12/8/15
Most of my analysis was done in python. Please look at the code in the following order:
- Autoencoder.ipynb. In this ipython notebook, I load the data, and then I train various autoencoders using Keras. I also tested another python package, Theanet, but it didn't give me as much control as I wanted. After training an autoencoder, I would save it and transfer it to a server where I can perform heavier computing.
- Method_pipeline.m. This MATLAB file loads an encoded data and the expression matrix. Its runs ADMM (jz_ADMM.m), which at some points uses a shrinkage operator (jz_shrink.m), solving Eq. 5 in the paper. This code outputs a series of U's for various lambdas.
- Analysis.ipynb. In this ipython notebook, I use various functions to perform kmeans clustering and visualization (PCA) on the matrices generated by Method_pipeline.m.
Please see the comments in the code for further detail.