Wait, I'm a bit confused. The submission file we are uploading should be the labels (i.e 0's and 1's) or just their probabilities? Because when I use functional API on Keras it doesn't give me the classes direct, so I end up tweaking the code to produce the classes.
That is not the problem.. I doubt; using predict method can never give a score of 0.74LB; your score would rather get stuck btw 45-60LB, though use the predict_proba function as said...your problem has to do with your algorithm(it seems you are using logistic regression)
Try catboost without tuning anything or adding any features, then from there start you feature engineering.
!pip install catboost==0.24.1
Not this
Your problem is maybe not using predict_proba method as @ImonEmmanuel as said clearly
Ok the problem is that you are using predict method nstead of predict_proba method
Try using it
Please are we predicting probability....?
yes
Wait, I'm a bit confused. The submission file we are uploading should be the labels (i.e 0's and 1's) or just their probabilities? Because when I use functional API on Keras it doesn't give me the classes direct, so I end up tweaking the code to produce the classes.
You're to submit the probabilities, read the COMPETITION INFO
That is not the problem.. I doubt; using predict method can never give a score of 0.74LB; your score would rather get stuck btw 45-60LB, though use the predict_proba function as said...your problem has to do with your algorithm(it seems you are using logistic regression)
Yes Emmanuel, I am using predict_proba and LogisticRegreesio, is there a better algorithm?
Ohh nooo...
Well I got 0.70 using predict method. Tho I later got 0.82 using probability. Tho tweaking his algorithm will still go a long