def xml_to_brat(input_folder, output_folder, overwrite=True):
print('input_folder: {0}'.format(input_folder))
start_time = time.time()
if overwrite:
shutil.rmtree(output_folder, ignore_errors=True)
utils.create_folder_if_not_exists(output_folder)
for input_filepath in sorted(glob.glob(os.path.join(input_folder, '*.xml'))):
filename = utils.get_basename_without_extension(input_filepath)
output_text_filepath = os.path.join(output_folder, '{0}.txt'.format(filename))
xmldoc = xml.etree.ElementTree.parse(input_filepath).getroot()
# Get text
text = xmldoc.findtext('TEXT')
with codecs.open(output_text_filepath, 'w', 'UTF-8') as f:
f.write(text)
# Get PHI tags
tags = xmldoc.findall('TAGS')[0] # [0] because there is only one <TAGS>...</TAGS>
entities = []
for tag in tags:
entity = {}
entity['label'] = tag.get('TYPE')
entity['text'] = tag.get('text')
entity['start'] = int(tag.get('start'))
entity['end'] = int(tag.get('end'))
entities.append(entity)
output_entities(output_folder, filename, entities, output_text_filepath, text, overwrite=overwrite)
time_spent = time.time() - start_time
print("Time spent formatting: {0:.2f} seconds".format(time_spent))
def generate_reference_text_file_for_conll(conll_input_filepath,
conll_output_filepath, text_folder):
'''
generates reference text files and adds the corresponding filename and token offsets to conll file.
conll_input_filepath: path to a conll-formatted file without filename and token offsets
text_folder: folder to write the reference text file to
'''
dataset_type = utils.get_basename_without_extension(conll_input_filepath)
conll_file = codecs.open(conll_input_filepath, 'r', 'UTF-8')
utils.create_folder_if_not_exists(text_folder)
text = ''
new_conll_string = ''
character_index = 0
document_count = 0
text_base_filename = '{0}_text_{1}'.format(dataset_type,
str(document_count).zfill(5))
for line in conll_file:
split_line = line.strip().split(' ')
# New document
if '-DOCSTART-' in split_line[0]:
new_conll_string += line
if len(text) != 0:
with codecs.open(
os.path.join(text_folder,
'{0}.txt'.format(text_base_filename)),
'w', 'UTF-8') as f:
f.write(text)
text = ''
character_index = 0
document_count += 1
text_base_filename = '{0}_text_{1}'.format(
dataset_type,
str(document_count).zfill(5))
continue
# New sentence
elif len(split_line) == 0 or len(split_line[0]) == 0:
new_conll_string += '\n'
if text != '':
text += '\n'
character_index += 1
continue
token = split_line[0]
start = character_index
end = start + len(token)
text += token + ' '
character_index += len(token) + 1
new_conll_string += ' '.join(
[token, text_base_filename,
str(start), str(end)] + split_line[1:]) + '\n'
if len(text) != 0:
with codecs.open(
os.path.join(text_folder,
'{0}.txt'.format(text_base_filename)), 'w',
'UTF-8') as f:
f.write(text)
conll_file.close()
with codecs.open(conll_output_filepath, 'w', 'UTF-8') as f:
f.write(new_conll_string)
def format_for_ann_old(dataset_base_filename, split):
'''
version without any dictionary or feature
'''
print("Started formatting for ann")
start_time = time.time()
filepaths = utils_deid.get_original_dataset_filepaths(dataset_base_filename, split=split)
output_folder = os.path.join('ann', 'data', dataset_base_filename, 'stanford', split)
create_folder_if_not_exists(output_folder)
number_of_unicode_characters = 0
open('unicode.txt','w').close()
for dataset_type in filepaths:
# if dataset_type == 'test':
# continue
output_filepath = os.path.join(output_folder, '{0}.txt'.format(dataset_type))
open(output_filepath, 'w').close()
# output_file.write('')
for filepath in filepaths[dataset_type]:
print("filepath: {0}".format(filepath))
# filepath = '../data/datasets/original/i2b2deid2016/60_40/training-PHI-Gold-Set1/0666_gs.xml'
xmldoc = xml.etree.ElementTree.parse(filepath).getroot()
# Get text
text = xmldoc.findtext('TEXT')#.replace(u"\u2019", "'")
# .encode('ascii', 'replace')
# try:
# print("text: {0}".format(text))
# except:
# number_of_unicode_characters += 1
# text_replaced = utils_nlp.normalize_unicode_text(text)
# with open('unicode.txt','a') as f:
# f.write(filepath+'\n\n')
# # f.write(text+'\n\n')
# f.write(text_replaced + '\n\n\n======================================================================{0}'.format(number_of_unicode_characters))
# text = text_replaced
# text = text.replace(u"\u2019", "'")
# text = text.replace(u"\u00E3", "a")
# Get stanford output
stanford_output = get_stanford_annotations(text, annotators='tokenize,ssplit')
# Get PHI tags
tags = xmldoc.findall('TAGS')[0] # [0] because there is only one <TAGS>...</TAGS>
phis = []
for tag in tags:
#print(tag)
phi = {}
phi['main_type'] = tag.tag
phi['type'] = tag.get('TYPE')
phi['text'] = tag.get('text')#.replace(u"\u00E3", "a")
phi['start'] = int(tag.get('start'))
phi['end'] = int(tag.get('end'))
phis.append(phi)
xml_filename = utils.get_basename_without_extension(filepath)
convert_stanford_output_to_ann_txt_old(output_filepath, xml_filename, stanford_output, phis)
# 0/0
time_spent = time.time() - start_time
print("Time spent formatting for ann: {0}".format(time_spent))
def check_validity_of_conll_bioes(bioes_filepath):
dataset_type = utils.get_basename_without_extension(bioes_filepath).split(
'_')[0]
print("Checking validity of CONLL BIOES format... ".format(dataset_type),
end='')
input_conll_file = codecs.open(bioes_filepath, 'r', 'UTF-8')
labels_bioes = []
labels_bio = []
for line in input_conll_file:
split_line = line.strip().split(' ')
# New sentence
if len(split_line) == 0 or len(
split_line[0]) == 0 or '-DOCSTART-' in split_line[0]:
if check_bio_bioes_compatibility(labels_bio, labels_bioes):
continue
return False
label_bioes = split_line[-1]
label_bio = split_line[-2]
labels_bioes.append(label_bioes)
labels_bio.append(label_bio)
input_conll_file.close()
if check_bio_bioes_compatibility(labels_bio, labels_bioes):
print("Done.")
return True
return False
def convert_conll_from_bio_to_bioes(input_conll_filepath, output_conll_filepath):
if os.path.exists(output_conll_filepath):
if check_validity_of_conll_bioes(output_conll_filepath):
return
dataset_type = utils.get_basename_without_extension(input_conll_filepath).split('_')[0]
print("Converting CONLL from BIO to BIOES format... ".format(dataset_type), end='')
input_conll_file = codecs.open(input_conll_filepath, 'r', 'UTF-8')
output_conll_file = codecs.open(output_conll_filepath, 'w', 'UTF-8')
labels = []
split_lines = []
for line in input_conll_file:
split_line = line.strip().split(' ')
# New sentence
if len(split_line) == 0 or len(split_line[0]) == 0 or '-DOCSTART-' in split_line[0]:
output_conll_lines_with_bioes(split_lines, labels, output_conll_file)
output_conll_file.write(line)
continue
label = split_line[-1]
labels.append(label)
split_lines.append(split_line)
output_conll_lines_with_bioes(split_lines, labels, output_conll_file)
input_conll_file.close()
output_conll_file.close()
print("Done.")
def predict(text):
# if prediction_count == 1:
parameters['dataset_text_folder'] = os.path.join('..', 'data', 'temp')
stats_graph_folder, _ = utils.create_stats_graph_folder(parameters)
# Update the deploy folder, file, and dataset
dataset_type = 'deploy'
### Delete all deployment data
for filepath in glob.glob(
os.path.join(parameters['dataset_text_folder'],
'{0}*'.format(dataset_type))):
if os.path.isdir(filepath):
shutil.rmtree(filepath)
else:
os.remove(filepath)
### Create brat folder and file
dataset_brat_deploy_folder = os.path.join(
parameters['dataset_text_folder'], dataset_type)
utils.create_folder_if_not_exists(dataset_brat_deploy_folder)
dataset_brat_deploy_filepath = os.path.join(
dataset_brat_deploy_folder,
'temp_{0}.txt'.format(str(prediction_count).zfill(5))
) # self._get_dataset_brat_deploy_filepath(dataset_brat_deploy_folder)
with codecs.open(dataset_brat_deploy_filepath, 'w', 'UTF-8') as f:
f.write(text)
### Update deploy filepaths
dataset_filepaths, dataset_brat_folders = utils.get_valid_dataset_filepaths(
parameters, dataset_types=[dataset_type])
dataset_filepaths.update(dataset_filepaths)
dataset_brat_folders.update(dataset_brat_folders)
### Update the dataset for the new deploy set
dataset.update_dataset(dataset_filepaths, [dataset_type])
# Predict labels and output brat
output_filepaths = {}
prediction_output = train.prediction_step(
sess, dataset, dataset_type, model, transition_params_trained,
stats_graph_folder, prediction_count, dataset_filepaths,
parameters['tagging_format'], parameters['main_evaluation_mode'])
_, _, output_filepaths[dataset_type] = prediction_output
conll2brat.output_brat(output_filepaths,
dataset_brat_folders,
stats_graph_folder,
overwrite=True)
# Print and output result
text_filepath = os.path.join(
stats_graph_folder, 'brat', 'deploy',
os.path.basename(dataset_brat_deploy_filepath))
annotation_filepath = os.path.join(
stats_graph_folder, 'brat', 'deploy', '{0}.ann'.format(
utils.get_basename_without_extension(
dataset_brat_deploy_filepath)))
text2, entities = brat2conll.get_entities_from_brat(text_filepath,
annotation_filepath,
verbose=True)
assert (text == text2)
return entities
def _create_stats_graph_folder(self, parameters):
# Initialize stats_graph_folder
experiment_timestamp = utils.get_current_time_in_miliseconds()
dataset_name = utils.get_basename_without_extension(parameters['dataset_text_folder'])
model_name = '{0}_{1}'.format(dataset_name, experiment_timestamp)
utils.create_folder_if_not_exists(parameters['output_folder'])
stats_graph_folder = os.path.join(parameters['output_folder'], model_name) # Folder where to save graphs
utils.create_folder_if_not_exists(stats_graph_folder)
return stats_graph_folder, experiment_timestamp
def check_compatibility_between_conll_and_brat_text(conll_filepath,
brat_folder):
'''
check if token offsets match between conll and brat .txt files.
conll_filepath: path to conll file
brat_folder: folder that contains the .txt (and .ann) files that are formatted according to brat.
'''
dataset_type = utils.get_basename_without_extension(conll_filepath)
print("Checking compatibility between CONLL and BRAT for {0} set ... ".
format(dataset_type),
end='')
print('**** conll_filepath=%s' % conll_filepath)
conll_file = codecs.open(conll_filepath, 'r', 'UTF-8')
previous_filename = ''
for line in conll_file:
line = line.strip().split(' ')
# New sentence
if len(line) == 0 or len(line[0]) == 0 or '-DOCSTART-' in line[0]:
continue
filename = str(line[1])
# New file
if filename != previous_filename:
text_filepath = os.path.join(brat_folder,
'{0}.txt'.format(filename))
try:
text = utils.read_file(text_filepath)
except:
print('$' * 80)
print(line)
raise
previous_filename = filename
label = str(line[-1]).replace('_', '-') # For LOCATION-OTHER
token = {}
token['text'] = str(line[0])
token['start'] = int(line[2])
token['end'] = int(line[3])
# check that the token text matches the original
if token['text'] != text[token['start']:token['end']]:
print("Warning: conll and brat text do not match.")
print("\tCONLL: {0}".format(token['text']))
print("\tBRAT : {0}".format(text[token['start']:token['end']]))
if token['text'] != text[token['start']:token['end']].replace(
' ', '-'):
raise AssertionError("CONLL and BRAT files are incompatible.")
print("Done.")
def _create_stats_graph_folder(self, parameters):
'''
Tạo folder output/en để chứa các file sinh ra trong quá trình huấn luyện
Trả về
(đường dẫn đến folder output, timestamp của lần chạy)
'''
experiment_timestamp = utils.get_current_time_in_miliseconds()
dataset_name = utils.get_basename_without_extension(parameters['dataset_text_folder'])
model_name = '{0}_{1}'.format(dataset_name, experiment_timestamp)
utils.create_folder_if_not_exists(parameters['output_folder'])
stats_graph_folder = os.path.join(parameters['output_folder'], model_name) # Folder where to save graphs
utils.create_folder_if_not_exists(stats_graph_folder)
return stats_graph_folder, experiment_timestamp
def generate_reference_text_file_for_conll(conll_input_filepath, conll_output_filepath, text_folder):
'''
Từ file conll ban đầu, tiến hành generate ngược lại các file text (nằm trong text_folder) và conll_reference (conll_output_filepath)
Cấu trúc conll_output_filepath:
- DOCSTART: bắt đầu văn bản
- <token> <tên file> <startIndex> <endIndex> < 3 nhãn chuẩn conll>
'''
dataset_type = utils.get_basename_without_extension(conll_input_filepath) # lấy filename mà không có phần mở rộng
conll_file = codecs.open(conll_input_filepath, 'r', 'UTF-8')
utils.create_folder_if_not_exists(text_folder)
text = ''
new_conll_string = ''
character_index = 0
document_count = 0
text_base_filename = '{0}_text_{1}'.format(dataset_type, str(document_count).zfill(5)) # zfill: dùng để format string đảm bảo luôn có 5 ký tự
for line in conll_file:
split_line = line.strip().split(' ')
# Bắt đầu 1 document: có chuỗi '-DOCSTART-' đầu câu
if '-DOCSTART-' in split_line[0]:
new_conll_string += line
if len(text) != 0:
with codecs.open(os.path.join(text_folder, '{0}.txt'.format(text_base_filename)), 'w', 'UTF-8') as f:
f.write(text)
text = ''
character_index = 0
document_count += 1
text_base_filename = '{0}_text_{1}'.format(dataset_type, str(document_count).zfill(5))
continue
# Bắt đầu 1 câu
elif len(split_line) == 0 or len(split_line[0]) == 0:
new_conll_string += '\n'
if text != '':
text += '\n'
character_index += 1
continue
token = split_line[0]
start = character_index
end = start + len(token)
text += token + ' '
character_index += len(token) + 1
new_conll_string += ' '.join([token, text_base_filename, str(start), str(end)] + split_line[1:]) + '\n'
if len(text) != 0:
with codecs.open(os.path.join(text_folder, '{0}.txt'.format(text_base_filename)), 'w', 'UTF-8') as f:
f.write(text)
conll_file.close()
with codecs.open(conll_output_filepath, 'w', 'UTF-8') as f:
f.write(new_conll_string)
def main():
'''
This is the main function
'''
#stats_graph_folder=os.path.join('..', 'stats_graphs', 'test')
stats_graph_folder = os.path.join('..', 'output')
# Getting a list of all subdirectories in the current directory. Not recursive.
subfolders = os.listdir(stats_graph_folder)
subfolders = sorted(os.listdir(stats_graph_folder), reverse=True)
# Recursive
#subfolders = [x[0] for x in os.walk(stats_graph_folder)][1:]
# Parameters
#metrics = ['accuracy_score', 'f1_score']
metrics = ['f1_score', 'f1_conll']
dataset_types = ['train', 'valid', 'test']
execution_details = [
'num_epochs', 'train_duration', 'keyboard_interrupt', 'early_stop'
]
# It's good to put the important fields (for your experiments) first,
# so that it appears right next to the test f1 score.
fields_of_interest = '''dataset_text_filepath all_emb pre_emb char_dim char_bidirect character_cnn_filter_height character_cnn_number_of_filters
word_dim using_token_lstm word_lstm_dim word_bidirect experiment_name
using_token_cnn token_cnn_filter_height token_cnn_number_of_filters using_character_lstm using_character_cnn patience char_lstm_dim
crf dropout lr_method training_set_size'''.replace('\n', '').split(' ')
fields_of_interest = filter(None, fields_of_interest)
result_tables = {}
print('subfolders: {0}'.format(subfolders))
# 0/0
# Define column_order, i.e. how the result table is presented
column_order = ['dataset_name', 'time_stamp']
for metric in metrics:
for dataset_type in dataset_types:
column_order.append('{0}_{1}'.format(dataset_type, metric))
column_order.append('{0}_{1} (based on valid)'.format('test', metric))
column_order.extend(fields_of_interest[:3])
for metric in metrics:
column_order.append(
'best_epoch_for_{0} (based on valid)'.format(metric))
column_order.extend(execution_details)
column_order.extend(fields_of_interest[3:])
print('fields_of_interest: {0}'.format(fields_of_interest))
print('column_order: {0}'.format(column_order))
for subfolder in subfolders:
result_row = {}
result_filepath = os.path.join(stats_graph_folder, subfolder,
'results.json')
if not os.path.isfile(result_filepath): continue
print('result_filepath: {0}'.format(result_filepath))
try:
result_json = json.load(open(result_filepath, 'r'))
except ValueError:
print('This file is skipped since it is in use or corrupted.')
