Closed Yanchuan99 closed 2 years ago
@Yanchuan99, you can do this with train_seq2seq.py
arguments here.
--vis_dropout=1.0
--lang_dropout=1.0
--zero_instr
--zero_goal
Thanks a lot!
And I'm wondering why default value of --vis_dropout
is 0.3.
Just some regularization to prevent overfitting: https://cs231n.github.io/neural-networks-2/
Hi, Thanks for your great work firstly! I'm going to do the ablation study including input ablation and model ablation. How can I achieve that? I mean, which parameter 's value should I change? Thank you.