def apply_multi_label_classification(data_frame, text_vectorizer,
class_vectorizer, classif_type,
classif_level, source_path):
################################################# classification: from text to sparse binary matrix [[0, 1, 0],[1, 0, 1]]
baseline_name = text_vectorizer + '/' + class_vectorizer + '/onevsrest/'
vectorizer_results = ch.apply_df_vectorizer(data_frame, text_vectorizer,
class_vectorizer,
'[both]' + baseline_name)
X_train, X_test, y_train, y_test, classes, n_classes, vocab_processor, len_vocabulary = vectorizer_results
y_train = pd.DataFrame(y_train, columns=classes)
y_test = pd.DataFrame(y_test, columns=classes)
###################################################### text: from text to sparse binary matrix [[0, 1, 0],[1, 0, 1]]
# temp_text = ch.apply_count_vectorizer(data_frame)
###################################################### text: from text to sparse binary matrix [[0, 1, 0],[1, 0, 1]]
apply_onevsrest(X_train, y_train, X_test, y_test, classes, baseline_name,
source_path)
# apply_binary_relevance(X_train, y_train, X_test, y_test, baseline_name, source_path) # needs lots of patents!
# apply_classifier_chain(X_train, y_train, X_test, y_test, baseline_name, source_path) # needs lots of patents!
# apply_label_powerset(X_train, y_train, X_test, y_test, baseline_name, source_path) # best one
apply_adapted_algorithm(X_train, y_train, X_test, y_test, baseline_name,
source_path)
def second_attempt_from_web(data_frame, text_vectorizer, class_vectorizer, classif_level, classif_type, dataset_location):
root_location = fh.get_root_location('data/convolutional_outcome/')
# imdb = keras.datasets.imdb
# (train_data, train_labels), (test_data, test_labels) = imdb.load_data(num_words=10000)
# explore data
# print("Training entries: {}, labels: {}".format(len(train_data), len(train_labels)))
# print("how data looks like: ", train_data[0]) # [1, 14, 34...]
# print("how labels looks like: ", train_labels[0]) # 0
# preprocess data
# A dictionary mapping words to an integer index
# word_index = imdb.get_word_index()
# The first indices are reserved, thus it increases all the indices by 3
# word_index = {k : (v + 3) for k, v in word_index.items()}
# word_index["<PAD>"] = 0
# word_index["<START>"] = 1
# word_index["<UNK>"] = 2 # unknown
# word_index["<UNUSED>"] = 3
# make both the train and the test dataset the same length
# train_data = keras.preprocessing.sequence.pad_sequences(train_data,
# value=word_index["<PAD>"],
# padding='post',
# maxlen=256)
# test_data = keras.preprocessing.sequence.pad_sequences(test_data,
# value=word_index["<PAD>"],
# padding='post',
# maxlen=256)
# print(train_data[0]) # [1 14 34 0 0 0] - with little difference: PAD,START,UNK,UNUSED
model_name = text_vectorizer+'/'+class_vectorizer+'/NN'
standard_results = ch.apply_df_vectorizer(data_frame, text_vectorizer, class_vectorizer, model_name)
train_data, test_data, train_labels, test_labels, classes, n_classes, vocab_processor, len_vocabulary = standard_results
train_data, val_data, train_labels, val_labels = ch.get_train_test_from_data(train_data, train_labels)
