Experiments

Experiments from the initial research plan

CNN Activations

These are the activations from all CNN layers in experiment 4 for highPrice predictions in the paper.

Trained model weights

The trained model weights, model architectures, training histories and performance visualizations will be available here shortly.

Additional experiments

Our network models include hundreds of parameters that have to be individually set to achieve an increased prediction performance.

We are currently developing a genetic framework that aims to automate neural architecture generation and hyperparameter optimization. Results of experiments conducted with this framework will be published here when available.