# Include time stamp of the experiments
result_row['time_stamp'] = result_json['execution_details'][
'time_stamp']
for field_of_interest in fields_of_interest:
if field_of_interest in result_json['model_options']:
if field_of_interest == 'pre_emb':
result_row[field_of_interest] = os.path.basename(
result_json['model_options'][field_of_interest])
else:
result_row[field_of_interest] = result_json[
'model_options'][field_of_interest]
for execution_detail in execution_details:
try:
result_row[execution_detail] = result_json[
'execution_details'][execution_detail]
except:
result_row[execution_detail] = 'NULL'
for metric in metrics:
for dataset_type in dataset_types:
result_row['{0}_{1}'.format(
dataset_type, metric)] = result_json[dataset_type].get(
'best_{0}'.format(metric), 'NULL')
if dataset_type == 'test':
result_row['{0}_{1} (based on valid)'.format(
dataset_type, metric)] = result_json[dataset_type].get(
'best_{0}_based_on_valid'.format(metric), 'NULL')
elif dataset_type == 'valid':
result_row['best_epoch_for_{0} (based on valid)'.format(
metric)] = result_json[dataset_type].get(
'epoch_for_best_{0}'.format(metric), 'NULL')
# Save row in table: one table per data set
dataset_name = utils.get_basename_without_extension(
result_row['dataset_text_filepath'])
result_row['dataset_name'] = dataset_name
if dataset_name not in result_tables:
result_tables[dataset_name] = []
result_row_ordered = []
for column_name in column_order:
if column_name in result_row:
result_row_ordered.append(result_row[column_name])
else:
result_row_ordered.append('NULL')
result_tables[dataset_name].append(result_row_ordered)
print('result_tables: {0}'.format(result_tables))
#print('\ncolumn_order: {0}'.format(column_order))
#print('result_table: {0}'.format(result_tables))
import MySQLdb as mdb
connection = mdb.connect('128.52.165.241', 'tc',
open('database_password.txt', 'r').readline(),
'tc')
cursor = connection.cursor()
for dataset_name, dataset_results in result_tables.items():
with open(
os.path.join(stats_graph_folder,
'results_{0}.csv'.format(dataset_name)),
'wb') as testfile:
csv_writer = csv.writer(testfile)
clean_column_names = map(clean_column_name, column_order)
csv_writer.writerow(clean_column_names)
for row in dataset_results:
csv_writer.writerow(row)
# Convert row values to some importable string
values = ''
for value_number, value in enumerate(row):
if value_number > 0:
values += ','
if isinstance(value,
(bool)): # Check if object is a boolean
if value:
value = '1'
else:
value = '0'
if not isinstance(
value, (int, long)
) and value != 'NULL': # http://stackoverflow.com/questions/3501382/checking-whether-a-variable-is-an-integer-or-not
value = '"{0}"'.format(str(value))
else:
value = '{0}'.format(str(value))
#print('value: {0}'.format(value))
values += value
# Make sure that train_duration is more than 0. (if 0 it means the training was interrupted)
#print('row[clean_column_names.index("train_duration"): {0}'.format(row[clean_column_names.index('train_duration')]))
train_duration = row[clean_column_names.index(
'train_duration')]
keyboard_interrupt = row[clean_column_names.index(
'keyboard_interrupt')]
#print('train_duration: {0}'.format(train_duration))
'''
if train_duration == 'NULL' and keyboard_interrupt == '0' or train_duration == '0':
print('The experiment has train_duration = {0}, so we skip it.'.format(train_duration))
continue
'''
# Make sure the experiment isn't already in the database
time_stamp = row[clean_column_names.index('time_stamp')]
sql = 'SELECT COUNT(*) FROM tc.results_neurodeid WHERE time_stamp = "{0}"'.format(
time_stamp)
cursor.execute(sql)
row = cursor.fetchone()
if row[0] >= 1:
print(
'The experiment with timestamp {0} is already in the database, so we skip it.'
.format(time_stamp))
continue
if time_stamp < '2016-08-17_18-20-05-836274':
print(
'The experiment with timestamp {0} is too old, so we skip it.'
.format(time_stamp))
continue
sql = 'INSERT INTO tc.results_neurodeid ({0}) VALUES ({1})'.format(
','.join(clean_column_names), values)
print('sql: {0}'.format(sql))
cursor.execute(sql)
connection.commit()
connection.commit()
connection.close()
def conll_to_brat(conll_input_filepath, conll_output_filepath, brat_original_folder, brat_output_folder,
overwrite=False):
'''
convert conll file in conll-filepath to brat annotations and output to brat_output_folder,
with reference to the existing text files in brat_original_folder
if brat_original_folder does not exist or contain any text file, then the text files are generated from conll files,
and conll file is updated with filenames and token offsets accordingly.
conll_input_filepath: path to conll file to convert to brat annotations
conll_output_filepath: path to output conll file with filename and offsets that are compatible with brat annotations
brat_original_folder: folder that contains the original .txt (and .ann) files that are formatted according to brat.
.txt files are used to check if the token offsets match and generate the annotation from conll.
brat_output_folder: folder to output the text and brat annotations
.txt files are copied from brat_original_folder to brat_output_folder
'''
verbose = False
dataset_type = utils.get_basename_without_extension(conll_input_filepath)
print("Formatting {0} set from CONLL to BRAT... ".format(dataset_type), end='')
# if brat_original_folder does not exist or have any text file
if not os.path.exists(brat_original_folder) or len(glob.glob(os.path.join(brat_original_folder, '*.txt'))) == 0:
assert (conll_input_filepath != conll_output_filepath)
generate_reference_text_file_for_conll(conll_input_filepath, conll_output_filepath, brat_original_folder)
utils.create_folder_if_not_exists(brat_output_folder)
conll_file = codecs.open(conll_output_filepath, 'r', 'latin-1', errors='replace')
previous_token_label = 'O'
previous_filename = ''
text_filepath = ''
text = ''
entity_id = 1
entities = []
entity = {}
line_count = 0
for line in conll_file:
line = line.strip().split(' ')
# New sentence
if len(line) == 0 or len(line[0]) == 0 or '-DOCSTART-' in line[0]:
# Add the last entity
if entity != {}:
if verbose: print("entity: {0}".format(entity))
entities.append(entity)
entity_id += 1
entity = {}
previous_token_label = 'O'
continue
filename = str(line[1])
# New file
if filename != previous_filename:
output_entities(brat_output_folder, previous_filename, entities, text_filepath, text, overwrite=overwrite)
text_filepath = os.path.join(brat_original_folder, '{0}.txt'.format(filename))
with codecs.open(text_filepath, 'r', 'latin-1', errors='replace') as f:
text = f.read()
previous_token_label = 'O'
previous_filename = filename
entity_id = 1
entities = []
entity = {}
label = str(line[-1]).replace('_', '-') # For LOCATION-OTHER
if label == 'O':
# Previous entity ended
if previous_token_label != 'O':
if verbose: print("entity: {0}".format(entity))
entities.append(entity)
entity_id += 1
entity = {}
previous_token_label = 'O'
continue
token = {}
token['text'] = str(line[0])
token['start'] = int(line[2])
token['end'] = int(line[3])
# check that the token text matches the original
if token['text'] != text[token['start']:token['end']].replace(' ', '-'):
print("Warning: conll and brat text do not match.")
print("\tCONLL: {0}".format(token['text']))
print("\tBRAT : {0}".format(text[token['start']:token['end']]))
token['label'] = label[2:]
if label[:2] == 'B-':
if previous_token_label != 'O':
# End the previous entity
if verbose: print("entity: {0}".format(entity))
entities.append(entity)
entity_id += 1
# Start a new entity
entity = token
elif label[:2] == 'I-':
# Entity continued
if previous_token_label == token['label']:
# if there is no newline between the entity and the token
if '\n' not in text[entity['end']:token['start']]:
# Update entity
entity['text'] = entity['text'] + ' ' + token['text']
entity['end'] = token['end']
else: # newline between the entity and the token
# End the previous entity
if verbose: print("entity: {0}".format(entity))
entities.append(entity)
entity_id += 1
# Start a new entity
entity = token
elif previous_token_label != 'O':
# TODO: count BI or II incompatibility
# End the previous entity
if verbose: print("entity: {0}".format(entity))
entities.append(entity)
entity_id += 1
# Start new entity
entity = token
else: # previous_token_label == 'O'
# TODO: count OI incompatibility
# Start new entity
entity = token
previous_token_label = token['label']
output_entities(brat_output_folder, previous_filename, entities, text_filepath, text, overwrite=overwrite)
conll_file.close()
print('Done.')
def main():
parameters, conf_parameters = load_parameters()
dataset_filepaths, dataset_brat_folders = get_valid_dataset_filepaths(parameters)
check_parameter_compatiblity(parameters, dataset_filepaths)
# Load dataset
dataset = ds.Dataset(verbose=parameters['verbose'], debug=parameters['debug'])
dataset.load_dataset(dataset_filepaths, parameters)
# Create graph and session
with tf.Graph().as_default():
session_conf = tf.ConfigProto(
intra_op_parallelism_threads=parameters['number_of_cpu_threads'],
inter_op_parallelism_threads=parameters['number_of_cpu_threads'],
device_count={'CPU': 1, 'GPU': parameters['number_of_gpus']},
allow_soft_placement=True, # automatically choose an existing and supported device to run the operations in case the specified one doesn't exist
log_device_placement=False
)
sess = tf.Session(config=session_conf)
with sess.as_default():
# Initialize and save execution details
start_time = time.time()
experiment_timestamp = utils.get_current_time_in_miliseconds()
results = {}
results['epoch'] = {}
results['execution_details'] = {}
results['execution_details']['train_start'] = start_time
results['execution_details']['time_stamp'] = experiment_timestamp
results['execution_details']['early_stop'] = False
results['execution_details']['keyboard_interrupt'] = False
results['execution_details']['num_epochs'] = 0
results['model_options'] = copy.copy(parameters)
dataset_name = utils.get_basename_without_extension(parameters['dataset_text_folder'])
model_name = '{0}_{1}'.format(dataset_name, results['execution_details']['time_stamp'])
output_folder=os.path.join('..', 'output')
utils.create_folder_if_not_exists(output_folder)
stats_graph_folder=os.path.join(output_folder, model_name) # Folder where to save graphs
utils.create_folder_if_not_exists(stats_graph_folder)
model_folder = os.path.join(stats_graph_folder, 'model')
utils.create_folder_if_not_exists(model_folder)
with open(os.path.join(model_folder, 'parameters.ini'), 'w') as parameters_file:
conf_parameters.write(parameters_file)
tensorboard_log_folder = os.path.join(stats_graph_folder, 'tensorboard_logs')
utils.create_folder_if_not_exists(tensorboard_log_folder)
tensorboard_log_folders = {}
for dataset_type in dataset_filepaths.keys():
tensorboard_log_folders[dataset_type] = os.path.join(stats_graph_folder, 'tensorboard_logs', dataset_type)
utils.create_folder_if_not_exists(tensorboard_log_folders[dataset_type])
pickle.dump(dataset, open(os.path.join(model_folder, 'dataset.pickle'), 'wb'))
# Instantiate the model
# graph initialization should be before FileWriter, otherwise the graph will not appear in TensorBoard
model = EntityLSTM(dataset, parameters)
# Instantiate the writers for TensorBoard
writers = {}
for dataset_type in dataset_filepaths.keys():
writers[dataset_type] = tf.summary.FileWriter(tensorboard_log_folders[dataset_type], graph=sess.graph)
embedding_writer = tf.summary.FileWriter(model_folder) # embedding_writer has to write in model_folder, otherwise TensorBoard won't be able to view embeddings
embeddings_projector_config = projector.ProjectorConfig()
tensorboard_token_embeddings = embeddings_projector_config.embeddings.add()
tensorboard_token_embeddings.tensor_name = model.token_embedding_weights.name
token_list_file_path = os.path.join(model_folder, 'tensorboard_metadata_tokens.tsv')
tensorboard_token_embeddings.metadata_path = os.path.relpath(token_list_file_path, '..')
tensorboard_character_embeddings = embeddings_projector_config.embeddings.add()
tensorboard_character_embeddings.tensor_name = model.character_embedding_weights.name
character_list_file_path = os.path.join(model_folder, 'tensorboard_metadata_characters.tsv')
tensorboard_character_embeddings.metadata_path = os.path.relpath(character_list_file_path, '..')
projector.visualize_embeddings(embedding_writer, embeddings_projector_config)
# Write metadata for TensorBoard embeddings
token_list_file = codecs.open(token_list_file_path,'w', 'UTF-8')
for token_index in range(dataset.vocabulary_size):
token_list_file.write('{0}\n'.format(dataset.index_to_token[token_index]))
token_list_file.close()
character_list_file = codecs.open(character_list_file_path,'w', 'UTF-8')
for character_index in range(dataset.alphabet_size):
if character_index == dataset.PADDING_CHARACTER_INDEX:
character_list_file.write('PADDING\n')
else:
character_list_file.write('{0}\n'.format(dataset.index_to_character[character_index]))
character_list_file.close()
# Initialize the model
sess.run(tf.global_variables_initializer())
if not parameters['use_pretrained_model']:
model.load_pretrained_token_embeddings(sess, dataset, parameters)
# Start training + evaluation loop. Each iteration corresponds to 1 epoch.
bad_counter = 0 # number of epochs with no improvement on the validation test in terms of F1-score
previous_best_valid_f1_score = 0
transition_params_trained = np.random.rand(len(dataset.unique_labels)+2,len(dataset.unique_labels)+2)
model_saver = tf.train.Saver(max_to_keep=parameters['maximum_number_of_epochs']) # defaults to saving all variables
epoch_number = -1
try:
while True:
step = 0
epoch_number += 1
print('\nStarting epoch {0}'.format(epoch_number))
epoch_start_time = time.time()
if parameters['use_pretrained_model'] and epoch_number == 0:
# Restore pretrained model parameters
transition_params_trained = train.restore_model_parameters_from_pretrained_model(parameters, dataset, sess, model, model_saver)
elif epoch_number != 0:
# Train model: loop over all sequences of training set with shuffling
sequence_numbers=list(range(len(dataset.token_indices['train'])))
random.shuffle(sequence_numbers)
for sequence_number in sequence_numbers:
transition_params_trained = train.train_step(sess, dataset, sequence_number, model, transition_params_trained, parameters)
step += 1
if step % 10 == 0:
print('Training {0:.2f}% done'.format(step/len(sequence_numbers)*100), end='\r', flush=True)
epoch_elapsed_training_time = time.time() - epoch_start_time
print('Training completed in {0:.2f} seconds'.format(epoch_elapsed_training_time), flush=True)
y_pred, y_true, output_filepaths = train.predict_labels(sess, model, transition_params_trained, parameters, dataset, epoch_number, stats_graph_folder, dataset_filepaths)
# Evaluate model: save and plot results
evaluate.evaluate_model(results, dataset, y_pred, y_true, stats_graph_folder, epoch_number, epoch_start_time, output_filepaths, parameters)
if parameters['use_pretrained_model'] and not parameters['train_model']:
conll_to_brat.output_brat(output_filepaths, dataset_brat_folders, stats_graph_folder)
break
# Save model
model_saver.save(sess, os.path.join(model_folder, 'model_{0:05d}.ckpt'.format(epoch_number)))
# Save TensorBoard logs
summary = sess.run(model.summary_op, feed_dict=None)
writers['train'].add_summary(summary, epoch_number)
writers['train'].flush()
utils.copytree(writers['train'].get_logdir(), model_folder)
# Early stop
valid_f1_score = results['epoch'][epoch_number][0]['valid']['f1_score']['micro']
if valid_f1_score > previous_best_valid_f1_score:
bad_counter = 0
previous_best_valid_f1_score = valid_f1_score
conll_to_brat.output_brat(output_filepaths, dataset_brat_folders, stats_graph_folder, overwrite=True)
else:
bad_counter += 1
print("The last {0} epochs have not shown improvements on the validation set.".format(bad_counter))
if bad_counter >= parameters['patience']:
print('Early Stop!')
