Tensorflow Simpsons Morph
Properly Trained Autoencoder

Tensorflow Simpsons Morph

A focus on hyperparameters to build the best model

Click here to try it out!

This experiment uses a convolutional autoencoder for performing principle component analysis on images of The Simpsons faces.

This time, I wanted to focus on hyper-parameter tuning. Instead of guessing hyper-parameters and then overfitting, instead I built methods for iteratively training different parameters, and then analyzing the results on validation data.