This is my implementation of a stacked regressor using optimized SVM and random Forest using Optuna.The actual inputs of the combined regressor is a latent representation of 59 numerical inputs compressed into 5 ,extracted using an auto-encoder implemented under Keras
My goal was to focus more on the model and the fine tuning of the hyper-parameters ,instead of the data itself and all the visualisation/preprocessing behind .
I actually scored 0.15,(top 20%) with a very simple auto-encoder architecture and a lazy data preprocessing .
With a deeper one ,the result would probably be better. (DM me if you achieve better with a more complex architecture).
-
pip install -r requirements.txt && mkdir sample
-
python3 preprocessing.py
-
python3 AE_train.py
-
python3 main.py