results['execution_details']['early_stop'] = True
break
if epoch_number >= parameters['maximum_number_of_epochs']: break
except KeyboardInterrupt:
results['execution_details']['keyboard_interrupt'] = True
print('Training interrupted')
print('Finishing the experiment')
end_time = time.time()
results['execution_details']['train_duration'] = end_time - start_time
results['execution_details']['train_end'] = end_time
print('ok1')
evaluate.save_results(results, stats_graph_folder)
print('ok2')
print('ok3')
#sess.close() # release the session's resources
print('ok4')
def main():
file_params = 'parameters_yelp_50k.ini'
if len(sys.argv) > 1 and '.ini' in sys.argv[1]:
file_params = sys.argv[1]
# Load config
parameters, conf_parameters = load_parameters(
parameters_filepath=os.path.join('.', file_params))
dataset_filepaths = get_valid_dataset_filepaths(parameters)
#check_parameter_compatiblity(parameters, dataset_filepaths)
if parameters['seed'] != -1:
random.seed(parameters['seed'])
# Create annotator
annotator = stanford_corenlp_pywrapper.CoreNLP(
configdict={
'annotators': 'tokenize, ssplit',
'ssplit.eolonly': True
},
corenlp_jars=[parameters['stanford_folder'] + '/*'])
# Load dataset
dataset = ds.Dataset(verbose=parameters['verbose'],
debug=parameters['debug'])
dataset.load_dataset(dataset_filepaths, parameters, annotator)
# Adapt train/valid/test to be multiple of batch_size
for size in ['train_size', 'valid_size', 'test_size']:
if parameters[size] % parameters['batch_size'] != 0:
parameters[size] = int(
parameters[size] /
parameters['batch_size']) * parameters['batch_size']
print('Changed {}'.format(size))
# Set GPU device if more GPUs are specified
if parameters['number_of_gpus'] > 1 and parameters['gpu_device'] != -1:
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
os.environ["CUDA_VISIBLE_DEVICES"] = parameters['gpu_device']
# GPUs
print(device_lib.list_local_devices())
# Create graph and session
with tf.Graph().as_default():
session_conf = tf.ConfigProto(
intra_op_parallelism_threads=parameters['number_of_cpu_threads'],
inter_op_parallelism_threads=parameters['number_of_cpu_threads'],
device_count={
'CPU': 1,
'GPU': parameters['number_of_gpus']
},
allow_soft_placement=
True, # automatically choose an existing and supported device to run the operations in case the specified one doesn't exist
log_device_placement=False)
sess = tf.Session(config=session_conf)
with sess.as_default():
if parameters['seed'] != -1:
tf.set_random_seed(parameters['seed'])
# Initialize and save execution details
start_time = time.time()
experiment_timestamp = utils.get_current_time_in_miliseconds()
results = {}
results['epoch'] = {}
results['execution_details'] = {}
results['execution_details']['train_start'] = start_time
results['execution_details'][
'time_stamp'] = experiment_timestamp
results['execution_details']['early_stop'] = False
results['execution_details']['keyboard_interrupt'] = False
results['execution_details']['num_epochs'] = 0
results['model_options'] = copy.copy(parameters)
dataset_name = utils.get_basename_without_extension(
parameters['dataset_folder'])
model_name = '{0}_{1}'.format(
dataset_name, results['execution_details']['time_stamp'])
output_folder = os.path.join('..', 'output')
utils.create_folder_if_not_exists(output_folder)
stats_graph_folder = os.path.join(
output_folder, model_name) # Folder where to save graphs
utils.create_folder_if_not_exists(stats_graph_folder)
model_folder = os.path.join(stats_graph_folder, 'model')
utils.create_folder_if_not_exists(model_folder)
with open(os.path.join(model_folder, file_params),
'w') as parameters_file:
conf_parameters.write(parameters_file)
pickle.dump(
dataset,
open(os.path.join(model_folder, 'dataset.pickle'), 'wb'))
# Instantiate the model
# graph initialization should be before FileWriter, otherwise the graph will not appear in TensorBoard
model = SelfSent(dataset, parameters)
# Initialize the model
sess.run(tf.global_variables_initializer())
if not parameters['use_pretrained_model']:
model.load_pretrained_token_embeddings(
sess, dataset, parameters)
# Start training + evaluation loop. Each iteration corresponds to 1 epoch.
bad_counter = 0 # number of epochs with no improvement on the validation test
previous_best_valid_accuracy = 0
previous_best_test_accuracy = 0
model_saver = tf.train.Saver(
max_to_keep=parameters['maximum_number_of_epochs']
) # defaults to saving all variables
epoch_number = -1
try:
while True:
epoch_number += 1
print('\nStarting epoch {0}'.format(epoch_number))
epoch_start_time = time.time()
if parameters[
'use_pretrained_model'] and epoch_number == 0:
# Restore pretrained model parameters
dataset = train.restore_model_parameters_from_pretrained_model(
parameters, dataset, sess, model_saver)
dataset.load_deploy(
os.path.join(parameters['dataset_folder'],
'{0}.json'.format('deploy')),
parameters, annotator)
y_pred, y_true, output_filepaths, attentions = train.predict_labels(
sess,
model,
parameters,
dataset,
epoch_number,
stats_graph_folder,
dataset_filepaths,
only_deploy=True)
y_pred = y_pred['deploy']
with open(
output_filepaths['deploy']
[:output_filepaths['deploy'].rfind('/') + 1] +
'attention.txt',
'w',
encoding='utf-8') as fp:
# Compute attention
tokens_with_attentions = []
for sample_id in range(len(y_pred)):
attention = attentions[int(
sample_id / parameters['batch_size'])][
sample_id %
parameters['batch_size']]
# Remove padded dimension
attention = attention[:dataset.
token_lengths[
'deploy']
[sample_id]]
# Save current attention
fp.write("{}\t{:05.2f}\t".format(
y_pred[sample_id][0],
y_pred[sample_id][1]))
fp.write(' '.join(dataset.tokens['deploy']
[sample_id]) + '\t')
fp.write(' '.join(
[str(a)
for a in attention.flatten()]) + '\n')
# Sum over columns (we combine all the annotation vectors)
attention = np.sum(attention, axis=1)
# Normalize to sum at 1
attention = attention / np.linalg.norm(
attention)
# Keep only high confidence
if y_pred[sample_id][1] >= parameters[
'attention_visualization_conf']:
tokens_with_attentions.append(
(y_pred[sample_id][0],
y_pred[sample_id][1],
dataset.tokens['deploy']
[sample_id], attention))
# Plot attention
utils_plots.visualize_attention(
tokens_with_attentions, dataset.unique_labels,
output_filepaths['deploy']
[:output_filepaths['deploy'].rfind('/') + 1],
parameters['attention_visualization_conf'])
break
elif epoch_number != 0:
total_loss, total_accuracy = train.train_step(
sess, dataset, model, parameters)
print('Mean loss: {:.2f}\tMean accuracy: {:.2f}'.
format(np.mean(total_loss),
100.0 * np.mean(total_accuracy)),
flush=True)
epoch_elapsed_training_time = time.time(
) - epoch_start_time
print('Training completed in {0:.2f} seconds'.format(
epoch_elapsed_training_time),
flush=True)
y_pred, y_true, output_filepaths, _ = train.predict_labels(
sess, model, parameters, dataset, epoch_number,
stats_graph_folder, dataset_filepaths)
# Evaluate model: save and plot results
evaluate.evaluate_model(results, dataset, y_pred,
y_true, stats_graph_folder,
epoch_number, epoch_start_time,
output_filepaths, parameters)
# Save model
model_saver.save(
sess,
os.path.join(
model_folder,
'model_{0:05d}.ckpt'.format(epoch_number)))
# Early stop
valid_accuracy = results['epoch'][epoch_number][0][
'valid']['accuracy_score']
if valid_accuracy > previous_best_valid_accuracy:
bad_counter = 0
previous_best_valid_accuracy = valid_accuracy
previous_best_test_accuracy = results['epoch'][
epoch_number][0]['test']['accuracy_score']
else:
bad_counter += 1
print(
"The last {0} epochs have not shown improvements on the validation set."
.format(bad_counter))
print("Best valid with test performances in epoch " +
str(epoch_number - bad_counter) +
": {:05.2f}%\t{:05.2f}%".format(
previous_best_valid_accuracy,
previous_best_test_accuracy))
if bad_counter >= parameters['patience']:
print('Early Stop!')
results['execution_details']['early_stop'] = True
break
if epoch_number >= parameters[
'maximum_number_of_epochs']:
break
except KeyboardInterrupt:
results['execution_details']['keyboard_interrupt'] = True
print('Training interrupted')
print('Finishing the experiment')
end_time = time.time()
results['execution_details'][
'train_duration'] = end_time - start_time
results['execution_details']['train_end'] = end_time
evaluate.save_results(results, stats_graph_folder)
sess.close() # release the session's resources
def main(argv=sys.argv):
arguments = parse_arguments(argv[1:])
parameters, conf_parameters = load_parameters(
arguments['parameters_filepath'], arguments=arguments)
dataset_filepaths, dataset_brat_folders = get_valid_dataset_filepaths(
parameters)
check_parameter_compatiblity(parameters, dataset_filepaths)
# Load dataset
dataset = ds.Dataset(verbose=parameters['verbose'],
debug=parameters['debug'])
dataset.load_dataset(dataset_filepaths, parameters)
# Create graph and session
with tf.device('/gpu:0'):
with tf.Graph().as_default():
session_conf = tf.ConfigProto(
intra_op_parallelism_threads=parameters[
'number_of_cpu_threads'],
inter_op_parallelism_threads=parameters[
'number_of_cpu_threads'],
device_count={
'CPU': 1,
'GPU': parameters['number_of_gpus']
},
allow_soft_placement=True,
log_device_placement=False)
sess = tf.Session(config=session_conf)
with sess.as_default():
start_time = time.time()
experiment_timestamp = utils.get_current_time_in_miliseconds()
results = {}
results['epoch'] = {}
results['execution_details'] = {}
results['execution_details']['train_start'] = start_time
results['execution_details'][
'time_stamp'] = experiment_timestamp
results['execution_details']['early_stop'] = False
results['execution_details']['keyboard_interrupt'] = False
results['execution_details']['num_epochs'] = 0
results['model_options'] = copy.copy(parameters)
dataset_name = utils.get_basename_without_extension(
parameters['dataset_text_folder'])
model_name = dataset_name
utils.create_folder_if_not_exists(parameters['output_folder'])
stats_graph_folder = os.path.join(
parameters['output_folder'],
model_name) # Folder where to save graphs
final_weights_folder = os.path.join(
parameters['output_folder'], 'weights')
utils.create_folder_if_not_exists(stats_graph_folder)
utils.create_folder_if_not_exists(final_weights_folder)
model_folder = os.path.join(stats_graph_folder, 'model')
utils.create_folder_if_not_exists(model_folder)
with open(os.path.join(model_folder, 'parameters.ini'),
'w') as parameters_file:
conf_parameters.write(parameters_file)
tensorboard_log_folder = os.path.join(stats_graph_folder,
'tensorboard_logs')
utils.create_folder_if_not_exists(tensorboard_log_folder)
tensorboard_log_folders = {}
for dataset_type in dataset_filepaths.keys():
tensorboard_log_folders[dataset_type] = os.path.join(
stats_graph_folder, 'tensorboard_logs', dataset_type)
utils.create_folder_if_not_exists(
tensorboard_log_folders[dataset_type])
pickle.dump(
dataset,
open(os.path.join(model_folder, 'dataset.pickle'), 'wb'))
model = EntityLSTM(dataset, parameters)
writers = {}
for dataset_type in dataset_filepaths.keys():
writers[dataset_type] = tf.summary.FileWriter(
tensorboard_log_folders[dataset_type],
graph=sess.graph)
embedding_writer = tf.summary.FileWriter(model_folder)
embeddings_projector_config = projector.ProjectorConfig()
tensorboard_token_embeddings = embeddings_projector_config.embeddings.add(
)
tensorboard_token_embeddings.tensor_name = model.token_embedding_weights.name
token_list_file_path = os.path.join(
model_folder, 'tensorboard_metadata_tokens.tsv')
tensorboard_token_embeddings.metadata_path = os.path.relpath(
token_list_file_path, '..')
tensorboard_character_embeddings = embeddings_projector_config.embeddings.add(
)
tensorboard_character_embeddings.tensor_name = model.character_embedding_weights.name
character_list_file_path = os.path.join(
model_folder, 'tensorboard_metadata_characters.tsv')
tensorboard_character_embeddings.metadata_path = os.path.relpath(
character_list_file_path, '..')
projector.visualize_embeddings(embedding_writer,
embeddings_projector_config)
token_list_file = codecs.open(token_list_file_path, 'w',
'latin-1')
for token_index in range(dataset.vocabulary_size):
token_list_file.write('{0}\n'.format(
dataset.index_to_token[token_index]))
token_list_file.close()
character_list_file = codecs.open(character_list_file_path,
'w', 'latin-1')
for character_index in range(dataset.alphabet_size):
if character_index == dataset.PADDING_CHARACTER_INDEX:
character_list_file.write('PADDING\n')
else:
character_list_file.write('{0}\n'.format(
dataset.index_to_character[character_index]))
character_list_file.close()
# Initialize the model
sess.run(tf.global_variables_initializer())
if not parameters['use_pretrained_model']:
model.load_pretrained_token_embeddings(
sess, dataset, parameters)
patience_counter = 0 # number of epochs with no improvement on the validation test in terms of F1-score
f1_score_best = 0
f1_scores = {'train-F1': [], 'valid-F1': [], 'test-F1': []}
transition_params_trained = np.random.rand(
len(dataset.unique_labels) + 2,
len(dataset.unique_labels) + 2)
model_saver = tf.train.Saver(
max_to_keep=parameters['num_of_model_to_keep']
) #, reshape= True) # defaults to saving all variables
epoch_number = -1
try:
while True:
step = 0
epoch_number += 1
print('\nStarting epoch {0}'.format(epoch_number))
epoch_start_time = time.time()
if parameters[
'use_pretrained_model'] and epoch_number == 0:
if parameters['use_corrector']:
parameters['use_corrector'] = False
transition_params_trained = train.restore_pretrained_model(
parameters, dataset, sess, model,
model_saver)
print(
'Getting the 3-label predictions from the step1 model.'
