curve.append(l2)
np.save('reg_curve', curve)
elif args.mb:
for mb in np.linspace(20, 50, 10):
model = RNN(vocab=lexicon, reg=1)
l1, l2 = model.train(X_trees_train,
max_iter=1000,
val_set=X_trees_dev,
strat='AdaGrad',
mini_batch_size=mb)
curve.append(l2)
np.save('mb_curve', curve)
else:
for lr in np.logspace(-2, 3, 10):
model = RNN(vocab=lexicon, reg=1)
l1, l2 = model.train(X_trees_train,
max_iter=1000,
val_set=X_trees_dev,
strat='AdaGrad',
mini_batch_size=30,
learning_rate=lr)
curve.append(l2)
np.save('lr_curve', curve)
sa_trn, sr_trn = model.score_fine(X_trees_train)
sa_val, sr_val = model.score_fine(X_trees_dev)
sa_tst, sr_tst = model.score_fine(X_trees_test)
print 'Fine grain\tTrain\tTest\tValidation'
print 'Overall\t\t{:.3}\t{:.3}\t{:.3}'.format(sa_trn, sa_tst, sa_val)
print 'Root\t\t{:.3}\t{:.3}\t{:.3}'.format(sr_trn, sr_tst, sr_val)
from RNN import RNN
curve = []
if args.reg:
for reg in np.logspace(-2, 3, 10):
model = RNN(vocab=lexicon, reg=reg)
l1, l2 = model.train(X_trees_train, max_iter=1000, val_set=X_trees_dev,
strat='AdaGrad', mini_batch_size=30)
curve.append(l2)
np.save('reg_curve', curve)
elif args.mb:
for mb in np.linspace(20, 50, 10):
model = RNN(vocab=lexicon, reg=1)
l1, l2 = model.train(X_trees_train, max_iter=1000, val_set=X_trees_dev,
strat='AdaGrad', mini_batch_size=mb)
curve.append(l2)
np.save('mb_curve', curve)
else:
for lr in np.logspace(-2, 3, 10):
model = RNN(vocab=lexicon, reg=1)
l1, l2 = model.train(X_trees_train, max_iter=1000, val_set=X_trees_dev,
strat='AdaGrad', mini_batch_size=30, learning_rate=lr)
curve.append(l2)
np.save('lr_curve', curve)
sa_trn, sr_trn = model.score_fine(X_trees_train)
sa_val, sr_val = model.score_fine(X_trees_dev)
sa_tst, sr_tst = model.score_fine(X_trees_test)
print 'Fine grain\tTrain\tTest\tValidation'
print 'Overall\t\t{:.3}\t{:.3}\t{:.3}'.format(sa_trn, sa_tst, sa_val)
print 'Root\t\t{:.3}\t{:.3}\t{:.3}'.format(sr_trn, sr_tst, sr_val)
