It's good sign if both has relative correclation.
let's share both score.
my score:
cv to lb
0.85 :::: 0.875,
0.86 :::: 0.865
0.87 :::: 0.851
0.83x :::: 0.82x
May i ask what your local train f1 is?
I think I(we) might be overfitting?
I did't f1 metric use for training samples instead log_loss, that is 0.08
ok
thanks
0.119
CV to LB
0.90xx ::::::0.81xx
while my train f1 is 0.95xx
CV = 0.8196 | LB = 0.803970223
What model are you using? NN or GBDT?
I am using GBDT. Mine is 0.8583, lb = 0.8512.
@nymfree are you using a cv approach?
Yes I am using a cv-like approach - just a bunch of 2d convs so far.
Alright, I will switch to a cv approach soon.
cv to lb
0.83x :::: 0.82x
May i ask what your local train f1 is?
I think I(we) might be overfitting?
I did't f1 metric use for training samples instead log_loss, that is 0.08
ok
thanks
0.119
CV to LB
0.90xx ::::::0.81xx
while my train f1 is 0.95xx
CV = 0.8196 | LB = 0.803970223
What model are you using? NN or GBDT?
I am using GBDT. Mine is 0.8583, lb = 0.8512.
@nymfree are you using a cv approach?
Yes I am using a cv-like approach - just a bunch of 2d convs so far.
Alright, I will switch to a cv approach soon.