)
all_pred_labels, y_pred_for_corrector, y_true_for_corrector, \
output_filepaths = train.predict_labels(sess, model,
transition_params_trained,
parameters, dataset,
epoch_number,
stats_graph_folder,
dataset_filepaths,
for_corrector = True)
all_pred_indices = {} #defaultdict(list)
for dataset_type in dataset_filepaths.keys():
all_pred_indices[dataset_type] = []
for i in range(
len(all_pred_labels[dataset_type])
):
indices = [
dataset.
label_corrector_to_index[label]
for label in
all_pred_labels[dataset_type][i]
]
all_pred_indices[dataset_type].append(
indices)
label_binarizer_corrector = sklearn.preprocessing.LabelBinarizer(
)
label_binarizer_corrector.fit(
range(
max(dataset.index_to_label_corrector.
keys()) + 1))
predicted_label_corrector_vector_indices = {}
for dataset_type in dataset_filepaths.keys():
predicted_label_corrector_vector_indices[
dataset_type] = []
for label_indices_sequence in all_pred_indices[
dataset_type]:
predicted_label_corrector_vector_indices[
dataset_type].append(
label_binarizer_corrector.
transform(
label_indices_sequence))
parameters['use_corrector'] = True
transition_params_trained, model, glo_step = \
train.restore_model_parameters_from_pretrained_model(parameters, dataset, sess, model, model_saver)
for dataset_type in dataset_filepaths.keys():
writers[dataset_type] = tf.summary.FileWriter(
tensorboard_log_folders[dataset_type],
graph=sess.graph)
embedding_writer = tf.summary.FileWriter(
model_folder)
init_new_vars_op = tf.initialize_variables(
[glo_step])
sess.run(init_new_vars_op)
elif epoch_number != 0:
sequence_numbers = list(
range(len(dataset.token_indices['train'])))
random.shuffle(sequence_numbers)
for sequence_number in sequence_numbers:
transition_params_trained, W_before_crf = train.train_step(
sess, dataset, sequence_number, model,
transition_params_trained, parameters)
step += 1
epoch_elapsed_training_time = time.time(
) - epoch_start_time
print('Training completed in {0:.2f} seconds'.format(
epoch_elapsed_training_time),
flush=False)
if parameters['use_corrector']:
original_label_corrector_vector_indices = dataset.label_corrector_vector_indices
dataset.label_corrector_vector_indices = predicted_label_corrector_vector_indices
y_pred, y_true, output_filepaths = train.predict_labels(
sess, model, transition_params_trained,
parameters, dataset, epoch_number,
stats_graph_folder, dataset_filepaths)
# Evaluate model: save and plot results
evaluate.evaluate_model(results, dataset, y_pred,
y_true, stats_graph_folder,
epoch_number,
epoch_start_time,
output_filepaths,
parameters)
dataset.label_corrector_vector_indices = original_label_corrector_vector_indices
else:
y_pred, y_true, output_filepaths = train.predict_labels(
sess, model, transition_params_trained,
parameters, dataset, epoch_number,
stats_graph_folder, dataset_filepaths)
# Evaluate model: save and plot results
evaluate.evaluate_model(results, dataset, y_pred,
y_true, stats_graph_folder,
epoch_number,
epoch_start_time,
output_filepaths,
parameters)
summary = sess.run(model.summary_op, feed_dict=None)
writers['train'].add_summary(summary, epoch_number)
writers['train'].flush()
utils.copytree(writers['train'].get_logdir(),
model_folder)
# Early stopping
train_f1_score = results['epoch'][epoch_number][0][
'train']['f1_score']['micro']
valid_f1_score = results['epoch'][epoch_number][0][
'valid']['f1_score']['micro']
test_f1_score = results['epoch'][epoch_number][0][
'test']['f1_score']['micro']
f1_scores['train-F1'].append(train_f1_score)
f1_scores['valid-F1'].append(valid_f1_score)
f1_scores['test-F1'].append(test_f1_score)
if valid_f1_score > f1_score_best:
patience_counter = 0
f1_score_best = valid_f1_score
# Save the best model
model_saver.save(
sess,
os.path.join(model_folder, 'best_model.ckpt'))
print(
'updated model to current epoch : epoch {:d}'.
format(epoch_number))
print('the model is saved in: {:s}'.format(
model_folder))
### newly deleted
else:
patience_counter += 1
print("In epoch {:d}, the valid F1 is : {:f}".format(
epoch_number, valid_f1_score))
print(
"The last {0} epochs have not shown improvements on the validation set."
.format(patience_counter))
if patience_counter >= parameters['patience']:
print('Early Stop!')
results['execution_details']['early_stop'] = True
if epoch_number >= parameters[
'maximum_number_of_epochs'] and parameters[
'refine_with_crf']:
model = train.refine_with_crf(
parameters, sess, model, model_saver)
print('refine model with CRF ...')
for additional_epoch in range(
parameters['additional_epochs_with_crf']):
print('Additional {:d}th epoch'.format(
additional_epoch))
sequence_numbers = list(
range(len(dataset.token_indices['train'])))
random.shuffle(sequence_numbers)
for sequence_number in sequence_numbers:
transition_params_trained, W_before_crf = train.train_step(
sess, dataset, sequence_number, model,
transition_params_trained, parameters)
step += 1
epoch_elapsed_training_time = time.time(
) - epoch_start_time
print(
'Additional training completed in {0:.2f} seconds'
.format(epoch_elapsed_training_time),
flush=False)
y_pred, y_true, output_filepaths = train.predict_labels(
sess, model, transition_params_trained,
parameters, dataset, epoch_number,
stats_graph_folder, dataset_filepaths)
evaluate.evaluate_model(
results, dataset, y_pred, y_true,
stats_graph_folder, epoch_number,
epoch_start_time, output_filepaths,
parameters)
summary = sess.run(model.summary_op,
feed_dict=None)
writers['train'].add_summary(
summary, epoch_number)
writers['train'].flush()
utils.copytree(writers['train'].get_logdir(),
model_folder)
if epoch_number >= parameters[
'maximum_number_of_epochs'] and not parameters[
'refine_with_crf']:
break
if not parameters['use_pretrained_model']:
plot_name = 'F1-summary-step1.svg'
else:
plot_name = 'F1-summary-step2.svg'
for k, l in f1_scores.items():
print(k, l)
utils_plots.plot_f1(
f1_scores,
os.path.join(stats_graph_folder, '..', plot_name),
'F1 score summary')
except KeyboardInterrupt:
results['execution_details']['keyboard_interrupt'] = True
print('Training interrupted')
print('Finishing the experiment')
end_time = time.time()
results['execution_details'][
'train_duration'] = end_time - start_time
results['execution_details']['train_end'] = end_time
evaluate.save_results(results, stats_graph_folder)
for dataset_type in dataset_filepaths.keys():
writers[dataset_type].close()
sess.close()
def main(languages):
#embeddings_type = ['polyglot', 'fasttext']
#embeddings_type = ['fasttext', 'fasttext_noOOV']
embeddings_type = ['fasttext_noOOV']
character_lstm = [True]
embedding_language = ['target', 'source']
combination = product(languages, embeddings_type, embedding_language, character_lstm)
create_folder_if_not_exists(os.path.join("..", "log"))
experiment_timestamp = utils.get_current_time_in_miliseconds()
log_file = os.path.join("..", "log", "experiment-{}.log".format(experiment_timestamp))
for language, emb_type, emb_language, char_lstm in combination:
conf_parameters = load_parameters()
conf_parameters = set_datasets(conf_parameters, language)
conf_parameters.set('ann','use_character_lstm', str(char_lstm))
conf_parameters.set('ann','embedding_type', emb_type)
conf_parameters.set('ann','embedding_language', emb_language)
if emb_type == 'polyglot':
conf_parameters.set('ann', 'embedding_dimension', str(64))
elif 'fasttext' in emb_type:
conf_parameters.set('ann', 'embedding_dimension', str(300))
else:
raise("Uknown embedding type")
if emb_language == 'source':
conf_parameters.set('dataset', 'language', constants.MAPPING_LANGUAGE[language])
else:
conf_parameters.set('dataset', 'language', language)
parameters, conf_parameters = parse_parameters(conf_parameters)
start_time = time.time()
experiment_timestamp = utils.get_current_time_in_miliseconds()
results = {}
results['epoch'] = {}
results['execution_details'] = {}
results['execution_details']['train_start'] = start_time
results['execution_details']['time_stamp'] = experiment_timestamp
results['execution_details']['early_stop'] = False
results['execution_details']['keyboard_interrupt'] = False
results['execution_details']['num_epochs'] = 0
results['model_options'] = copy.copy(parameters)
dataset_name = utils.get_basename_without_extension(parameters['dataset_train'])
model_name = '{0}_{1}_{2}_{3}_{4}'.format(language, emb_type, char_lstm, emb_language,
results['execution_details']['time_stamp'])
sys.stdout = open(os.path.join("..", "log", model_name), "w")
print(language, emb_type, char_lstm, emb_language)
with open(log_file, "a") as file:
file.write("Experiment: {}\n".format(model_name))
file.write("Start time:{}\n".format(experiment_timestamp))
file.write("-------------------------------------\n\n")
pprint(parameters)
dataset_filepaths = get_valid_dataset_filepaths(parameters)
check_parameter_compatiblity(parameters, dataset_filepaths)
previous_best_valid_epoch = -1
# Load dataset
dataset = ds.Dataset(verbose=parameters['verbose'], debug=parameters['debug'])
dataset.load_vocab_word_embeddings(parameters)
dataset.load_dataset(dataset_filepaths, parameters)
# Create graph and session
with tf.Graph().as_default():
session_conf = tf.ConfigProto(
intra_op_parallelism_threads=parameters['number_of_cpu_threads'],
inter_op_parallelism_threads=parameters['number_of_cpu_threads'],
device_count={'CPU': 1, 'GPU': parameters['number_of_gpus']},
allow_soft_placement=True,
# automatically choose an existing and supported device to run the operations in case the specified one doesn't exist
log_device_placement=False
)
session_conf.gpu_options.allow_growth = True
sess = tf.Session(config=session_conf)
with sess.as_default():
# Initialize and save execution details
print(model_name)
output_folder = os.path.join('..', 'output')
utils.create_folder_if_not_exists(output_folder)
stats_graph_folder = os.path.join(output_folder, model_name) # Folder where to save graphs
utils.create_folder_if_not_exists(stats_graph_folder)
model_folder = os.path.join(stats_graph_folder, 'model')
utils.create_folder_if_not_exists(model_folder)
with open(os.path.join(model_folder, 'parameters.ini'), 'w') as parameters_file:
conf_parameters.write(parameters_file)
tensorboard_log_folder = os.path.join(stats_graph_folder, 'tensorboard_logs')
utils.create_folder_if_not_exists(tensorboard_log_folder)
tensorboard_log_folders = {}
for dataset_type in dataset_filepaths.keys():
tensorboard_log_folders[dataset_type] = os.path.join(stats_graph_folder, 'tensorboard_logs',
dataset_type)
utils.create_folder_if_not_exists(tensorboard_log_folders[dataset_type])
# del dataset.embeddings_matrix
if not parameters['use_pretrained_model']:
pickle.dump(dataset, open(os.path.join(model_folder, 'dataset.pickle'), 'wb'))
# dataset.load_pretrained_word_embeddings(parameters)
# Instantiate the model
# graph initialization should be before FileWriter, otherwise the graph will not appear in TensorBoard
model = EntityLSTM(dataset, parameters)
# Instantiate the writers for TensorBoard
writers = {}
for dataset_type in dataset_filepaths.keys():
writers[dataset_type] = tf.summary.FileWriter(tensorboard_log_folders[dataset_type],
graph=sess.graph)
embedding_writer = tf.summary.FileWriter(
model_folder) # embedding_writer has to write in model_folder, otherwise TensorBoard won't be able to view embeddings
embeddings_projector_config = projector.ProjectorConfig()
tensorboard_token_embeddings = embeddings_projector_config.embeddings.add()
tensorboard_token_embeddings.tensor_name = model.token_embedding_weights.name
token_list_file_path = os.path.join(model_folder, 'tensorboard_metadata_tokens.tsv')
tensorboard_token_embeddings.metadata_path = os.path.relpath(token_list_file_path, '..')
if parameters['use_character_lstm']:
tensorboard_character_embeddings = embeddings_projector_config.embeddings.add()
tensorboard_character_embeddings.tensor_name = model.character_embedding_weights.name
character_list_file_path = os.path.join(model_folder, 'tensorboard_metadata_characters.tsv')
tensorboard_character_embeddings.metadata_path = os.path.relpath(character_list_file_path, '..')
projector.visualize_embeddings(embedding_writer, embeddings_projector_config)
# Write metadata for TensorBoard embeddings
token_list_file = codecs.open(token_list_file_path, 'w', 'UTF-8')
for token_index in range(len(dataset.index_to_token)):
token_list_file.write('{0}\n'.format(dataset.index_to_token[token_index]))
token_list_file.close()
if parameters['use_character_lstm']:
character_list_file = codecs.open(character_list_file_path, 'w', 'UTF-8')
for character_index in range(dataset.alphabet_size):
if character_index == dataset.PADDING_CHARACTER_INDEX:
character_list_file.write('PADDING\n')
else:
character_list_file.write('{0}\n'.format(dataset.index_to_character[character_index]))
character_list_file.close()
try:
# Initialize the model
sess.run(tf.global_variables_initializer())
if not parameters['use_pretrained_model']:
model.load_pretrained_token_embeddings(sess, dataset, parameters)
# Start training + evaluation loop. Each iteration corresponds to 1 epoch.
bad_counter = 0 # number of epochs with no improvement on the validation test in terms of F1-score
previous_best_valid_f1_score = -1
transition_params_trained = np.random.rand(len(dataset.unique_labels), len(
dataset.unique_labels)) # TODO np.random.rand(len(dataset.unique_labels)+2,len(dataset.unique_labels)+2)
model_saver = tf.train.Saver(
max_to_keep=None) # parameters['maximum_number_of_epochs']) # defaults to saving all variables
epoch_number = 0
while True:
step = 0
epoch_number += 1
print('\nStarting epoch {0}'.format(epoch_number))
epoch_start_time = time.time()
if parameters['use_pretrained_model'] and epoch_number == 1:
# Restore pretrained model parameters
transition_params_trained = train.restore_model_parameters_from_pretrained_model(parameters,
dataset,
sess,
model,
model_saver)
elif epoch_number != 0:
# Train model: loop over all sequences of training set with shuffling
sequence_numbers = list(range(len(dataset.token_indices['train'])))
random.shuffle(sequence_numbers)
data_counter = 0
sub_id = 0
for i in tqdm(range(0, len(sequence_numbers), parameters['batch_size']), "Training epoch {}".format(epoch_number),
mininterval=1):
data_counter += parameters['batch_size']
if data_counter >= 20000:
data_counter = 0
sub_id += 0.001
print("Intermediate evaluation number: ", sub_id)
epoch_elapsed_training_time = time.time() - epoch_start_time
print('Training completed in {0:.2f} seconds'.format(epoch_elapsed_training_time),
flush=True)
y_pred, y_true, output_filepaths = train.predict_labels(sess, model,
transition_params_trained,
parameters, dataset,
epoch_number + sub_id,
stats_graph_folder,
dataset_filepaths)
# Evaluate model: save and plot results
evaluate.evaluate_model(results, dataset, y_pred, y_true, stats_graph_folder,
epoch_number, epoch_start_time, output_filepaths,
parameters)
# Save model
model_saver.save(sess, os.path.join(model_folder,
'model_{0:07.3f}.ckpt'.format(
epoch_number + sub_id)))
# Save TensorBoard logs
summary = sess.run(model.summary_op, feed_dict=None)
writers['train'].add_summary(summary, epoch_number)
writers['train'].flush()
utils.copytree(writers['train'].get_logdir(), model_folder)
# Early stop
valid_f1_score = results['epoch'][epoch_number][0]['valid']['f1_score']['micro']
if valid_f1_score > previous_best_valid_f1_score:
bad_counter = 0
previous_best_valid_f1_score = valid_f1_score
else:
bad_counter += 1
sequence_number = sequence_numbers[i: i + parameters['batch_size']]
transition_params_trained, loss = train.train_step(sess, dataset, sequence_number,
model, transition_params_trained,
parameters)
epoch_elapsed_training_time = time.time() - epoch_start_time
print('Training completed in {0:.2f} seconds'.format(epoch_elapsed_training_time), flush=True)
y_pred, y_true, output_filepaths = train.predict_labels(sess, model, transition_params_trained,
parameters, dataset, epoch_number,
stats_graph_folder, dataset_filepaths)
# Evaluate model: save and plot results
evaluate.evaluate_model(results, dataset, y_pred, y_true, stats_graph_folder, epoch_number,
epoch_start_time, output_filepaths, parameters)
# Save model
model_saver.save(sess, os.path.join(model_folder, 'model_{0:05d}.ckpt'.format(epoch_number)))
# Save TensorBoard logs
summary = sess.run(model.summary_op, feed_dict=None)
writers['train'].add_summary(summary, epoch_number)
writers['train'].flush()
utils.copytree(writers['train'].get_logdir(), model_folder)
# Early stop
valid_f1_score = results['epoch'][epoch_number][0]['valid']['f1_score']['micro']
if valid_f1_score > previous_best_valid_f1_score:
bad_counter = 0
previous_best_valid_f1_score = valid_f1_score
previous_best_valid_epoch = epoch_number
else:
bad_counter += 1
print("The last {0} epochs have not shown improvements on the validation set.".format(
bad_counter))
if bad_counter >= parameters['patience']:
print('Early Stop!')
