My codes that I used for the ADIA Lab competition
Some of the notebooks that they provided are located here: https://github.com/crunchdao/adialab-notebooks/. These were used to generate ideas of things to test.
I've uploaded three notebooks: Solutions 000 and 001 are both feed forward deep neural networks with different parameters, activation functions, etc. Solution 002 implements feature selection with the boruta method. The results I received on the training data were not promising with this. Finally, the autoencoder model is a customized model that uses the loss from an autoencoder as well as the loss from standard multi layer perceptron to inform back propagation.
For my submissions I went with solution 000.