hyperopt / hyperopt-nnet

Hyperparameter optimization for neural networks

hyperopt-nnet

This package provides a hyperopt-compatible neural network implementation.

Currently, it can be used to tune neural network hyperparameters for data sets provided as skdata protocols.

See the ./examples subdirectory for sample training scripts (e.g. nips2011_nnet.py) and a plotting script (plot_trials.py).

The hpnnet.nips2011 file implements the search parameterization used in Bergstra, Bardenet, Bengio, and Kegl ("Algorithms for Hyper-parameter Optimization") from NIPS 2011.

Dependencies

• NumPy
• Sklearn
• Theano
• Skdata (github master, not PyPI)
• Hyperopt (github master, not PyPI)
• matplotlib (for plotting)
• IPython (for parallel search, option 1)
• MongoDB (for parallel search, option 2)