model_save_path = os.path.join(model_dir,
'model-'+str(args.minibatch)+'-'+args.optimizer+'-'+str(args.epochs)+'-'+str(args.step)+'-'+str(fileIdx))
model_save_pre_path = os.path.join(model_dir,
'model-'+str(args.minibatch)+'-'+args.optimizer+'-'+str(args.epochs)+'-'+str(args.step)+'-'+str(fileIdx-1))
if not os.path.exists(model_save_path+".span"):
break
fileIdx += 1
input_var = T.itensor3('inputs')
target_var = T.fmatrix('targets')
wordEmbeddings = loadWord2VecMap(os.path.join(data_dir, 'word2vec.bin'))
wordEmbeddings = wordEmbeddings.astype(np.float32)
if args.mode == "train":
print("Loading training data...")
X_train, Y_labels_train, seqlen, num_feats = read_sequence_dataset_onehot(data_dir, "train")
X_dev, Y_labels_dev,_,_ = read_sequence_dataset_onehot(data_dir, "dev")
print "window_size is %d"%((seqlen-1)/2)
print "number features is %d"%num_feats
train_fn_span, val_fn_span, network_span, train_fn_dcr, val_fn_dcr, network_dcr, \
train_fn_type, val_fn_type, network_type, train_fn_degree, val_fn_degree, network_degree, \
train_fn_pol, val_fn_pol, network_pol, train_fn_cm, val_fn_cm, network_cm, \
model_dir, 'model-' + str(args.minibatch) + '-' + args.optimizer +
'-' + str(args.epochs) + '-' + str(args.step) + '-' + str(fileIdx))
model_save_pre_path = os.path.join(
model_dir,
'model-' + str(args.minibatch) + '-' + args.optimizer + '-' +
str(args.epochs) + '-' + str(args.step) + '-' + str(fileIdx - 1))
if not os.path.exists(model_save_path + ".span"):
break
fileIdx += 1
input_var = T.itensor3('inputs')
target_var = T.fmatrix('targets')
wordEmbeddings = loadWord2VecMap(os.path.join(data_dir, 'word2vec.bin'))
wordEmbeddings = wordEmbeddings.astype(np.float32)
if args.mode == "train":
print("Loading training data...")
X_train, Y_labels_train, seqlen, num_feats = read_sequence_dataset_onehot(
data_dir, "train")
X_dev, Y_labels_dev, _, _ = read_sequence_dataset_onehot(
data_dir, "dev")
print "window_size is %d" % ((seqlen - 1) / 2)
print "number features is %d" % num_feats
train_fn_span, val_fn_span, network_span, \
sick_dir, "train"
)
X1_dev, X1_mask_dev, X2_dev, X2_mask_dev, Y_labels_dev, Y_scores_dev, Y_scores_pred_dev = read_sequence_dataset(
sick_dir, "dev"
)
X1_test, X1_mask_test, X2_test, X2_mask_test, Y_labels_test, Y_scores_test, Y_scores_pred_test = read_sequence_dataset(
sick_dir, "test"
)
input1_var = T.imatrix("inputs_1")
input2_var = T.imatrix("inputs_2")
input1_mask_var = T.matrix("inputs_mask_1")
input2_mask_var = T.matrix("inputs_mask_2")
target_var = T.fmatrix("targets")
wordEmbeddings = loadWord2VecMap(os.path.join(sick_dir, "word2vec.bin"))
wordEmbeddings = wordEmbeddings.astype(np.float32)
network, penalty, input_dict = build_network_double_lstm(
args, input1_var, input1_mask_var, input2_var, input2_mask_var, wordEmbeddings
)
train_fn, val_fn = generate_theano_func(args, network, penalty, input_dict, target_var)
"""
train_fn, val_fn = build_network_2dconv(args, input1_var, input1_mask_var, input2_var, input2_mask_var,
target_var, wordEmbeddings)
"""
epsilon = 1.0e-7
print ("Starting training...")
best_val_acc = 0