if os.path.exists(constant.save_path + record_file):
checkpoint = True
## read checkpoint
with open(constant.save_path + record_file, newline='') as csvfile:
mLines = csvfile.readlines()
## get current split
targetLine = mLines[-1]
currentSplit = targetLine.split(',')[0]
##read F1 score records
rLines = mLines[-currentSplit - 1:]
for line in rLines:
microF1s += float(line.split(',')[1])
currentSplit += 1
model = get_classifier(ty=classifier,
c=parameter_list[0],
n_estimators=parameter_list[1],
max_depth=parameter_list[2])
for i in range(constant.num_split):
## confirm checkpoint
if checkpoint == True and i < currentSplit:
print("Split {} is skipped because it has been run!".format(i))
continue
## prepare feature for model
X_train, y_train, X_val, y_val, X_test, ind, X_text = get_features_for_prediction(
features, i, use_pca=False)
print('shape of X_train', X_train.shape)
print('shape of X_test', X_test.shape)
print("###### Running folder %d ######" % (i + 1))
## distinguish twitter glove and common glove
if item[:5] == 'glove':
ty = item[6:]
feature = item[:5]
elif item.find('-') > 0:
ty = 'common'
feature = item[:(item.find('-') - 1)]
else:
ty = 'common'
feature = item
## compute Micro F1 score for each feature
for j in range(1, 10):
c = j / 1000
model = get_classifier(ty='LR', c=c)
microF1s = 0
for i in range(constant.num_split):
## prepare data for feature-10 folders
vocab = generate_vocab()
train, val, dev_no_lab = read_data(is_shuffle=True, random_state=i)
## feature_list: glove emoji elmo bert deepmoji emo2vec
## if you want twitter glove or common glove use ty='twitter' and ty='common'
X_train, y_train = get_feature(
train,
vocab,
feature_list=[feature],
mode=['sum'],
split="train",