mappers
• added .halveStepEvery() method for a more intuitive way of initializing decay rate
• added random seed setter for mini-batch selection

Gradient Streaming Mixture Model

• collapsed streaming and static models into a single one
• cached data for better performance
• improved trajectory logs for debugging

Gradient-based gaussian mixture with streaming version

v1.0

gradient-based gaussian mixture model with streaming version