results['execution_details']['early_stop'] = True
break
if epoch_number >= parameters['maximum_number_of_epochs']: break
keep_only_best_model(model_folder,previous_best_valid_epoch ,parameters['maximum_number_of_epochs']+1)
except KeyboardInterrupt:
results['execution_details']['keyboard_interrupt'] = True
print('Training interrupted')
# remove the experiment
remove_experiment = input("Do you want to remove the experiment? (yes/y/Yes)")
if remove_experiment in ["Yes", "yes", "y"]:
shutil.rmtree(stats_graph_folder)
print("Folder removed")
else:
print('Finishing the experiment')
end_time = time.time()
results['execution_details']['train_duration'] = end_time - start_time
results['execution_details']['train_end'] = end_time
evaluate.save_results(results, stats_graph_folder)
sys.stdout.close()
except Exception:
logging.exception("")
remove_experiment = input("Do you want to remove the experiment? (yes/y/Yes)")
if remove_experiment in ["Yes", "yes", "y"]:
shutil.rmtree(stats_graph_folder)
print("Folder removed")
sys.stdout.close()
sess.close() # release the session's resources
sys.stdout.close()
def _get_valid_dataset_filepaths(self, parameters, dataset_types=['train', 'valid', 'test', 'deploy']):
'''
Tiền xử lý dataset đầu vào, nếu data chuẩn conll thì chuyển sang brat
Tham số:
- parameters: parameters của toàn bộ chương trình
Return:
(
{ // dataset_filepaths các fields bên dưới là optional, không nhất thiết phải đủ 4
"train": "data_text_folder/train[_compatible_with_brat][_bioes].txt",
"valid": "data_text_folder/valid[_compatible_with_brat][_bioes].txt",
"test": "data_text_folder/test[_compatible_with_brat][_bioes].txt",
"deploy": "data_text_folder/deploy[_compatible_with_brat][_bioes].txt"
},
{ // dataset_brat_folders, các fields bên dưới là optional, không nhất thiết phải đủ 4
"train": "data_text_folder/train",
"valid": "data_text_folder/valid",
"test": "data_text_folder/test",
"deploy": "data_text_folder/deploy"
}
)
'''
dataset_filepaths = {}
dataset_brat_folders = {}
for dataset_type in dataset_types:
dataset_filepaths[dataset_type] = os.path.join(parameters['dataset_text_folder'], '{0}.txt'.format(dataset_type))
dataset_brat_folders[dataset_type] = os.path.join(parameters['dataset_text_folder'], dataset_type)
dataset_compatible_with_brat_filepath = os.path.join(parameters['dataset_text_folder'], '{0}_compatible_with_brat.txt'.format(dataset_type))
# Conll file exists
if os.path.isfile(dataset_filepaths[dataset_type]) and os.path.getsize(dataset_filepaths[dataset_type]) > 0:
# Brat text files exist
if os.path.exists(dataset_brat_folders[dataset_type]) and len(glob.glob(os.path.join(dataset_brat_folders[dataset_type], '*.txt'))) > 0:
# Check compatibility between conll and brat files
brat_to_conll.check_brat_annotation_and_text_compatibility(dataset_brat_folders[dataset_type])
if os.path.exists(dataset_compatible_with_brat_filepath):
dataset_filepaths[dataset_type] = dataset_compatible_with_brat_filepath
conll_to_brat.check_compatibility_between_conll_and_brat_text(dataset_filepaths[dataset_type], dataset_brat_folders[dataset_type])
# Brat text files do not exist
else:
# Populate brat text and annotation files based on conll file
conll_to_brat.conll_to_brat(dataset_filepaths[dataset_type], dataset_compatible_with_brat_filepath, dataset_brat_folders[dataset_type], dataset_brat_folders[dataset_type])
dataset_filepaths[dataset_type] = dataset_compatible_with_brat_filepath
# Conll file does not exist
else:
# Brat text files exist
if os.path.exists(dataset_brat_folders[dataset_type]) and len(glob.glob(os.path.join(dataset_brat_folders[dataset_type], '*.txt'))) > 0:
dataset_filepath_for_tokenizer = os.path.join(parameters['dataset_text_folder'], '{0}_{1}.txt'.format(dataset_type, parameters['tokenizer']))
if os.path.exists(dataset_filepath_for_tokenizer):
conll_to_brat.check_compatibility_between_conll_and_brat_text(dataset_filepath_for_tokenizer, dataset_brat_folders[dataset_type])
else:
# Populate conll file based on brat files
brat_to_conll.brat_to_conll(dataset_brat_folders[dataset_type], dataset_filepath_for_tokenizer, parameters['tokenizer'], parameters['spacylanguage'])
dataset_filepaths[dataset_type] = dataset_filepath_for_tokenizer
# Brat text files do not exist
else:
del dataset_filepaths[dataset_type]
del dataset_brat_folders[dataset_type]
continue
if parameters['tagging_format'] == 'bioes':
# Generate conll file with BIOES format
bioes_filepath = os.path.join(parameters['dataset_text_folder'], '{0}_bioes.txt'.format(utils.get_basename_without_extension(dataset_filepaths[dataset_type])))
utils_nlp.convert_conll_from_bio_to_bioes(dataset_filepaths[dataset_type], bioes_filepath)
dataset_filepaths[dataset_type] = bioes_filepath
return dataset_filepaths, dataset_brat_folders
def main(argv=sys.argv):
''' NeuroNER main method
Args:
parameters_filepath the path to the parameters file
output_folder the path to the output folder
'''
arguments = parse_arguments(argv[1:])
parameters, conf_parameters = load_parameters(
arguments['parameters_filepath'], arguments=arguments)
dataset_filepaths, dataset_brat_folders = get_valid_dataset_filepaths(
parameters)
check_parameter_compatiblity(parameters, dataset_filepaths)
# Load dataset
dataset = ds.Dataset(verbose=parameters['verbose'],
debug=parameters['debug'])
dataset.load_dataset(dataset_filepaths, parameters)
# Create graph and session
with tf.device('/gpu:0'):
with tf.Graph().as_default():
session_conf = tf.ConfigProto(
intra_op_parallelism_threads=parameters[
'number_of_cpu_threads'],
inter_op_parallelism_threads=parameters[
'number_of_cpu_threads'],
device_count={
'CPU': 1,
'GPU': parameters['number_of_gpus']
},
allow_soft_placement=True,
# automatically choose an existing and supported device to run the operations in case the specified one doesn't exist
log_device_placement=False)
sess = tf.Session(config=session_conf)
with sess.as_default():
start_time = time.time()
experiment_timestamp = utils.get_current_time_in_miliseconds()
results = {}
results['epoch'] = {}
results['execution_details'] = {}
results['execution_details']['train_start'] = start_time
results['execution_details'][
'time_stamp'] = experiment_timestamp
results['execution_details']['early_stop'] = False
results['execution_details']['keyboard_interrupt'] = False
results['execution_details']['num_epochs'] = 0
results['model_options'] = copy.copy(parameters)
dataset_name = utils.get_basename_without_extension(
parameters['dataset_text_folder'])
model_name = dataset_name
utils.create_folder_if_not_exists(parameters['output_folder'])
stats_graph_folder = os.path.join(
parameters['output_folder'],
model_name) # Folder where to save graphs
final_weights_folder = os.path.join(
parameters['output_folder'], 'weights')
utils.create_folder_if_not_exists(stats_graph_folder)
utils.create_folder_if_not_exists(final_weights_folder)
model_folder = os.path.join(stats_graph_folder, 'model')
utils.create_folder_if_not_exists(model_folder)
# saving the parameter setting to the output model dir. For later resuming training
with open(os.path.join(model_folder, 'parameters.ini'),
'w') as parameters_file:
conf_parameters.write(parameters_file)
tensorboard_log_folder = os.path.join(stats_graph_folder,
'tensorboard_logs')
utils.create_folder_if_not_exists(tensorboard_log_folder)
tensorboard_log_folders = {}
for dataset_type in dataset_filepaths.keys():
tensorboard_log_folders[dataset_type] = os.path.join(
stats_graph_folder, 'tensorboard_logs', dataset_type)
utils.create_folder_if_not_exists(
tensorboard_log_folders[dataset_type])
pickle.dump(
dataset,
open(os.path.join(model_folder, 'dataset.pickle'), 'wb'))
# Instantiate the model
# graph initialization should be before FileWriter, otherwise the graph will not appear in TensorBoard
model = EntityLSTM(dataset, parameters)
# Instantiate the writers for TensorBoard
writers = {}
for dataset_type in dataset_filepaths.keys():
writers[dataset_type] = tf.summary.FileWriter(
tensorboard_log_folders[dataset_type],
graph=sess.graph)
# embedding_writer has to write in model_folder, otherwise TensorBoard won't be able to view embeddings
embedding_writer = tf.summary.FileWriter(model_folder)
embeddings_projector_config = projector.ProjectorConfig()
tensorboard_token_embeddings = embeddings_projector_config.embeddings.add(
)
tensorboard_token_embeddings.tensor_name = model.token_embedding_weights.name
token_list_file_path = os.path.join(
model_folder, 'tensorboard_metadata_tokens.tsv')
tensorboard_token_embeddings.metadata_path = os.path.relpath(
token_list_file_path, '..')
tensorboard_character_embeddings = embeddings_projector_config.embeddings.add(
)
tensorboard_character_embeddings.tensor_name = model.character_embedding_weights.name
character_list_file_path = os.path.join(
model_folder, 'tensorboard_metadata_characters.tsv')
tensorboard_character_embeddings.metadata_path = os.path.relpath(
character_list_file_path, '..')
projector.visualize_embeddings(embedding_writer,
embeddings_projector_config)
# Write metadata for TensorBoard embeddings
token_list_file = codecs.open(token_list_file_path, 'w',
'latin-1')
for token_index in range(dataset.vocabulary_size):
token_list_file.write('{0}\n'.format(
dataset.index_to_token[token_index]))
token_list_file.close()
character_list_file = codecs.open(character_list_file_path,
'w', 'latin-1')
for character_index in range(dataset.alphabet_size):
if character_index == dataset.PADDING_CHARACTER_INDEX:
character_list_file.write('PADDING\n')
else:
character_list_file.write('{0}\n'.format(
dataset.index_to_character[character_index]))
character_list_file.close()
# Initialize the model
sess.run(tf.global_variables_initializer())
if not parameters['use_pretrained_model']:
model.load_pretrained_token_embeddings(
sess, dataset, parameters)
# Start training + evaluation loop. Each iteration corresponds to 1 epoch.
patience_counter = 0
f1_score_best = 0
f1_scores = {'train-F1': [], 'valid-F1': [], 'test-F1': []}
f1_scores_conll = {
'train-F1': [],
'valid-F1': [],
'test-F1': []
}
transition_params_trained = np.random.rand(
len(dataset.unique_labels) + 2,
len(dataset.unique_labels) + 2)
model_saver = tf.train.Saver(
max_to_keep=parameters['num_of_model_to_keep'])
epoch_number = -1
try:
while True:
step = 0
epoch_number += 1
print('\nStarting epoch {0}'.format(epoch_number))
epoch_start_time = time.time()
# use pre-trained model and epoch_number = 0
if parameters[
'use_pretrained_model'] and epoch_number == 0:
if parameters['use_adapter']:
parameters['use_adapter'] = False
transition_params_trained = train.restore_pretrained_model(
parameters, dataset, sess, model,
model_saver)
print(
'Getting the 3-label predictions from the step1 model.'
)
all_pred_labels, y_pred_for_adapter, y_true_for_adapter, \
output_filepaths = train.predict_labels(sess, model,
transition_params_trained,
parameters, dataset,
epoch_number,
stats_graph_folder,
dataset_filepaths,
for_adapter=True)
# use the label2idx mapping (for adapter) in the dataset to transform all_pred_labels
all_pred_indices = {}
for dataset_type in dataset_filepaths.keys():
all_pred_indices[dataset_type] = []
for i in range(
len(all_pred_labels[dataset_type])
):
indices = [
dataset.
label_adapter_to_index[label]
for label in
all_pred_labels[dataset_type][i]
]
all_pred_indices[dataset_type].append(
indices)
# and use binarizer to transform to ndarray
label_binarizer_adapter = sklearn.preprocessing.LabelBinarizer(
)
label_binarizer_adapter.fit(
range(
max(dataset.index_to_label_adapter.
keys()) + 1))
predicted_label_adapter_vector_indices = {}
for dataset_type in dataset_filepaths.keys():
predicted_label_adapter_vector_indices[
dataset_type] = []
for label_indices_sequence in all_pred_indices[
dataset_type]:
predicted_label_adapter_vector_indices[
dataset_type].append(
label_binarizer_adapter.
