Closed Pourbahman closed 3 years ago
Hi Zahra, the number of training instances used by the model is niters
* itersize
, so I agree increasing niters
is the right strategy.
The values we used in the arxiv paper are here: https://github.com/capreolus-ir/capreolus/blob/master/docs/reproduction/config_parade_long-robust04_title.txt#L19
Hi Andrew,
Avvording to your paper, you trained your model on 36 epochs. According to the following https://github.com/capreolus-ir/capreolus/blob/1ef78b2ed327e848305df283ed3f62a8c451061f/capreolus/trainer/tensorflow.py#L259 your current implementation trains the model on 1 epoch. So, if I want to train your model on 2 epochs, I should make twice niter in https://github.com/capreolus-ir/capreolus/blob/3521171ecf38cebfec5e19e22621bf9dfabf58d9/capreolus/trainer/tensorflow.py#L70
Would you please tell me if I am right?
Thanks in advance, Kind Regards