Auto-encoder trained with MNIST Data Set in order to compress data set before Classification with Clustering Method. Final purpose is to explain why some data are misclassified.
This repository contain some scripts to build Auto-encoder using Keras. ANN Models differ by activation function and number of output variables, and models are evaluated on several point : Distribution of values, Entropy (2-base), Loss values. The Auto-encoder is trained on MNIST Data base.
In order to use it, make sure Keras, ggplot2, reshape2 and dplyr packages are installed