Info about hyperparameters about dynamic-coattention-network-for-squad HOT 1 CLOSED

Hi! As I stated in the README, I achieved the mentioned EM and F1 via

python code/train.py --rnn_state_size=150

This means that the state size of the RNN was 150, and all other hyperparameters had its "default value". If you check code/train.py you will see lines like

tf.app.flags.DEFINE_string("model", "DCN", "Choose which model to use baseline/DCN") tf.app.flags.DEFINE_float("learning_rate", 0.001, "Learning rate.") tf.app.flags.DEFINE_integer("batch_size", 64, "Batch size to use during training.") tf.app.flags.DEFINE_integer("epochs", 20, "Number of epochs to train.")

which tell you what these default values are.

from dynamic-coattention-network-for-squad.