v 0.1.0, July 19, 2019 -- Initial development release.
v 0.2.0, July 29, 2019 -- Changed syntax for accessing free parameter values, setting free parameters and setting which parameters are trainable. Free parameters values are now accessed using dot syntax and the trainability of a parameter is set using the new method `trainable` (`freeze` and `thaw` have been removed). The valid fit modes are now `cold` and `hot`. Cold uses a random initialization for trainable parameters while hot initializes trainable parameters at their current values.
v 0.2.1, July 29, 2019 -- Patch for evaluation method. Will now return inf instead of nan.
v 0.2.2, August 5, 2019 -- Patch for loading old models. Changed deprecated variable use in active selection generator. Reworked preprocess module. Reworked visualize module to more closely align with Matplotlib conventions. Added timing information to verbose output for psiz.models.fit, psiz.models.posterior_samples and psiz.generators.generate methods.
v 0.3.0, April 29, 2020 -- Updated to TensorFlow 2.0 idioms. Added session ID attribute to psiz.trials.Observations object. Removed log_likelihood method from PsychologicalEmbedding API since it can easily be derived from outcome_probability. Deprecated various load_x functions in favor of module-scoped load functions. Created new dimensionality.search function that uses a better encapsulation of arguments. Setup file now installs tensorflow package since this package now supports cpu only and gpu use.