def prediction_step(sess, dataset, dataset_type, model, transition_params_trained, stats_graph_folder, epoch_number, parameters, dataset_filepaths):
if dataset_type == 'deploy':
print('Predict labels for the {0} set'.format(dataset_type))
else:
print('Evaluate model on the {0} set'.format(dataset_type))
all_predictions = []
all_y_true = []
output_filepath = os.path.join(stats_graph_folder, '{1:03d}_{0}.txt'.format(dataset_type,epoch_number))
output_file = codecs.open(output_filepath, 'w', 'UTF-8')
original_conll_file = codecs.open(dataset_filepaths[dataset_type], 'r', 'UTF-8')
for i in range(len(dataset.token_indices[dataset_type])):
feed_dict = {
model.input_token_indices: dataset.token_indices[dataset_type][i],
model.input_token_character_indices: dataset.character_indices_padded[dataset_type][i],
model.input_token_lengths: dataset.token_lengths[dataset_type][i],
model.input_label_indices_vector: dataset.label_vector_indices[dataset_type][i],
model.dropout_keep_prob: 1.
}
unary_scores, predictions = sess.run([model.unary_scores, model.predictions], feed_dict)
if parameters['use_crf']:
predictions, _ = tf.contrib.crf.viterbi_decode(unary_scores, transition_params_trained)
predictions = predictions[1:-1]
else:
predictions = predictions.tolist()
assert(len(predictions) == len(dataset.tokens[dataset_type][i]))
output_string = ''
prediction_labels = [dataset.index_to_label[prediction] for prediction in predictions]
gold_labels = dataset.labels[dataset_type][i]
if parameters['tagging_format'] == 'bioes':
prediction_labels = utils_nlp.bioes_to_bio(prediction_labels)
gold_labels = utils_nlp.bioes_to_bio(gold_labels)
for prediction, token, gold_label in zip(prediction_labels, dataset.tokens[dataset_type][i], gold_labels):
while True:
line = original_conll_file.readline()
split_line = line.strip().split(' ')
if '-DOCSTART-' in split_line[0] or len(split_line) == 0 or len(split_line[0]) == 0:
continue
else:
token_original = split_line[0]
if parameters['tagging_format'] == 'bioes':
split_line.pop()
gold_label_original = split_line[-1]
assert(token == token_original and gold_label == gold_label_original)
break
split_line.append(prediction)
output_string += ' '.join(split_line) + '\n'
output_file.write(output_string+'\n')
all_predictions.extend(predictions)
all_y_true.extend(dataset.label_indices[dataset_type][i])
output_file.close()
original_conll_file.close()
if dataset_type != 'deploy':
if parameters['main_evaluation_mode'] == 'conll':
conll_evaluation_script = os.path.join('.', 'conlleval')
conll_output_filepath = '{0}_conll_evaluation.txt'.format(output_filepath)
shell_command = 'perl {0} < {1} > {2}'.format(conll_evaluation_script, output_filepath, conll_output_filepath)
os.system(shell_command)
with open(conll_output_filepath, 'r') as f:
classification_report = f.read()
print(classification_report)
else:
new_y_pred, new_y_true, new_label_indices, new_label_names, _, _ = remap_labels(all_predictions, all_y_true, dataset, parameters['main_evaluation_mode'])
print(sklearn.metrics.classification_report(new_y_true, new_y_pred, digits=4, labels=new_label_indices, target_names=new_label_names))
return all_predictions, all_y_true, output_filepath
def prediction_step(sess,
dataset,
dataset_type,
model,
transition_params_trained,
stats_graph_folder,
epoch_number,
parameters,
dataset_filepaths,
for_adapter=False):
if dataset_type == 'deploy':
print('Predict labels for the {0} set'.format(dataset_type))
else:
print('Evaluate model on the {0} set'.format(dataset_type))
if for_adapter == True:
all_predictions_per_sentence = []
all_y_true_per_sentence = []
all_prediction_labels_per_sentence = []
all_predictions = []
all_y_true = []
output_filepath = os.path.join(
stats_graph_folder, '{1:03d}_{0}.txt'.format(dataset_type,
epoch_number))
output_file = codecs.open(output_filepath,
'w',
'latin-1',
errors='replace')
original_conll_file = codecs.open(dataset_filepaths[dataset_type],
'r',
'latin-1',
errors='replace')
for i in range(len(dataset.token_indices[dataset_type])):
if parameters['use_adapter']:
feed_dict = {
model.input_token_indices:
dataset.token_indices[dataset_type][i],
model.input_token_character_indices:
dataset.character_indices_padded[dataset_type][i],
model.input_token_lengths:
dataset.token_lengths[dataset_type][i],
model.input_label_indices_vector:
dataset.label_vector_indices[dataset_type][i],
model.input_label_adapter_indices_vector:
dataset.label_adapter_vector_indices[dataset_type][i],
model.dropout_keep_prob:
1.,
model.adapter_keep_prob:
1.