transform(
label_indices_sequence))
parameters['use_adapter'] = True
if parameters['train_model'] and parameters[
'add_class']:
transition_params_trained, model, glo_step = \
train.restore_model_parameters_from_pretrained_model(parameters, dataset, sess,
model, model_saver)
init_new_vars_op = tf.initialize_variables(
[glo_step])
sess.run(init_new_vars_op)
else:
transition_params_trained = \
train.restore_pretrained_model(parameters, dataset, sess, model, model_saver)
for dataset_type in dataset_filepaths.keys():
writers[dataset_type] = tf.summary.FileWriter(
tensorboard_log_folders[dataset_type],
graph=sess.graph)
# embedding_writer has to write in model_folder, otherwise TensorBoard won't be able to view embeddings
embedding_writer = tf.summary.FileWriter(
model_folder)
# epoch_number != 0, no matter use or not use pre-trained model
elif epoch_number != 0:
# Train model: loop over all sequences of training set with shuffling
sequence_numbers = list(
range(len(dataset.token_indices['train'])))
random.shuffle(sequence_numbers)
for sequence_number in sequence_numbers:
transition_params_trained, W_before_crf = train.train_step(
sess, dataset, sequence_number, model,
transition_params_trained, parameters)
step += 1
epoch_elapsed_training_time = time.time(
) - epoch_start_time
print('Training completed in {0:.2f} seconds'.format(
epoch_elapsed_training_time),
flush=False)
if parameters[
'use_adapter']: # model evaluation, using adapter
# pass the pred_for_adapter as label_indices vector
original_label_adapter_vector_indices = dataset.label_adapter_vector_indices
dataset.label_adapter_vector_indices = predicted_label_adapter_vector_indices
y_pred, y_true, output_filepaths = train.predict_labels(
sess, model, transition_params_trained,
parameters, dataset, epoch_number,
stats_graph_folder, dataset_filepaths)
evaluate.evaluate_model(results, dataset, y_pred,
y_true, stats_graph_folder,
epoch_number,
epoch_start_time,
output_filepaths,
parameters)
dataset.label_adapter_vector_indices = original_label_adapter_vector_indices
else: # model evaluation, not using adapter
y_pred, y_true, output_filepaths = train.predict_labels(
sess, model, transition_params_trained,
parameters, dataset, epoch_number,
stats_graph_folder, dataset_filepaths)
# Evaluate model: save and plot results
evaluate.evaluate_model(results, dataset, y_pred,
y_true, stats_graph_folder,
epoch_number,
epoch_start_time,
output_filepaths,
parameters)
summary = sess.run(model.summary_op, feed_dict=None)
writers['train'].add_summary(summary, epoch_number)
writers['train'].flush()
utils.copytree(writers['train'].get_logdir(),
model_folder)
# Early stopping
train_f1_score = results['epoch'][epoch_number][0][
'train']['f1_score']['weighted']
valid_f1_score = results['epoch'][epoch_number][0][
'valid']['f1_score']['weighted']
test_f1_score = results['epoch'][epoch_number][0][
'test']['f1_score']['weighted']
f1_scores['train-F1'].append(train_f1_score)
f1_scores['valid-F1'].append(valid_f1_score)
f1_scores['test-F1'].append(test_f1_score)
train_f1_score_conll = results['epoch'][epoch_number][
0]['train']['f1_conll']['micro']
valid_f1_score_conll = results['epoch'][epoch_number][
0]['valid']['f1_conll']['micro']
test_f1_score_conll = results['epoch'][epoch_number][
0]['test']['f1_conll']['micro']
f1_scores_conll['train-F1'].append(
train_f1_score_conll)
f1_scores_conll['valid-F1'].append(
valid_f1_score_conll)
f1_scores_conll['test-F1'].append(test_f1_score_conll)
if valid_f1_score > f1_score_best:
patience_counter = 0
f1_score_best = valid_f1_score
# Save the best model
model_saver.save(
sess,
os.path.join(model_folder, 'best_model.ckpt'))
print(
'updated model to current epoch : epoch {:d}'.
format(epoch_number))
print('the model is saved in: {:s}'.format(
model_folder))
else:
patience_counter += 1
print("In epoch {:d}, the valid F1 is : {:f}".format(
epoch_number, valid_f1_score))
print(
"The last {0} epochs have not shown improvements on the validation set."
.format(patience_counter))
if patience_counter >= parameters['patience']:
print('Early Stop!')
results['execution_details']['early_stop'] = True
# save last model
model_saver.save(
sess,
os.path.join(model_folder, 'last_model.ckpt'))
print('the last model is saved in: {:s}'.format(
model_folder))
break
if epoch_number >= parameters[
'maximum_number_of_epochs'] and not parameters[
'refine_with_crf']:
break
if not parameters['use_pretrained_model']:
plot_name = 'F1-summary-step1.svg'
else:
plot_name = 'F1-summary-step2.svg'
print('Sklearn result:')
for k, l in f1_scores.items():
print(k, l)
print('Conll result:')
for k, l in f1_scores_conll.items():
print(k, l)
utils_plots.plot_f1(
f1_scores,
os.path.join(stats_graph_folder, '..', plot_name),
'F1 score summary')
# TODO: in step 1, for task a, add the best deploy data to step 2 train set, and call script
print('(sklearn micro) test F1:')
micro_f1 = ','.join([
str(results['epoch'][ep][0]['test']['f1_score']
['micro']) for ep in range(epoch_number + 1)
])
print(micro_f1)
print('(sklearn macro) test F1:')
macro_f1 = ','.join([
str(results['epoch'][ep][0]['test']['f1_score']
['macro']) for ep in range(epoch_number + 1)
])
print(macro_f1)
except KeyboardInterrupt:
results['execution_details']['keyboard_interrupt'] = True
print('Training interrupted')
print('Finishing the experiment')
end_time = time.time()
results['execution_details'][
'train_duration'] = end_time - start_time
results['execution_details']['train_end'] = end_time
evaluate.save_results(results, stats_graph_folder)
for dataset_type in dataset_filepaths.keys():
writers[dataset_type].close()
sess.close() # release the session's resources
def main():
parameters, dataset_filepaths = load_parameters()
# Load dataset
dataset = ds.Dataset()
dataset.load_dataset(dataset_filepaths, parameters)
# Create graph and session
with tf.Graph().as_default():
session_conf = tf.ConfigProto(
device_count={
'CPU': 1,
'GPU': 1
},
allow_soft_placement=
True, # automatically choose an existing and supported device to run the operations in case the specified one doesn't exist
log_device_placement=False)
sess = tf.Session(config=session_conf)
with sess.as_default():
# Initialize and save execution details
start_time = time.time()
experiment_timestamp = utils.get_current_time_in_miliseconds()
results = {}
results['epoch'] = {}
results['execution_details'] = {}
results['execution_details']['train_start'] = start_time
results['execution_details']['time_stamp'] = experiment_timestamp
results['execution_details']['early_stop'] = False
results['execution_details']['keyboard_interrupt'] = False
results['execution_details']['num_epochs'] = 0
results['model_options'] = copy.copy(parameters)
dataset_name = utils.get_basename_without_extension(
parameters['dataset_text_folder'])
model_name = '{0}_{1}'.format(
dataset_name, results['execution_details']['time_stamp'])
output_folder = os.path.join('..', 'output')
utils.create_folder_if_not_exists(output_folder)
stats_graph_folder = os.path.join(
output_folder, model_name) # Folder where to save graphs
utils.create_folder_if_not_exists(stats_graph_folder)
model_folder = os.path.join(stats_graph_folder, 'model')
utils.create_folder_if_not_exists(model_folder)
tensorboard_log_folder = os.path.join(stats_graph_folder,
'tensorboard_logs')
utils.create_folder_if_not_exists(tensorboard_log_folder)
tensorboard_log_folders = {}
for dataset_type in ['train', 'valid', 'test']:
tensorboard_log_folders[dataset_type] = os.path.join(
stats_graph_folder, 'tensorboard_logs', dataset_type)
utils.create_folder_if_not_exists(
tensorboard_log_folders[dataset_type])
pickle.dump(
dataset,
open(os.path.join(stats_graph_folder, 'dataset.pickle'), 'wb'))
# Instantiate the model
# graph initialization should be before FileWriter, otherwise the graph will not appear in TensorBoard
model = EntityLSTM(dataset, parameters)
# Instantiate the writers for TensorBoard
writers = {}
for dataset_type in ['train', 'valid', 'test']:
writers[dataset_type] = tf.summary.FileWriter(
tensorboard_log_folders[dataset_type], graph=sess.graph)
embedding_writer = tf.summary.FileWriter(
model_folder
) # embedding_writer has to write in model_folder, otherwise TensorBoard won't be able to view embeddings
embeddings_projector_config = projector.ProjectorConfig()
tensorboard_token_embeddings = embeddings_projector_config.embeddings.add(
)
tensorboard_token_embeddings.tensor_name = model.token_embedding_weights.name
token_list_file_path = os.path.join(
model_folder, 'tensorboard_metadata_tokens.tsv')
tensorboard_token_embeddings.metadata_path = os.path.relpath(
token_list_file_path, '..')
tensorboard_character_embeddings = embeddings_projector_config.embeddings.add(
)
tensorboard_character_embeddings.tensor_name = model.character_embedding_weights.name
character_list_file_path = os.path.join(
model_folder,
'tensorboard_metadata_characters.tsv') # 'metadata.tsv'
tensorboard_character_embeddings.metadata_path = os.path.relpath(
character_list_file_path, '..')
projector.visualize_embeddings(embedding_writer,
embeddings_projector_config)
# Write metadata for TensorBoard embeddings
token_list_file = open(token_list_file_path, 'w')
for token_index in range(dataset.vocabulary_size):
token_list_file.write('{0}\n'.format(
dataset.index_to_token[token_index]))
token_list_file.close()
character_list_file = open(character_list_file_path, 'w')
print('len(dataset.character_to_index): {0}'.format(
len(dataset.character_to_index)))
print('len(dataset.index_to_character): {0}'.format(
len(dataset.index_to_character)))
for character_index in range(dataset.alphabet_size):
if character_index == dataset.PADDING_CHARACTER_INDEX:
character_list_file.write('PADDING\n')
else:
character_list_file.write('{0}\n'.format(
dataset.index_to_character[character_index]))
character_list_file.close()
# Initialize the model
sess.run(tf.global_variables_initializer())
model.load_pretrained_token_embeddings(sess, dataset, parameters)
# Start training + evaluation loop. Each iteration corresponds to 1 epoch.
step = 0
bad_counter = 0 # number of epochs with no improvement on the validation test in terms of F1-score
previous_best_valid_f1_score = 0
transition_params_trained = np.random.rand(
len(dataset.unique_labels), len(dataset.unique_labels))
model_saver = tf.train.Saver(
max_to_keep=parameters['maximum_number_of_epochs']
) # defaults to saving all variables
epoch_number = -1
try:
while True:
epoch_number += 1
#epoch_number = math.floor(step / len(dataset.token_indices['train']))
print('\nStarting epoch {0}'.format(epoch_number), end='')
epoch_start_time = time.time()
#print('step: {0}'.format(step))
# Train model: loop over all sequences of training set with shuffling
sequence_numbers = list(
range(len(dataset.token_indices['train'])))
random.shuffle(sequence_numbers)
for sequence_number in sequence_numbers:
transition_params_trained = train.train_step(
sess, dataset, sequence_number, model,
transition_params_trained, parameters)
step += 1
if step % 100 == 0:
print('.', end='', flush=True)
#break
print('.', flush=True)
#print('step: {0}'.format(step))
# Predict labels using trained model
y_pred = {}
y_true = {}
output_filepaths = {}
for dataset_type in ['train', 'valid', 'test']:
#print('dataset_type: {0}'.format(dataset_type))
prediction_output = train.prediction_step(
sess, dataset, dataset_type, model,
transition_params_trained, step,
stats_graph_folder, epoch_number, parameters)
y_pred[dataset_type], y_true[
dataset_type], output_filepaths[
dataset_type] = prediction_output
# model_options = None
epoch_elapsed_training_time = time.time(
) - epoch_start_time
print(
'epoch_elapsed_training_time: {0:.2f} seconds'.format(
epoch_elapsed_training_time))
results['execution_details']['num_epochs'] = epoch_number
# Evaluate model: save and plot results
evaluate.evaluate_model(results, dataset, y_pred, y_true,
stats_graph_folder, epoch_number,
epoch_start_time, output_filepaths,
parameters)
# Save model
model_saver.save(
sess,
os.path.join(model_folder,
'model_{0:05d}.ckpt'.format(epoch_number))
) #, global_step, latest_filename, meta_graph_suffix, write_meta_graph, write_state)
# Save TensorBoard logs
summary = sess.run(model.summary_op, feed_dict=None)
writers['train'].add_summary(summary, epoch_number)
# Early stop
valid_f1_score = results['epoch'][epoch_number][0][
'valid']['f1_score']['micro']
if valid_f1_score > previous_best_valid_f1_score:
bad_counter = 0
previous_best_valid_f1_score = valid_f1_score
else:
bad_counter += 1
if bad_counter > parameters['patience']:
print('Early Stop!')
results['execution_details']['early_stop'] = True
break
if epoch_number > parameters['maximum_number_of_epochs']:
break
# break # debugging
except KeyboardInterrupt:
results['execution_details']['keyboard_interrupt'] = True
# assess_model.save_results(results, stats_graph_folder)
print('Training interrupted')
print('Finishing the experiment')
end_time = time.time()
results['execution_details'][
'train_duration'] = end_time - start_time
results['execution_details']['train_end'] = end_time
evaluate.save_results(results, stats_graph_folder)
sess.close() # release the session's resources
def main():
#### Parameters - start
conf_parameters = configparser.ConfigParser()
conf_parameters.read(os.path.join('.', 'parameters.ini'))
nested_parameters = utils.convert_configparser_to_dictionary(
conf_parameters)
parameters = {}
for k, v in nested_parameters.items():
parameters.update(v)
for k, v in parameters.items():
if k in [
'remove_unknown_tokens', 'character_embedding_dimension',
'character_lstm_hidden_state_dimension',
'token_embedding_dimension',
'token_lstm_hidden_state_dimension', 'patience',
'maximum_number_of_epochs', 'maximum_training_time',
'number_of_cpu_threads', 'number_of_gpus'
]:
parameters[k] = int(v)
if k in ['dropout_rate']:
parameters[k] = float(v)
if k in [
'use_character_lstm', 'is_character_lstm_bidirect',
'is_token_lstm_bidirect', 'use_crf'
]:
parameters[k] = distutils.util.strtobool(v)
pprint(parameters)
# Load dataset
dataset_filepaths = {}
dataset_filepaths['train'] = os.path.join(
parameters['dataset_text_folder'], 'train.txt')
dataset_filepaths['valid'] = os.path.join(
parameters['dataset_text_folder'], 'valid.txt')
dataset_filepaths['test'] = os.path.join(parameters['dataset_text_folder'],
'test.txt')
dataset = ds.Dataset()
dataset.load_dataset(dataset_filepaths, parameters)
with tf.Graph().as_default():
session_conf = tf.ConfigProto(
device_count={
'CPU': 1,
'GPU': 1
},
allow_soft_placement=
True, # automatically choose an existing and supported device to run the operations in case the specified one doesn't exist
log_device_placement=False)
sess = tf.Session(config=session_conf)
with sess.as_default():
# Instantiate model
model = EntityLSTM(dataset, parameters)
sess.run(tf.global_variables_initializer())
model.load_pretrained_token_embeddings(sess, dataset, parameters)
# Initialize and save execution details
start_time = time.time()
experiment_timestamp = utils.get_current_time_in_miliseconds()
results = {}
#results['model_options'] = copy.copy(model_options)
#results['model_options'].pop('optimizer', None)
results['epoch'] = {}
results['execution_details'] = {}
results['execution_details']['train_start'] = start_time
results['execution_details']['time_stamp'] = experiment_timestamp
results['execution_details']['early_stop'] = False
results['execution_details']['keyboard_interrupt'] = False
results['execution_details']['num_epochs'] = 0
results['model_options'] = copy.copy(parameters)
dataset_name = utils.get_basename_without_extension(
parameters['dataset_text_folder']
) #opts.train.replace('/', '_').split('.')[0] # 'conll2003en'
model_name = '{0}_{1}'.format(
dataset_name, results['execution_details']['time_stamp'])
output_folder = os.path.join('..', 'output')
utils.create_folder_if_not_exists(output_folder)
stats_graph_folder = os.path.join(
output_folder, model_name) # Folder where to save graphs
#print('stats_graph_folder: {0}'.format(stats_graph_folder))
utils.create_folder_if_not_exists(stats_graph_folder)
# model_folder = os.path.join(stats_graph_folder, 'model')
# utils.create_folder_if_not_exists(model_folder)
step = 0
bad_counter = 0
previous_best_valid_f1_score = 0
transition_params_trained = np.random.rand(
len(dataset.unique_labels), len(dataset.unique_labels))
try:
while True:
epoch_number = math.floor(
step / len(dataset.token_indices['train']))
print('\nStarting epoch {0}'.format(epoch_number), end='')
epoch_start_time = time.time()
#print('step: {0}'.format(step))
# Train model: loop over all sequences of training set with shuffling
sequence_numbers = list(
range(len(dataset.token_indices['train'])))
random.shuffle(sequence_numbers)
for sequence_number in sequence_numbers:
transition_params_trained = train.train_step(
sess, dataset, sequence_number, model,
transition_params_trained, parameters)
step += 1
if step % 100 == 0:
print('.', end='', flush=True)
#break
print('.', flush=True)
#print('step: {0}'.format(step))
# Predict labels using trained model
all_predictions = {}
all_y_true = {}
output_filepaths = {}
for dataset_type in ['train', 'valid', 'test']:
#print('dataset_type: {0}'.format(dataset_type))
prediction_output = train.prediction_step(
sess, dataset, dataset_type, model,
transition_params_trained, step,
stats_graph_folder, epoch_number, parameters)
all_predictions[dataset_type], all_y_true[
dataset_type], output_filepaths[
dataset_type] = prediction_output
# model_options = None
epoch_elapsed_training_time = time.time(
) - epoch_start_time
print(
'epoch_elapsed_training_time: {0:.2f} seconds'.format(
epoch_elapsed_training_time))
results['execution_details']['num_epochs'] = epoch_number
# Evaluate model: save and plot results
evaluate.evaluate_model(results, dataset, all_predictions,
all_y_true, stats_graph_folder,
epoch_number, epoch_start_time,
output_filepaths)
# Early stop
valid_f1_score = results['epoch'][epoch_number][0][
'valid']['f1_score']['micro']
if valid_f1_score > previous_best_valid_f1_score:
bad_counter = 0
previous_best_valid_f1_score = valid_f1_score
else:
bad_counter += 1
if bad_counter > parameters['patience']:
print('Early Stop!')