# print("Training entries: {}, labels: {}".format(len(train_data), len(train_labels)))
# print("how data looks like: ", train_data[0]) # [1 14 34 0 0 0]
# print("how labels looks like: ", train_labels[0:5]) # list of lists [[0 1 0 0 0 1 0 1], ...]
model = pmh.get_text_convolutional_from_web(len_vocabulary, n_classes)
metrics, predictions = pmh.run_text_cnn_model(model, train_data, train_labels, test_data, test_labels)
classifier_name, layers = ch.get_sequential_classifier_information(model)
mh.display_convolutional_metrics(classifier_name, metrics[0], metrics[1], metrics[2], test_labels, predictions)
ch.save_results(classifier_name, metrics, layers, model_name, classif_level, classif_type, dataset_location)
def fourth_attemp_from_web(data_frame, text_vectorizer, class_vectorizer, classif_level, classif_type, dataset_location):
model_name = text_vectorizer+'/'+class_vectorizer+'/CNN'
standard_results = ch.apply_df_vectorizer(data_frame, text_vectorizer, class_vectorizer, model_name)
train_data, test_data, train_labels, test_labels, classes, n_classes, vocab_processor, len_vocabulary = standard_results
# TODO: is it useful?
# test_labels = np_utils.to_categorical(test_labels, n_classes)
model = pmh.get_fourth_attempt_model_from_web(train_data, n_classes)
y_train_predclass, y_test_predclass, train_metrics, test_metrics, train_predictions, test_predictions = pmh.run_fourth_attempt_model(model, train_data, train_labels, test_data, test_labels)
# mh.display_convolution_metrics_fourth_attempt(train_labels, test_labels, y_train_predclass, y_test_predclass)
# mh.display_convolutional_metrics(classifier_name, train_metrics[0], train_metrics[1], train_metrics[2], train_labels, train_predictions)
mh.display_convolutional_metrics(classifier_name, test_metrics[0], test_metrics[1], test_metrics[2], test_labels, test_predictions)
classifier_name, layers = ch.get_sequential_classifier_information(model)
ch.save_results(classifier_name, test_metrics, layers, model_name, classif_level, classif_type, dataset_location)
def train_testing_convolution(data_frame, text_vectorizer, class_vectorizer):
save_standard_sets = True
root_location = fh.get_root_location('data/convolutional_outcome/')
sets_location = fh.join_paths(root_location, "model_sets")
checkpoint_path = fh.join_paths(root_location, "model_checkpoints")
model_path = fh.link_paths(checkpoint_path, 'convolution_model')
weights_path = fh.link_paths(checkpoint_path, 'convolution_weights')
# get sets
model_name = text_vectorizer+'/'+class_vectorizer+'/NN'
standard_results = ch.apply_df_vectorizer(data_frame, text_vectorizer, class_vectorizer, model_name)
train_data, test_data, train_labels, test_labels, classes, n_classes, vocab_processor, len_vocabulary = standard_results
train_data, val_data, train_labels, val_labels = ch.get_train_test_from_data(train_data, train_labels)
# save sets
# ch.save_sets(sets_location, train_data, test_data, val_data, train_labels, test_labels, val_labels,
# [classes, n_classes, vocab_processor, len_vocabulary])
# this is for test
train_data, test_data, val_data, train_labels, test_labels, val_labels, _ = ch.load_sets(sets_location)
# it could be that a label is only in the test/data data, might be a problem
sequence_length = train_data.shape[1]
# define the model
model = pmh.get_cnn_test(len_vocabulary, n_classes, sequence_length)