}
if parameters['include_pos']:
feed_dict[
model.
input_label_pos_indices_vector] = dataset.label_pos_vector_indices[
dataset_type][i]
elif for_adapter == True:
# use for pred/eval step, not to provide the gold labels in dataset
feed_dict = {
model.input_token_indices:
dataset.token_indices[dataset_type][i],
model.input_token_character_indices:
dataset.character_indices_padded[dataset_type][i],
model.input_token_lengths:
dataset.token_lengths[dataset_type][i],
model.dropout_keep_prob:
1.
}
else:
feed_dict = {
model.input_token_indices:
dataset.token_indices[dataset_type][i],
model.input_token_character_indices:
dataset.character_indices_padded[dataset_type][i],
model.input_token_lengths:
dataset.token_lengths[dataset_type][i],
model.input_label_indices_vector:
dataset.label_vector_indices[dataset_type][i],
model.dropout_keep_prob:
1.
}
unary_scores, predictions = sess.run(
[model.unary_scores, model.predictions], feed_dict)
if parameters['use_crf']:
predictions, _ = tf.contrib.crf.viterbi_decode(
unary_scores, transition_params_trained)
predictions = predictions[1:-1]
else:
predictions = predictions.tolist()
assert (len(predictions) == len(dataset.tokens[dataset_type][i]))
output_string = ''
prediction_labels = [
dataset.index_to_label[prediction] for prediction in predictions
]
gold_labels = dataset.labels[dataset_type][i]
if parameters['tagging_format'] == 'bioes':
prediction_labels = utils_nlp.bioes_to_bio(prediction_labels)
gold_labels = utils_nlp.bioes_to_bio(gold_labels)
try:
assert len(prediction_labels) == len(gold_labels)
except AssertionError:
print(dataset.tokens[dataset_type][i])
print(gold_labels)
print(prediction_labels)
for z, (prediction, token, gold_label) in enumerate(
zip(prediction_labels, dataset.tokens[dataset_type][i],
gold_labels)):
while True:
line = original_conll_file.readline()
split_line = line.strip().split(' ')
if '-DOCSTART-' in split_line[0] or len(
split_line) == 0 or len(split_line[0]) == 0:
continue
else:
token_original = split_line[0]
if parameters['tagging_format'] == 'bioes':
split_line.pop()
gold_label_original = split_line[-1]
try:
assert (token == dataset.tokenize(token_original)
and gold_label == gold_label_original)
except AssertionError:
print(' '.join([
dataset.tokens[dataset_type][i][x] + '/' +
gold_labels[x] for x in range(len(gold_labels))
]))
print(
'token: {:s} - gold_label: {:s} - gold_label_original: {:s}'
.format(dataset.tokens[dataset_type][i][z],
gold_label, gold_label_original))
break
split_line.append(prediction)
output_string += ' '.join(split_line) + '\n'
newstr = output_string + '\n'
output_file.write(newstr)
if for_adapter == True:
all_predictions_per_sentence.append(predictions)
all_y_true_per_sentence.append(
dataset.label_indices[dataset_type][i])
all_prediction_labels_per_sentence.append(prediction_labels)
all_predictions.extend(predictions)
all_y_true.extend(dataset.label_indices[dataset_type][i])
output_file.close()
original_conll_file.close()
if dataset_type != 'deploy':
if parameters['main_evaluation_mode'] == 'conll':
conll_evaluation_script = os.path.join('.', 'conlleval')
conll_output_filepath = '{0}_conll_evaluation.txt'.format(
output_filepath)
shell_command = '/usr/bin/perl {0} < {1} > {2}'.format(
conll_evaluation_script, output_filepath,
conll_output_filepath)
os.system(shell_command)
with open(conll_output_filepath, 'r') as f:
classification_report = f.read()
print(classification_report)
else:
new_y_pred, new_y_true, new_label_indices, new_label_names, _, _ = remap_labels(
all_predictions, all_y_true, dataset,
parameters['main_evaluation_mode'])
print(
sklearn.metrics.classification_report(
new_y_true,
new_y_pred,
digits=4,
labels=new_label_indices,
target_names=new_label_names))
if for_adapter == True:
return all_prediction_labels_per_sentence, all_predictions, all_y_true, output_filepath
else:
return all_predictions, all_y_true, output_filepath
