Bayesian-Optimization-for-hyper-parameter-tuning

Spearmint integrated Bayesian Optimization for hyper parameter tuning of Auto sparse encoder embedded with Softmax Classifier for MNIST digit Classification.

Instructions to run

1. Download and install Spearmint package (instructions are on 'https://github.com/JasperSnoek/spearmint').
2. Download the MNIST dataset (from 'http://yann.lecun.com/exdb/mnist/') in the same folder with the rest of Matlab files.
3. Run the spearmint optimization module.

• Implementation of Classification module is in Matlab.
• STL_opt is the matlab wrapper required for spearmint package.
• config.json is the configuration file with specifications as per the spearmint instruction.
• Bayesian Optimization used is minimizing the classification error.
• L-BFGS algorithm is used to minimize the cost function for weights training in Softmax Classifier and Sparse Auto-encoder.