Notebook: Hyperparameter / Grid Search

[DiogenesAnalytics]

Problem

Would be worthwhile to have a notebook that demonstrates how to find the optimal parameters/architecture for a an autoencoder mode.

References

• https://github.com/CodeReclaimers/neat-python
• https://keras.io/keras_tuner/
• https://keras.io/guides/keras_tuner/
• http://www.cs.ucf.edu/~kstanley/#publications
• Keras Tuner: Layer Type
• https://www.tensorflow.org/tutorials/keras/keras_tuner
• Minimize Max Recon Error

[DiogenesAnalytics]

MinNDAE Tuning

See: https://github.com/DiogenesAnalytics/autoencoder/blob/master/notebooks/tuning/minndae.ipynb

Loss Function

$$loss + encode * reg$$

Plot

[DiogenesAnalytics]

MinNDAE Tuning

See: https://github.com/DiogenesAnalytics/autoencoder/blob/master/notebooks/tuning/minndae.ipynb

Loss Function

$$loss + (loss * encode * reg)$$

Plot

[DiogenesAnalytics]