def prediction_step(sess, dataset, dataset_type, model, transition_params_trained, stats_graph_folder, epoch_number, parameters, dataset_filepaths, demo=False):
if dataset_type == 'deploy':
print('Predict labels for the {0} set'.format(dataset_type))
else:
print('Evaluate model on the {0} set'.format(dataset_type))
all_predictions = []
all_predictions_label = []
all_y_true = []
all_y_true_label = []
output_filepath = os.path.join(stats_graph_folder, '{1:07.3f}_{0}.txt'.format(dataset_type,epoch_number))
encoding = "utf-8"
output_file = codecs.open(output_filepath, 'w', encoding)
if "combined" in dataset_filepaths[dataset_type]:
label_idx = 2
else:
label_idx = -1
#original_conll_file = codecs.open(dataset_filepaths[dataset_type], 'r', encoding=encoding)
sequence_numbers = list(range(len(dataset.token_indices[dataset_type])))
for i in tqdm(range(0,len(dataset.token_indices[dataset_type]), parameters['batch_size']), "Testing on {} at epoch {}".format(dataset_type, epoch_number)): #
sequence_number = sequence_numbers[i: i + parameters['batch_size']]
batch = utils.pad_batch(dataset, sequence_number, dataset_type)
feed_dict = {
model.input_token_indices: batch['token_indices_padded'],
model.input_sequence_lengths: batch['sequence_lengths'],
model.input_token_character_indices: batch['character_indices_padded'],
model.input_token_lengths: batch['token_lengths'],
model.input_label_indices_flat: batch['label_indices'],
model.input_label_indices_vector: batch['label_vector_indices'],
model.dropout_keep_prob: 1
}
batch_unary_scores, batch_predictions = sess.run([model.unary_scores, model.predictions], feed_dict)
for unary_score, y, sequence_length, predictions, j in zip(batch_unary_scores, batch['label_indices'],
batch['sequence_lengths'], batch_predictions, sequence_number):
if parameters['use_crf']:
# Remove padding from the scores and tag sequence.
unary_score =unary_score[:sequence_length]
y = y[:sequence_length]
# Compute the highest scoring sequence.
predictions, _ = tf.contrib.crf.viterbi_decode(unary_score, transition_params_trained)
else:
predictions = predictions[:sequence_length].tolist()
assert(len(predictions) == len(np.array(dataset.tokens[dataset_type])[j]))
output_string = []
prediction_labels = [dataset.index_to_label[pred] for pred in predictions]
gold_labels = np.array(dataset.labels[dataset_type])[j]
if parameters['tagging_format'] == 'bioes':
prediction_labels = utils_nlp.bioes_to_bio(prediction_labels)
gold_labels = utils_nlp.bioes_to_bio(gold_labels)
for prediction, token, gold_label in zip(prediction_labels, np.array(dataset.tokens[dataset_type])[j], gold_labels):
#while True:
# line = original_conll_file.readline()
# split_line = line.strip().split(' ')
# if "combined" in dataset_filepaths[dataset_type]:
# split_line = split_line[:label_idx+1]
# if len(split_line) == 0 or '-DOCSTART-' in split_line[0] or len(split_line[0]) == 0:
# continue
# else:
# token_original = split_line[0]
# if parameters['tagging_format'] == 'bioes':
# split_line.pop()
# gold_label_original = split_line[label_idx]
# #if token != token_original or gold_label != gold_label_original:
# assert(token == token_original and gold_label == gold_label_original)
# break
#split_line.append(prediction)
#output_string.append(' '.join(split_line))
output_string.append(' '.join([token, gold_label, prediction]))
output_file.write("\n".join(output_string))
output_file.write("\n\n")
all_predictions.extend(predictions)
all_y_true.extend(np.array(dataset.label_indices[dataset_type])[j])
all_predictions_label.append(prediction_labels)
all_y_true_label.append(np.array(dataset.labels[dataset_type])[j])
output_file.close()
#original_conll_file.close()
if demo:
return all_predictions, all_y_true, output_filepath
if dataset_type != 'deploy':
if parameters['main_evaluation_mode'] == 'conll':
conll_evaluation_script = os.path.join('.', 'conlleval')
