Disclaimer: I apologize for this but my laptop doesn't have the processing power needed to work on the test projects that's why:
- Test 1: I only trained the model for one epoch and it achieved around 0.75 ROC AUC, I believe it could reach the desired 0.8 score.
- Test 2 and 3: my memory overflows whenever I load the parquet files I couldn't even peak at the structure of an instance, so the dataset loading part doesn't work and throws errors for sure and I didn't execute the training part
In this project I used a Vision Transformer Regressor architecture that I have experimented with in machine learning competitions and by far it gives better results than using CNN for regression