results['execution_details']['early_stop'] = True
break
if epoch_number > parameters['maximum_number_of_epochs']:
break
# break # debugging
except KeyboardInterrupt:
results['execution_details']['keyboard_interrupt'] = True
# assess_model.save_results(results, stats_graph_folder)
print('Training interrupted')
print('Finishing the experiment')
end_time = time.time()
results['execution_details'][
'train_duration'] = end_time - start_time
results['execution_details']['train_end'] = end_time
evaluate.save_results(results, stats_graph_folder)
sess.close() # release the session's resources
def main():
parameters, conf_parameters = load_parameters()
pprint(parameters)
dataset_filepaths = get_valid_dataset_filepaths(parameters)
check_parameter_compatiblity(parameters, dataset_filepaths)
cross_validation = parameters[
'cross_validation'] if 'cross_validation' in parameters else 1
valid_fscores = []
valid_precisions = []
valid_recalls = []
for cv in range(0, cross_validation):
if "als" in dataset_filepaths['train'] and cross_validation > 1:
train_files = list(range(0, cv)) + list(
range(cv + 1, cross_validation))
test_file = cv
file_train = "tmp_combined.train"
file_valid = "tmp_combined.test"
output = []
for i in train_files:
with open(dataset_filepaths['train'] + "_" + str(i),
"r",
encoding="utf-8") as file:
output.append(file.read())
with open(file_train, "w", encoding="utf-8") as file:
file.write("\n\n".join(output))
output = []
with open(dataset_filepaths['train'] + "_" + str(test_file),
"r",
encoding="utf-8") as file:
output.append(file.read())
with open(file_valid, "w", encoding="utf-8") as file:
file.write("\n\n".join(output))
dataset_filepaths['train'] = file_train
dataset_filepaths['valid'] = file_valid
# Load dataset
dataset = ds.Dataset(verbose=parameters['verbose'],
debug=parameters['debug'])
dataset.load_vocab_word_embeddings(parameters)
dataset.load_dataset(dataset_filepaths, parameters)
# Create graph and session
with tf.Graph().as_default():
session_conf = tf.ConfigProto(
intra_op_parallelism_threads=parameters[
'number_of_cpu_threads'],
inter_op_parallelism_threads=parameters[
'number_of_cpu_threads'],
device_count={
'CPU': 1,
'GPU': parameters['number_of_gpus']
},
allow_soft_placement=
True, # automatically choose an existing and supported device to run the operations in case the specified one doesn't exist
log_device_placement=False)
session_conf.gpu_options.allow_growth = True
sess = tf.Session(config=session_conf)
with sess.as_default():
# Initialize and save execution details
start_time = time.time()
experiment_timestamp = utils.get_current_time_in_miliseconds()
results = {}
results['epoch'] = {}
results['execution_details'] = {}
results['execution_details']['train_start'] = start_time
results['execution_details'][
'time_stamp'] = experiment_timestamp
results['execution_details']['early_stop'] = False
results['execution_details']['keyboard_interrupt'] = False
results['execution_details']['num_epochs'] = 0
results['model_options'] = copy.copy(parameters)
dataset_name = utils.get_basename_without_extension(
parameters['dataset_train'])
if 'data_to_use' in parameters:
model_name = '{0}_{1}'.format(
parameters['language'] + "_" + dataset_name + "_small",
results['execution_details']['time_stamp'])
else:
model_name = '{0}_{1}'.format(
parameters['language'] + "_" + dataset_name,
results['execution_details']['time_stamp'])
output_folder = os.path.join('..', 'output')
utils.create_folder_if_not_exists(output_folder)
stats_graph_folder = os.path.join(
output_folder, model_name) # Folder where to save graphs
utils.create_folder_if_not_exists(stats_graph_folder)
model_folder = os.path.join(stats_graph_folder, 'model')
utils.create_folder_if_not_exists(model_folder)
with open(os.path.join(model_folder, 'parameters.ini'),
'w') as parameters_file:
conf_parameters.write(parameters_file)
tensorboard_log_folder = os.path.join(stats_graph_folder,
'tensorboard_logs')
utils.create_folder_if_not_exists(tensorboard_log_folder)
tensorboard_log_folders = {}
for dataset_type in dataset_filepaths.keys():
tensorboard_log_folders[dataset_type] = os.path.join(
stats_graph_folder, 'tensorboard_logs', dataset_type)
utils.create_folder_if_not_exists(
tensorboard_log_folders[dataset_type])
#del dataset.embeddings_matrix
if not parameters['use_pretrained_model']:
pickle.dump(
dataset,
open(os.path.join(model_folder, 'dataset.pickle'),
'wb'))
#dataset.load_pretrained_word_embeddings(parameters)
# Instantiate the model
# graph initialization should be before FileWriter, otherwise the graph will not appear in TensorBoard
model = EntityLSTM(dataset, parameters)
# Instantiate the writers for TensorBoard
writers = {}
for dataset_type in dataset_filepaths.keys():
writers[dataset_type] = tf.summary.FileWriter(
tensorboard_log_folders[dataset_type],
graph=sess.graph)
embedding_writer = tf.summary.FileWriter(
model_folder
) # embedding_writer has to write in model_folder, otherwise TensorBoard won't be able to view embeddings
embeddings_projector_config = projector.ProjectorConfig()
tensorboard_token_embeddings = embeddings_projector_config.embeddings.add(
)
tensorboard_token_embeddings.tensor_name = model.token_embedding_weights.name
token_list_file_path = os.path.join(
model_folder, 'tensorboard_metadata_tokens.tsv')
tensorboard_token_embeddings.metadata_path = os.path.relpath(
token_list_file_path, '..')
if parameters['use_character_lstm']:
tensorboard_character_embeddings = embeddings_projector_config.embeddings.add(
)
tensorboard_character_embeddings.tensor_name = model.character_embedding_weights.name
character_list_file_path = os.path.join(
model_folder, 'tensorboard_metadata_characters.tsv')
tensorboard_character_embeddings.metadata_path = os.path.relpath(
character_list_file_path, '..')
projector.visualize_embeddings(embedding_writer,
embeddings_projector_config)
# Write metadata for TensorBoard embeddings
token_list_file = codecs.open(token_list_file_path, 'w',
'UTF-8')
for token_index in range(len(dataset.index_to_token)):
token_list_file.write('{0}\n'.format(
dataset.index_to_token[token_index]))
token_list_file.close()
if parameters['use_character_lstm']:
character_list_file = codecs.open(character_list_file_path,
'w', 'UTF-8')
for character_index in range(dataset.alphabet_size):
if character_index == dataset.PADDING_CHARACTER_INDEX:
character_list_file.write('PADDING\n')
else:
character_list_file.write('{0}\n'.format(
dataset.index_to_character[character_index]))
character_list_file.close()
try:
# Initialize the model
sess.run(tf.global_variables_initializer())
if not parameters['use_pretrained_model']:
model.load_pretrained_token_embeddings(
sess, dataset, parameters)
# Start training + evaluation loop. Each iteration corresponds to 1 epoch.
bad_counter = 0 # number of epochs with no improvement on the validation test in terms of F1-score
previous_best_valid_f1_score = 0
transition_params_trained = np.random.rand(
len(dataset.unique_labels), len(dataset.unique_labels)
) #TODO np.random.rand(len(dataset.unique_labels)+2,len(dataset.unique_labels)+2)
model_saver = tf.train.Saver(
max_to_keep=None
) #parameters['maximum_number_of_epochs']) # defaults to saving all variables
epoch_number = 0
while True:
epoch_number += 1
print('\nStarting epoch {0}'.format(epoch_number))
epoch_start_time = time.time()
if parameters[
'use_pretrained_model'] and epoch_number == 1:
# Restore pretrained model parameters
transition_params_trained = train.restore_model_parameters_from_pretrained_model(
parameters, dataset, sess, model, model_saver)
elif epoch_number != 0:
# Train model: loop over all sequences of training set with shuffling
sequence_numbers = list(
range(len(dataset.token_indices['train'])))
random.shuffle(sequence_numbers)
data_counter = 0
sub_id = 0
for i in tqdm(range(0, len(sequence_numbers),
parameters['batch_size']),
"Training",
mininterval=1):
data_counter += parameters['batch_size']
if data_counter >= 20000:
data_counter = 0
sub_id += 0.001
print("Intermediate evaluation number: ",
sub_id)
#model_saver.save(sess,
# os.path.join(model_folder, 'model_{0:05d}_{1}.ckpt'.format(epoch_number, len(sequence_numbers)/4/len(sequence_numbers))))
epoch_elapsed_training_time = time.time(
) - epoch_start_time
print(
'Training completed in {0:.2f} seconds'
.format(epoch_elapsed_training_time),
flush=True)
y_pred, y_true, output_filepaths = train.predict_labels(
sess, model, transition_params_trained,
parameters, dataset,
epoch_number + sub_id,
stats_graph_folder, dataset_filepaths)
# Evaluate model: save and plot results
evaluate.evaluate_model(
results, dataset, y_pred, y_true,
stats_graph_folder, epoch_number,
epoch_start_time, output_filepaths,
parameters)
# Save model
model_saver.save(
sess,
os.path.join(
model_folder,
'model_{0:07.3f}.ckpt'.format(
epoch_number + sub_id)))
# Save TensorBoard logs
summary = sess.run(model.summary_op,
feed_dict=None)
writers['train'].add_summary(
summary, epoch_number)
writers['train'].flush()
utils.copytree(
writers['train'].get_logdir(),
model_folder)
# Early stop
valid_f1_score = results['epoch'][
epoch_number][0]['valid']['f1_score'][
'micro']
# valid_precision = results['epoch'][epoch_number][0]['valid']['precision']['micro']
# valid_recall = results['epoch'][epoch_number][0]['valid']['recall']['micro']
# valid_fscores.append(valid_f1_score)
if valid_f1_score > previous_best_valid_f1_score:
bad_counter = 0
previous_best_valid_f1_score = valid_f1_score
# previous_best_valid_precision = valid_precision
# previous_best_valid_recall = valid_recall
else:
bad_counter += 1
sequence_number = sequence_numbers[
i:i + parameters['batch_size']]
transition_params_trained, loss = train.train_step(
sess, dataset, sequence_number, model,
transition_params_trained, parameters)
epoch_elapsed_training_time = time.time(
) - epoch_start_time
print('Training completed in {0:.2f} seconds'.format(
epoch_elapsed_training_time),
flush=True)
y_pred, y_true, output_filepaths = train.predict_labels(
sess, model, transition_params_trained, parameters,
dataset, epoch_number, stats_graph_folder,
dataset_filepaths)
# Evaluate model: save and plot results
evaluate.evaluate_model(results, dataset, y_pred,
y_true, stats_graph_folder,
epoch_number, epoch_start_time,
output_filepaths, parameters)
# Save model
model_saver.save(
sess,
os.path.join(
model_folder,
'model_{0:05d}.ckpt'.format(epoch_number)))
# Save TensorBoard logs
summary = sess.run(model.summary_op, feed_dict=None)
writers['train'].add_summary(summary, epoch_number)
writers['train'].flush()
utils.copytree(writers['train'].get_logdir(),
model_folder)
# Early stop
valid_f1_score = results['epoch'][epoch_number][0][
'valid']['f1_score']['micro']
#valid_precision = results['epoch'][epoch_number][0]['valid']['precision']['micro']
#valid_recall = results['epoch'][epoch_number][0]['valid']['recall']['micro']
#valid_fscores.append(valid_f1_score)
if valid_f1_score > previous_best_valid_f1_score:
bad_counter = 0
previous_best_valid_f1_score = valid_f1_score
#previous_best_valid_precision = valid_precision
#previous_best_valid_recall = valid_recall
else:
bad_counter += 1
print(
"The last {0} epochs have not shown improvements on the validation set."
.format(bad_counter))
if bad_counter >= parameters['patience']:
print('Early Stop!')
results['execution_details']['early_stop'] = True
break
if epoch_number >= parameters[
'maximum_number_of_epochs']:
break
except KeyboardInterrupt:
results['execution_details']['keyboard_interrupt'] = True
print('Training interrupted')
# remove the experiment
remove_experiment = input(
"Do you want to remove the experiment? (yes/y/Yes)")
if remove_experiment in ["Yes", "yes", "y"]:
shutil.rmtree(stats_graph_folder)
print("Folder removed")
else:
print('Finishing the experiment')
end_time = time.time()
results['execution_details'][
'train_duration'] = end_time - start_time
results['execution_details']['train_end'] = end_time
evaluate.save_results(results, stats_graph_folder)
except Exception:
logging.exception("")
remove_experiment = input(
"Do you want to remove the experiment? (yes/y/Yes)")
if remove_experiment in ["Yes", "yes", "y"]:
shutil.rmtree(stats_graph_folder)
print("Folder removed")
sess.close() # release the session's resources
if 'cross_validation' in parameters and parameters[
'cross_validation'] > 1:
valid_fscores.append(previous_best_valid_f1_score)
#valid_precisions.append(previous_best_valid_precision)
#valid_recalls.append(previous_best_valid_recall)
if 'cross_validation' in parameters and parameters['cross_validation'] > 1:
print("mean f1score:", np.mean(valid_fscores))
#print("mean precision:", np.mean(valid_precisions))
#print("mean recall:", np.mean(valid_recalls))
with codecs.open(os.path.join(stats_graph_folder, "result_cv.txt"),
"w") as file:
file.write("F1score " + ", ".join(map(str, valid_fscores)))
# file.write("Precision " + valid_precisions)
# file.write("Recall " + valid_recalls)
file.write("Mean F1score " + str(np.mean(valid_fscores)))
def predict(self, text):
"""
Predict
Args:
text (str): Description.