conll_output_filepath = '{0}_conll_evaluation.txt'.format(output_filepath)
#if "labelled_yelp_tips_th06" in parameters["dataset_train"]:
# shell_command = 'perl {0} -r < {1} > {2}'.format(conll_evaluation_script, output_filepath, conll_output_filepath)
#else:
shell_command = 'perl {0} < {1} > {2}'.format(conll_evaluation_script, output_filepath,
conll_output_filepath)
os.system(shell_command)
with open(conll_output_filepath, 'r') as f:
classification_report = f.read()
print(classification_report)
else:
raise AssertionError("Not implemented")
new_y_pred, new_y_true, new_label_indices, new_label_names, _, _ = remap_labels(all_predictions, all_y_true, dataset, parameters['main_evaluation_mode'])
print(sklearn.metrics.classification_report(new_y_true, new_y_pred, digits=4, labels=new_label_indices, target_names=new_label_names))
exact_score, inexact_score = report_fscore(all_y_true_label, all_predictions_label)
exact_inexact_evaluation = '{0}_exact_inexact_evaluation.txt'.format(output_filepath)
with open(exact_inexact_evaluation, "w") as file:
file.write("Exact score\n")
file.write(json.dumps(exact_score) + "\n")
file.write(json.dumps(inexact_score) + "\n")
return all_predictions, all_y_true, output_filepath
def prediction_step(sess, dataset, dataset_type, model,
transition_params_trained, stats_graph_folder,
epoch_number, parameters, dataset_filepaths):
"""
Predict.
"""
if dataset_type == 'deploy':
print('=> Predict labels for the {0} set'.format(dataset_type))
else:
print('Evaluate model on the {0} set'.format(dataset_type))
"""
Comment out writing to file only for efficiency experiment
"""
all_predictions = []
all_y_true = []
output_filepath = os.path.join(
stats_graph_folder, '{1:03d}_{0}.txt'.format(dataset_type,
epoch_number))
mentions_output_filepath = os.path.join(stats_graph_folder,
'mentions_output.txt')
print('output_file: ', output_filepath)
print('mentions_output_file: ', mentions_output_filepath)
output_file = codecs.open(output_filepath, 'w', 'UTF-8')
original_conll_file = codecs.open(dataset_filepaths[dataset_type], 'r',
'UTF-8')
# mentions_file= codecs.open(mentions_output_filepath, 'w', 'UTF-8')
output_tokens_list = []
token_list_outer = []
token_list_inner = []
for i in range(len(dataset.token_indices[dataset_type])):
feed_dict = {
model.input_token_indices:
dataset.token_indices[dataset_type][i],
model.input_token_character_indices:
dataset.character_indices_padded[dataset_type][i],
model.input_token_lengths:
dataset.token_lengths[dataset_type][i],
model.input_label_indices_vector:
dataset.label_vector_indices[dataset_type][i],
model.dropout_keep_prob:
1.
}
unary_scores, predictions = sess.run(
[model.unary_scores, model.predictions], feed_dict)
if parameters['use_crf']:
predictions, _ = tf.contrib.crf.viterbi_decode(
unary_scores, transition_params_trained)
predictions = predictions[1:-1]
else:
predictions = predictions.tolist()
assert (len(predictions) == len(dataset.tokens[dataset_type][i]))
output_string = ''
prediction_labels = [
dataset.index_to_label[prediction] for prediction in predictions
]
unary_score_list = unary_scores.tolist()[1:-1]
gold_labels = dataset.labels[dataset_type][i]
if parameters['tagging_format'] == 'bioes':
prediction_labels = utils_nlp.bioes_to_bio(prediction_labels)
gold_labels = utils_nlp.bioes_to_bio(gold_labels)
for prediction, token, gold_label, scores in zip(
prediction_labels, dataset.tokens[dataset_type][i],
gold_labels, unary_score_list):
while True:
line = original_conll_file.readline()
split_line = line.strip().split(' ')
if '-DOCSTART-' in split_line[0] or len(split_line) == 0 \
or len(split_line[0]) == 0:
continue
else:
token_original = split_line[0]
if parameters['tagging_format'] == 'bioes':
split_line.pop()
gold_label_original = split_line[-1]
assert (token == token_original
and gold_label == gold_label_original)
# print('prediction and label: ', prediction, token)
if (token != '--eosc'):
