search

sunilsu / cs224n

0 stars 0 forks source link

hyper parameter tuning #2

Open sunilsu opened 8 years ago

sunilsu commented 8 years ago

tune learning rate, regularization, input window size, epochs and hidden layer size. Write a program to iterate through different combinations and save the best model parameters.

printStats can be commented out for hyper parameter tuning.

sunilsu commented 8 years ago

WindowModel model = new WindowModel(5, 100, 0.01, 0.001, 10, false);

This is giving a F1 of 81 on dev set

jeasenrys commented 8 years ago

It seems that if we change C from 0.001 to 0.0001, 5 epochs gives F1 of 0.82 with base model-:) Actually C = 0 gives a F1 of 0.82 as well.