In this repository I present my solution for the ODSC 2020 hackathon https://odsc.com/hackathon/
The final solution notebook can be found here.
The pairplot is available here.
The correlation heatmap is available here.
In this challenge the competitors need propose a model that predicts the temperature of 4 components of an electric car motor. There is a high correlation between the temperatures of the 4 components. Therefore, at first I trained a fully connected network on only one temperature target variable. I took the predictions from my first model and used it as a feature to the other 3 components models.
In other words:
first stage:
y1=f(X)+error (best model = stator_yoke)
y2=h(x)+error
y3=g(x)+error
y4=j(x)+error
second stage:
y2=h(x + f(x))+error
y3=g(x + f(x))+error
y4=j(x + f(x))+error
The final RMSE results on the test set are:
RMSE_pm = 1.011369259
RMSE_stator_yoke = 0.192175924
RMSE_stator_tooth = 0.36982214
RMSE_stator_winding = 0.470236891
adding up to 2.043604214 total RMSE
A video explaining my solution can be found here:
youtube link