token_list_inner.append(token)
if (prediction != 'O'):
output_tokens_list.append(token + '//' +
prediction)
else:
token_list_inner.append(token)
token_list_outer.append(token_list_inner)
token_list_inner = []
output_tokens_list.append(token + '//' + prediction)
break
split_line.append(prediction)
# print('========================')
if parameters['output_scores']:
# space separated scores
scores = ' '.join([str(i) for i in scores])
split_line.append('{}'.format(scores))
# print(split_line)
output_string += ' '.join(split_line) + '\n'
output_file.write(output_string + '\n')
all_predictions.extend(predictions)
all_y_true.extend(dataset.label_indices[dataset_type][i])
output_file.close()
original_conll_file.close()
mention_count = 0
mentions_list_inner = []
mentions_list_outer = []
candidateMention = ""
for outputStr in output_tokens_list:
#outputStr=output[index]
candidate = " ".join((outputStr.split("//"))[:-1])
tag = (outputStr.split("//"))[-1]
#text="".join(candidate)
if (candidate != '--eosc'):
if (tag.startswith('B-')):
#print candidateMention
if (candidateMention != ""):
mentions_list_inner.append(candidateMention)
candidateMention = candidate
else:
candidateMention += " " + candidate
else:
if (candidateMention != ""):
mentions_list_inner.append(candidateMention)
candidateMention = ""
mentions_list_outer.append(mentions_list_inner)
mentions_list_inner = []
# print('sentence list: ', token_list_outer)
# print('mentions: ', mentions_list_outer)
for inner_list in mentions_list_outer:
mention_count += len(inner_list)
mentions_string = ','.join(inner_list) + '\n'
# mentions_file.write(mentions_string)
# mentions_string=''
# mentions_string += ' '.join(mentions_list_outer) + '\n'
# mentions_file.close()
print('tally:', len(token_list_outer), len(mentions_list_outer),
'total mentions discovered:', mention_count)
if dataset_type != 'deploy':
if parameters['main_evaluation_mode'] == 'conll':
# run perl evaluation script in python package
# conll_evaluation_script = os.path.join('.', 'conlleval')
package_name = 'neuroner'
root_dir = os.path.dirname(
pkg_resources.resource_filename(package_name, '__init__.py'))
conll_evaluation_script = os.path.join(root_dir, 'conlleval')
conll_output_filepath = '{0}_conll_evaluation.txt'.format(
output_filepath)
shell_command = 'perl {0} < {1} > {2}'.format(
conll_evaluation_script, output_filepath,
conll_output_filepath)
os.system(shell_command)
with open(conll_output_filepath, 'r') as f:
classification_report = f.read()
print(classification_report)
else:
new_y_pred, new_y_true, new_label_indices, new_label_names, _, _ = remap_labels(
all_predictions, all_y_true, dataset,
parameters['main_evaluation_mode'])
print(
sklearn.metrics.classification_report(
new_y_true,
new_y_pred,
digits=4,
labels=new_label_indices,
target_names=new_label_names))
return all_predictions, all_y_true, output_filepath
def prediction_step(sess, dataset, dataset_type, model,
transition_params_trained, stats_graph_folder,
epoch_number, parameters, dataset_filepaths):
if dataset_type == 'deploy':
print('Predict labels for the {0} set'.format(dataset_type))
else:
print('Evaluate model on the {0} set'.format(dataset_type))
all_predictions = []
all_y_true = []
output_filepath = os.path.join(
stats_graph_folder, '{1:03d}_{0}.txt'.format(dataset_type,
epoch_number))
output_file = codecs.open(output_filepath, 'w', 'UTF-8')
original_conll_file = codecs.open(dataset_filepaths[dataset_type], 'r',
'UTF-8')
#
A = []
# B=[]
C = []
# ti=[]
# chi=[]
# tl=[]
# lvi=[]
#
for i in range(len(dataset.token_indices[dataset_type])):
feed_dict = {
model.input_token_indices:
dataset.token_indices[dataset_type][i],
model.input_token_character_indices:
dataset.character_indices_padded[dataset_type][i],
model.input_token_lengths:
dataset.token_lengths[dataset_type][i],
model.input_label_indices_vector:
dataset.label_vector_indices[dataset_type][i],
model.dropout_keep_prob:
1.