# calculates metrics with validating data
model, val_predictions = pmh.run_cnn_test(model,
train_data, train_labels, val_data, val_labels, val_data, val_labels,
model_path, weights_path, True)
binary_val_predictions = mh.get_binary_0_5(val_predictions)
print(val_labels.shape)
print(val_predictions.shape)
# display validation metrics
metrics = mh.get_sequential_metrics(val_labels, val_predictions, binary_predictions)
mh.display_sequential_metrics('validation convolution sequence', metrics)
def test_LSTM_from_web(data_frame, text_vectorizer, class_vectorizer, classif_level, classif_type, dataset_location):
print('### LSTM Doing Testing ###')
root_location = fh.get_root_location('data/lstm_outcome/')
nn_parameter_search_location = fh.join_paths(root_location, "nn_fhv_parameter_search")
doc2vec_model_location = fh.link_paths(fh.join_paths(root_location, "doc2vec_model/vocab_model/"), "doc2vec_model")
save_results = True
sequence_size = 1
EMBEDDING_SIZE = 150
model_name = text_vectorizer+'/'+class_vectorizer+'/LSTM'
results = ch.apply_df_vectorizer(data_frame, text_vectorizer, class_vectorizer, model_name)
X_train, X_test, y_train, y_test, classes, n_classes, vocab_processor, len_vocabulary = results
X_train, X_val, y_train, y_val = ch.get_train_test_from_data(X_train, y_train)
training_docs_list = X_train['patent_id']
test_docs_list = X_test['patent_id']
val_docs_list = X_val['patent_id']
X_data, Xv_data, Xt_data = ch.get_df_data(3, training_docs_list, val_docs_list, test_docs_list, sequence_size, EMBEDDING_SIZE, doc2vec_model_location)
GLOBAL_VARS.DOC2VEC_MODEL_NAME, GLOBAL_VARS.MODEL_NAME = wmh.set_parameters_lstm_doc2vec(nn_parameter_search_location, classif_level, classif_type)
NN_INPUT_NEURONS, NN_SEQUENCE_SIZE, NN_OUTPUT_NEURONS = pmh.get_lstm_shapes(X_data, n_classes)
NN_BATCH_SIZE, PARTS_LEVEL, NN_MAX_EPOCHS, QUEUE_SIZE = pmh.get_lstm_basic_parameters()
params = pmh.get_lstm_testing_parameters()
lstm_output_size,w_dropout_do,u_dropout_do, stack_layers, conv_size, conv_filter_length, conv_max_pooling_length = params
EARLY_STOPPER_MIN_DELTA, EARLY_STOPPER_PATIENCE = pmh.get_early_stopping_parameters()
TEST_METRICS_FILENAME = '{}_level_{}_standard_nn_test_metrics_dict.pkl'
test_metrics_dict = dict()
test_metrics_path = fh.link_paths(fh.link_paths(nn_parameter_search_location, GLOBAL_VARS.MODEL_NAME), TEST_METRICS_FILENAME.format(classif_type, PARTS_LEVEL))
param_results_path = fh.link_paths(fh.link_paths(nn_parameter_search_location, GLOBAL_VARS.MODEL_NAME), NN_PARAMETER_SEARCH_PREFIX.format(classif_type, classif_level, NN_BATCH_SIZE))
param_results_dict = pickle.load(open(param_results_path, 'rb'))
GLOBAL_VARS.NN_MODEL_NAME = 'lstm_size_{}_w-drop_{}_u-drop_{}_stack_{}_conv_{}'.format(lstm_output_size,
w_dropout_do,
u_dropout_do,
stack_layers,
str(conv_size)
)
if conv_size:
GLOBAL_VARS.NN_MODEL_NAME += '_conv-filter-length_{}_max-pooling-size_{}'.format(conv_filter_length,
conv_max_pooling_length)
if GLOBAL_VARS.NN_MODEL_NAME not in param_results_dict.keys():
print("Can't find model: {}".format(GLOBAL_VARS.NN_MODEL_NAME))
raise Exception()
if fh.ensure_exists_path_location(test_metrics_path):
test_metrics_dict = pickle.load(open(fh.link_paths(fh.link_paths(nn_parameter_search_location, GLOBAL_VARS.MODEL_NAME), TEST_METRICS_FILENAME.format(classif_type,PARTS_LEVEL)), 'rb'))
if GLOBAL_VARS.NN_MODEL_NAME in test_metrics_dict.keys():