"""
self.prediction_count += 1
if self.prediction_count == 1:
self.parameters['dataset_text_folder'] = os.path.join(
'.', 'data', 'temp')
self.stats_graph_folder, _ = self._create_stats_graph_folder(
self.parameters)
# Update the deploy folder, file, and modeldata
dataset_type = 'deploy'
# Delete all deployment data
for filepath in glob.glob(
os.path.join(self.parameters['dataset_text_folder'],
'{0}*'.format(dataset_type))):
if os.path.isdir(filepath):
shutil.rmtree(filepath)
else:
os.remove(filepath)
# Create brat folder and file
dataset_brat_deploy_folder = os.path.join(
self.parameters['dataset_text_folder'], dataset_type)
utils.create_folder_if_not_exists(dataset_brat_deploy_folder)
dataset_brat_deploy_filepath = os.path.join(
dataset_brat_deploy_folder,
'temp_{0}.txt'.format(str(self.prediction_count).zfill(5)))
#self._get_dataset_brat_deploy_filepath(dataset_brat_deploy_folder)
# print('over here: ',dataset_brat_deploy_filepath)
with codecs.open(dataset_brat_deploy_filepath, 'w', 'UTF-8') as f:
f.write(text)
# Update deploy filepaths
dataset_filepaths, dataset_brat_folders = self._get_valid_dataset_filepaths(
self.parameters, dataset_types=[dataset_type])
self.dataset_filepaths.update(dataset_filepaths)
self.dataset_brat_folders.update(dataset_brat_folders)
# Update the dataset for the new deploy set
self.modeldata.update_dataset(self.dataset_filepaths, [dataset_type])
# Predict labels and output brat
output_filepaths = {}
prediction_output = train.prediction_step(
self.sess, self.modeldata, dataset_type, self.model,
self.transition_params_trained, self.stats_graph_folder,
self.prediction_count, self.parameters, self.dataset_filepaths)
_, _, output_filepaths[dataset_type] = prediction_output
conll_to_brat.output_brat(output_filepaths,
self.dataset_brat_folders,
self.stats_graph_folder,
overwrite=True)
# Print and output result
text_filepath = os.path.join(
self.stats_graph_folder, 'brat', 'deploy',
os.path.basename(dataset_brat_deploy_filepath))
annotation_filepath = os.path.join(
self.stats_graph_folder, 'brat', 'deploy', '{0}.ann'.format(
utils.get_basename_without_extension(
dataset_brat_deploy_filepath)))
text2, entities = brat_to_conll.get_entities_from_brat(
text_filepath, annotation_filepath, verbose=True)
assert (text == text2)
return entities
def predict(self, test_file_path):
# Not use
text = ''
with open(test_file_path, "r") as f:
text = f.read()
test_file_path = test_file_path.split('/')[-1]
self.prediction_count += 1
if self.prediction_count == 1:
self.parameters['dataset_text_folder'] = os.path.join('..', 'data', 'temp')
self.stats_graph_folder, _ = self._create_stats_graph_folder(self.parameters)
# Update the deploy folder, file, and dataset
dataset_type = 'deploy'
### Delete all deployment data
for filepath in glob.glob(os.path.join(self.parameters['dataset_text_folder'], '{0}*'.format(dataset_type))):
if os.path.isdir(filepath):
shutil.rmtree(filepath)
else:
os.remove(filepath)
### Create brat folder and file
dataset_brat_deploy_folder = os.path.join(self.parameters['dataset_text_folder'], dataset_type)
utils.create_folder_if_not_exists(dataset_brat_deploy_folder)
dataset_brat_deploy_filepath = os.path.join(dataset_brat_deploy_folder, test_file_path.format(str(self.prediction_count).zfill(5)))#self._get_dataset_brat_deploy_filepath(dataset_brat_deploy_folder)
with codecs.open(dataset_brat_deploy_filepath, 'w', 'UTF-8') as f:
f.write(text)
### Update deploy filepaths
dataset_filepaths, dataset_brat_folders = self._get_valid_dataset_filepaths(self.parameters, dataset_types=[dataset_type])
self.dataset_filepaths.update(dataset_filepaths)
self.dataset_brat_folders.update(dataset_brat_folders)
### Update the dataset for the new deploy set
self.dataset.update_dataset(self.dataset_filepaths, [dataset_type])
# Predict labels and output brat
output_filepaths = {}
prediction_output = train.prediction_step(self.sess, self.dataset, dataset_type, self.model, self.transition_params_trained, self.stats_graph_folder, self.prediction_count, self.parameters, self.dataset_filepaths)
_, _, output_filepaths[dataset_type] = prediction_output
conll_to_brat.output_brat(output_filepaths, self.dataset_brat_folders, self.stats_graph_folder, overwrite=True)
# Print and output result
text_filepath = os.path.join(self.stats_graph_folder, 'brat', 'deploy', os.path.basename(dataset_brat_deploy_filepath))
annotation_filepath = os.path.join(self.stats_graph_folder, 'brat', 'deploy', '{0}.ann'.format(utils.get_basename_without_extension(dataset_brat_deploy_filepath)))
text2, entities = brat_to_conll.get_entities_from_brat(text_filepath, annotation_filepath, verbose=True)
assert(text == text2)
#print (entities)
os.rename(self.stats_graph_folder, "../data/" + self.stats_graph_folder.split('/')[-1])
print("Use brat tool to see result at ", "../data/" + self.stats_graph_folder.split('/')[-1])
def get_valid_dataset_filepaths(parameters):
dataset_filepaths = {}
dataset_brat_folders = {}
for dataset_type in ['train', 'valid', 'test', 'deploy']:
dataset_filepaths[dataset_type] = os.path.join(
parameters['dataset_text_folder'], '{0}.txt'.format(dataset_type))
dataset_brat_folders[dataset_type] = os.path.join(
parameters['dataset_text_folder'], dataset_type)
dataset_compatible_with_brat_filepath = os.path.join(
parameters['dataset_text_folder'],
'{0}_compatible_with_brat.txt'.format(dataset_type))
# Conll file exists
if os.path.isfile(dataset_filepaths[dataset_type]) and os.path.getsize(
dataset_filepaths[dataset_type]) > 0:
# Brat text files exist
if os.path.exists(dataset_brat_folders[dataset_type]) and len(
glob.glob(
os.path.join(dataset_brat_folders[dataset_type],
'*.txt'))) > 0:
# Check compatibility between conll and brat files
brat_to_conll.check_brat_annotation_and_text_compatibility(
dataset_brat_folders[dataset_type])
if os.path.exists(dataset_compatible_with_brat_filepath):
dataset_filepaths[
dataset_type] = dataset_compatible_with_brat_filepath
conll_to_brat.check_compatibility_between_conll_and_brat_text(
dataset_filepaths[dataset_type],
dataset_brat_folders[dataset_type])
# Brat text files do not exist
else:
# Populate brat text and annotation files based on conll file
conll_to_brat.conll_to_brat(
dataset_filepaths[dataset_type],
dataset_compatible_with_brat_filepath,
dataset_brat_folders[dataset_type],
dataset_brat_folders[dataset_type])
dataset_filepaths[
dataset_type] = dataset_compatible_with_brat_filepath
# Conll file does not exist
else:
# Brat text files exist
if os.path.exists(dataset_brat_folders[dataset_type]) and len(
glob.glob(
os.path.join(dataset_brat_folders[dataset_type],
'*.txt'))) > 0:
dataset_filepath_for_tokenizer = os.path.join(
parameters['dataset_text_folder'],
'{0}_{1}.txt'.format(dataset_type,
parameters['tokenizer']))
if os.path.exists(dataset_filepath_for_tokenizer):
conll_to_brat.check_compatibility_between_conll_and_brat_text(
dataset_filepath_for_tokenizer,
dataset_brat_folders[dataset_type])
else:
# Populate conll file based on brat files
brat_to_conll.brat_to_conll(
dataset_brat_folders[dataset_type],
dataset_filepath_for_tokenizer,
parameters['tokenizer'], parameters['spacylanguage'])
dataset_filepaths[
dataset_type] = dataset_filepath_for_tokenizer
# Brat text files do not exist
else:
del dataset_filepaths[dataset_type]
del dataset_brat_folders[dataset_type]
continue
if parameters['tagging_format'] == 'bioes':
# Generate conll file with BIOES format
bioes_filepath = os.path.join(
parameters['dataset_text_folder'], '{0}_bioes.txt'.format(
utils.get_basename_without_extension(
dataset_filepaths[dataset_type])))
utils_nlp.convert_conll_from_bio_to_bioes(
dataset_filepaths[dataset_type], bioes_filepath)
dataset_filepaths[dataset_type] = bioes_filepath
return dataset_filepaths, dataset_brat_folders
def main():
#### Parameters - start
conf_parameters = configparser.ConfigParser()
conf_parameters.read(os.path.join('.','parameters.ini'))
nested_parameters = utils.convert_configparser_to_dictionary(conf_parameters)
parameters = {}
for k,v in nested_parameters.items():
parameters.update(v)
for k,v in parameters.items():
if k in ['remove_unknown_tokens','character_embedding_dimension','character_lstm_hidden_state_dimension','token_embedding_dimension','token_lstm_hidden_state_dimension',
'patience','maximum_number_of_epochs','maximum_training_time','number_of_cpu_threads','number_of_gpus']:
parameters[k] = int(v)
if k in ['dropout_rate']:
parameters[k] = float(v)
if k in ['use_character_lstm','is_character_lstm_bidirect','is_token_lstm_bidirect','use_crf']:
parameters[k] = distutils.util.strtobool(v)
pprint(parameters)
# Load dataset
dataset_filepaths = {}
dataset_filepaths['train'] = os.path.join(parameters['dataset_text_folder'], 'train.txt')
dataset_filepaths['valid'] = os.path.join(parameters['dataset_text_folder'], 'valid.txt')
dataset_filepaths['test'] = os.path.join(parameters['dataset_text_folder'], 'test.txt')
dataset = ds.Dataset()
dataset.load_dataset(dataset_filepaths, parameters)
with tf.Graph().as_default():
session_conf = tf.ConfigProto(
device_count={'CPU': 1, 'GPU': 1},
allow_soft_placement=True, # automatically choose an existing and supported device to run the operations in case the specified one doesn't exist
log_device_placement=False
)
sess = tf.Session(config=session_conf)
with sess.as_default():
model = EntityLSTM(dataset, parameters)
# Define training procedure
global_step = tf.Variable(0, name="global_step", trainable=False)
if parameters['optimizer'] == 'adam':
optimizer = tf.train.AdamOptimizer(1e-3)
elif parameters['optimizer'] == 'sgd':
optimizer = tf.train.GradientDescentOptimizer(0.005)
else:
raise ValueError("The lr_method parameter must be either adam or sgd.")
# https://github.com/google/prettytensor/issues/6
# https://www.tensorflow.org/api_docs/python/framework/graph_collections
#print('tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES) : {0}'.format(tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES) ))
#print('tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES) : {0}'.format(tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES) ))
#print('tf.get_collection(tf.GraphKeys.MODEL_VARIABLES) : {0}'.format(tf.get_collection(tf.GraphKeys.MODEL_VARIABLES) ))
# https://github.com/blei-lab/edward/issues/286#ref-pullrequest-181330211 : utility function to get all tensorflow variables a node depends on
grads_and_vars = optimizer.compute_gradients(model.loss)
# By defining a global_step variable and passing it to the optimizer we allow TensorFlow handle the counting of training steps for us.
# The global step will be automatically incremented by one every time you execute train_op.
train_op = optimizer.apply_gradients(grads_and_vars, global_step=global_step)
# Initialize all variables
sess.run(tf.global_variables_initializer())
# Load pretrained token embeddings
if not parameters['token_pretrained_embedding_filepath'] == '':
load_token_embeddings(sess, model.W, dataset, parameters)
estop = False # early stop
start_time = time.time()
experiment_timestamp = utils.get_current_time_in_miliseconds()
results = {}
#results['model_options'] = copy.copy(model_options)
#results['model_options'].pop('optimizer', None)
results['epoch'] = {}
# save/initialize execution details
results['execution_details'] = {}
results['execution_details']['train_start'] = start_time
results['execution_details']['time_stamp'] = experiment_timestamp
results['execution_details']['early_stop'] = False
results['execution_details']['keyboard_interrupt'] = False
results['execution_details']['num_epochs'] = 0
results['model_options'] = copy.copy(parameters)
dataset_name = utils.get_basename_without_extension(parameters['dataset_text_folder']) #opts.train.replace('/', '_').split('.')[0] # 'conll2003en'
model_name = '{0}_{1}'.format(dataset_name, results['execution_details']['time_stamp'])
output_folder=os.path.join('..', 'output')
stats_graph_folder=os.path.join(output_folder, model_name) # Folder where to save graphs
utils.create_folder_if_not_exists(output_folder)
print('stats_graph_folder: {0}'.format(stats_graph_folder))
utils.create_folder_if_not_exists(stats_graph_folder)
model_folder = os.path.join(stats_graph_folder, 'model')
utils.create_folder_if_not_exists(model_folder)
step = 0
bad_counter = 0
previous_best_valid_f1_score = 0
transition_params_trained = np.random.rand(len(dataset.unique_labels),len(dataset.unique_labels))
try:
while True:
epoch_number = math.floor(step / len(dataset.token_indices['train']))
print('epoch_number: {0}'.format(epoch_number))
epoch_start_time = time.time()
#print('step: {0}'.format(step))
# Train model: loop over all sequences of training set with shuffling
sequence_numbers=list(range(len(dataset.token_indices['train'])))
random.shuffle(sequence_numbers)
for sequence_number in sequence_numbers:
transition_params_trained = train_step(sess, dataset, sequence_number, train_op, global_step, model, transition_params_trained, parameters)
step += 1
if sequence_number % 100 == 0:
print('.',end='', flush=True)
#break
# Evaluate model
print('step: {0}'.format(step))
all_predictions = {}
all_y_true = {}
output_filepaths = {}
for dataset_type in ['train', 'valid', 'test']:
print('dataset_type: {0}'.format(dataset_type))
all_predictions[dataset_type], all_y_true[dataset_type], output_filepaths[dataset_type] = evaluate_model(sess, dataset, dataset_type, model, transition_params_trained, step, stats_graph_folder, epoch_number, parameters)
model_options = None
# Save and plot results
# TODO: remove uidx
uidx = 0
results['epoch'][epoch_number] = []
results['execution_details']['num_epochs'] = epoch_number
epoch_elapsed_training_time = time.time() - epoch_start_time
print('epoch_elapsed_training_time: {0:02f} seconds'.format(epoch_elapsed_training_time))
assess_model.assess_and_save(results, dataset, model_options, all_predictions, all_y_true, stats_graph_folder, epoch_number, uidx, epoch_start_time)
assess_model.plot_f1_vs_epoch(results, stats_graph_folder, 'f1_score')
assess_model.plot_f1_vs_epoch(results, stats_graph_folder, 'accuracy_score')
# CoNLL evaluation script
for dataset_type in ['train', 'valid', 'test']:
conll_evaluation_script = os.path.join('.', 'conlleval')
conll_output_filepath = '{0}_conll_evaluation.txt'.format(output_filepaths[dataset_type])
shell_command = 'perl {0} < {1} > {2}'.format(conll_evaluation_script, output_filepaths[dataset_type], conll_output_filepath)
print('shell_command: {0}'.format(shell_command))
#subprocess.call([shell_command])
os.system(shell_command)
conll_parsed_output = utils_nlp.get_parsed_conll_output(conll_output_filepath)
print('conll_parsed_output: {0}'.format(conll_parsed_output))
results['epoch'][epoch_number][0][dataset_type]['conll'] = conll_parsed_output
results['epoch'][epoch_number][0][dataset_type]['f1_conll'] = {}
results['epoch'][epoch_number][0][dataset_type]['f1_conll']['micro'] = results['epoch'][epoch_number][0][dataset_type]['conll']['all']['f1']
assess_model.plot_f1_vs_epoch(results, stats_graph_folder, 'f1_conll', from_json=False)
#end_time = time.time()
#results['execution_details']['train_duration'] = end_time - start_time
#results['execution_details']['train_end'] = end_time
# Early stop
valid_f1_score = results['epoch'][epoch_number][0]['valid']['f1_score']['micro']
if valid_f1_score > previous_best_valid_f1_score:
bad_counter = 0
previous_best_valid_f1_score = valid_f1_score
else:
bad_counter += 1
if bad_counter > parameters['patience']:
print('Early Stop!')
results['execution_details']['early_stop'] = True
break
if epoch_number > parameters['maximum_number_of_epochs']: break
except KeyboardInterrupt:
results['execution_details']['keyboard_interrupt'] = True
# assess_model.save_results(results, stats_graph_folder)
print('Training interrupted')
print('Finishing the experiment')
end_time = time.time()
results['execution_details']['train_duration'] = end_time - start_time
results['execution_details']['train_end'] = end_time
assess_model.save_results(results, stats_graph_folder)
sess.close() # release the session's resources
