Model_v1_8_2 (0, 1) : cv 0.556777(only posi and nega)
Model_v1_10_0 (1, 0) : cv 0.520439 (only posi and nega)
Model_v1_10_1 (1, 0.2) : cv 0.547114(only posi and nega)
Model_v1_10_2 (1, 0.5) : cv 0.551860(only posi and nega)
Model_v1_10_3 (1, 1) : cv 0.546657(only posi and nega)
Other conditions
multi linear head (hidden=[128], dropout=0.1)
dropout=0.1
Learnable weight of averaging hidden layer, n_hid=12, average, learn=False
implement consideration of text_areas
implement remove_excessive_padding
train only positive and negative
label smoothing 0.05
lr 1e-5
different learning rate (x30)
https://www.kaggle.com/koza4ukdmitrij/jaccard-expectation-loss
(JaccardExpectationLoss rate, IndexLoss rate)
Model_v1_8_2 (0, 1) : cv 0.556777(only posi and nega) Model_v1_10_0 (1, 0) : cv 0.520439 (only posi and nega) Model_v1_10_1 (1, 0.2) : cv 0.547114(only posi and nega) Model_v1_10_2 (1, 0.5) : cv 0.551860(only posi and nega) Model_v1_10_3 (1, 1) : cv 0.546657(only posi and nega)
Other conditions multi linear head (hidden=[128], dropout=0.1) dropout=0.1 Learnable weight of averaging hidden layer, n_hid=12, average, learn=False implement consideration of text_areas implement remove_excessive_padding train only positive and negative label smoothing 0.05 lr 1e-5 different learning rate (x30)