A quite disorganised repo containing some useful code for hyperparameter search/tuning.
really simple code for a random search.
Code implementing a quick and dirty genetic algorithm for hyperparameter search Uses the Tox21 dataset from deepchem. For relevant packages, should run in a virtual environment as per deepchem instructions for 'Easy install with Conda'. NN model code for the Deepchem dataset is from Deep Learning with Tensorflow.
It's a rough proof of concept, as it is there's a lot of code redundancy for each of the different hyperparameters, which is slow and a bit error-prone. To make it more concise and generally useful it needs an object-oriented reimplementation where each hyperparameter gets its own class, and hyperparams is a dictionary containing all hyperparameter classes.
Would also probably be better (and more pythonic) if implemented as an iterable, so you could just compute another generation by calling .next()