}
unary_scores, predictions = sess.run(
[model.unary_scores, model.predictions], feed_dict)
#
#np.save("SalmanTest/DICT%s%s"%(i,dataset_type),feed_dict)
#save_path = saver.save(sess,"./SalmanTest/CHECKDIC.ckpt")
#np.save("SalmanTest/SICT%s%s"%(i,dataset_type),predictions)
#print("SALMAAAAANNNNNNNNNNNNNN = %s"%feed_dict)
#print("SALMAAAAAN11111111NNNNNNNNNNNNN = %s"%dataset.token_indices[dataset_type][i])
#print("SALMAAAAAN22222222NNNNNNNNNNNNN = %s"%dataset.character_indices_padded[dataset_type][i])
#print("SALMAAAAAN33333333NNNNNNNNNNNNN = %s"%dataset.token_lengths[dataset_type][i])
#print("SALMAAAAAN44444444NNNNNNNNNNNNN = %s"%dataset.label_vector_indices[dataset_type][i])
# ti.append(dataset.token_indices[dataset_type][i])
# chi.append(dataset.character_indices_padded[dataset_type][i])
# tl.append(dataset.token_lengths[dataset_type][i])
# lvi.append(dataset.label_vector_indices[dataset_type][i])
A.append(unary_scores[1:, :-2][:-1])
# B.append(predictions)
#
if parameters['use_crf']:
predictions, _ = tf.contrib.crf.viterbi_decode(
unary_scores, transition_params_trained)
predictions = predictions[1:-1]
else:
predictions = predictions.tolist()
#
C.append(predictions)
#
assert (len(predictions) == len(dataset.tokens[dataset_type][i]))
output_string = ''
prediction_labels = [
dataset.index_to_label[prediction] for prediction in predictions
]
gold_labels = dataset.labels[dataset_type][i]
if parameters['tagging_format'] == 'bioes':
prediction_labels = utils_nlp.bioes_to_bio(prediction_labels)
gold_labels = utils_nlp.bioes_to_bio(gold_labels)
for prediction, token, gold_label in zip(
prediction_labels, dataset.tokens[dataset_type][i],
gold_labels):
while True:
line = original_conll_file.readline()
split_line = line.strip().split(' ')
if '-DOCSTART-' in split_line[0] or len(
split_line) == 0 or len(split_line[0]) == 0:
continue
else:
token_original = split_line[0]
if parameters['tagging_format'] == 'bioes':
split_line.pop()
gold_label_original = split_line[-1]
assert (token == token_original
and gold_label == gold_label_original)
break
split_line.append(prediction)
output_string += ' '.join(split_line) + '\n'
output_file.write(output_string + '\n')
all_predictions.extend(predictions)
all_y_true.extend(dataset.label_indices[dataset_type][i])
#
# np.save('SalmanTest/SalmanPred%s%s'%(i,dataset_type),predictions)
#A.append(np.zeros(99))
#C.append(np.zeros(99))
np.save('SalmanTest/NAMEMr1mainSalmanUnary_scores%s' % dataset_type, A)
# np.save('SalmanTest/SalmanUnary_scores%s%s'%(i,dataset_type),unary_scores)
# np.save('SalmanTest/myx3mainSalmanPred%s'%dataset_type,B)
np.save('SalmanTest/NAMEMr1mainSalmanCCC%s' % dataset_type, C)
# np.save('SalmanTest/myx3ti',ti)
# np.save('SalmanTest/myx3chi',chi)
# np.save('SalmanTest/myx3tl',tl)
# np.save('SalmanTest/myx3lvi',lvi)
#
output_file.close()
original_conll_file.close()
if dataset_type != 'deploy':
if parameters['main_evaluation_mode'] == 'conll':
conll_evaluation_script = os.path.join('.', 'conlleval')
conll_output_filepath = '{0}_conll_evaluation.txt'.format(
output_filepath)
shell_command = 'perl {0} < {1} > {2}'.format(
conll_evaluation_script, output_filepath,
conll_output_filepath)
os.system(shell_command)
with open(conll_output_filepath, 'r') as f:
classification_report = f.read()
print(classification_report)
else:
new_y_pred, new_y_true, new_label_indices, new_label_names, _, _ = remap_labels(
all_predictions, all_y_true, dataset,
parameters['main_evaluation_mode'])
print(
sklearn.metrics.classification_report(
new_y_true,
new_y_pred,
digits=4,
labels=new_label_indices,
target_names=new_label_names))
return all_predictions, all_y_true, output_filepath