print("Test metrics already exist for: {}".format(GLOBAL_VARS.NN_MODEL_NAME))
test_metrics = test_metrics_dict[GLOBAL_VARS.NN_MODEL_NAME]
print("** Test Metrics: Cov Err: {:.3f}, Avg Labels: {:.3f}, \n\t\t Top 1: {:.3f}, Top 3: {:.3f}, Top 5: {:.3f}, \n\t\t F1 Micro: {:.3f}, F1 Macro: {:.3f}".format(
test_metrics['coverage_error'], test_metrics['average_num_of_labels'],
test_metrics['top_1'], test_metrics['top_3'], test_metrics['top_5'],
test_metrics['f1_micro'], test_metrics['f1_macro']))
raise Exception()
print('***************************************************************************************')
print(GLOBAL_VARS.NN_MODEL_NAME)
model = pmh.get_keras_rnn_model(NN_INPUT_NEURONS, NN_SEQUENCE_SIZE, NN_OUTPUT_NEURONS,
lstm_output_size, w_dropout_do, u_dropout_do, stack_layers, conv_size,
conv_filter_length, conv_max_pooling_length)
# get model best weights
weights = param_results_dict[GLOBAL_VARS.NN_MODEL_NAME]['best_weights']
model.set_weights(weights)
print('Evaluating on Test Data using best weights')
_, ytp, ytp_binary = pmh.predict_generator(None, model, Xt_data, y_test, NN_BATCH_SIZE, QUEUE_SIZE, test_docs_list)
print('Generating Test Metrics')
test_metrics = mh.get_sequential_metrics(y_test, ytp, ytp_binary)
mh.display_sequential_metrics(test_metrics)
if save_results:
classifier_name, parameters = ch.get_sequential_classifier_information(model)
ch.save_results(classifier_name+'_LSTM', test_metrics, parameters, model_name, classif_level, classif_type, dataset_location)
test_metrics_dict[GLOBAL_VARS.NN_MODEL_NAME] = test_metrics
pickle.dump(test_metrics_dict, open(test_metrics_path, 'wb'))
def train_LSTM_from_web(data_frame, text_vectorizer, class_vectorizer, classif_level, classif_type, dataset_location):
print('### LSTM Doing Training ###')
root_location = fh.get_root_location('data/lstm_outcome/')
exports_location = fh.join_paths(root_location, "exported_data/")
matrices_save_location = fh.join_paths(root_location, "fhv_matrices/")
nn_parameter_search_location = fh.join_paths(root_location, "nn_fhv_parameter_search")
doc2vec_model_location = fh.link_paths(fh.join_paths(root_location, "doc2vec_model/vocab_model/"), "doc2vec_model")
load_existing_results = True # it was True
save_results = True
sequence_size = 1
EMBEDDING_SIZE = 150
model_name = text_vectorizer+'/'+class_vectorizer+'/LSTM'
results = ch.apply_df_vectorizer(data_frame, text_vectorizer, class_vectorizer, model_name)
X_train, X_val, y_train, y_val, classes, n_classes, vocab_processor, len_vocabulary = results
training_docs_list = X_train['patent_id']
val_docs_list = X_val['patent_id']
X_data, Xv_data, _ = ch.get_df_data(2, training_docs_list, val_docs_list, None, sequence_size, EMBEDDING_SIZE, doc2vec_model_location)
GLOBAL_VARS.DOC2VEC_MODEL_NAME, GLOBAL_VARS.MODEL_NAME = wmh.set_parameters_lstm_doc2vec(nn_parameter_search_location, classif_level, classif_type)
# print(X_data.shape) # 64, 1, 200
# print(Xt_data.shape) # 20, 1, 200
# print(Xv_data.shape) # 16, 1, 200
NN_INPUT_NEURONS, NN_SEQUENCE_SIZE, NN_OUTPUT_NEURONS = pmh.get_lstm_shapes(X_data, n_classes)
NN_BATCH_SIZE, PARTS_LEVEL, NN_MAX_EPOCHS, QUEUE_SIZE = pmh.get_lstm_basic_parameters()
param_sampler = pmh.get_lstm_training_parameters()
EARLY_STOPPER_MIN_DELTA, EARLY_STOPPER_PATIENCE = pmh.get_early_stopping_parameters()
param_results_dict = dict()
param_results_path = fh.link_paths(fh.join_paths(nn_parameter_search_location, GLOBAL_VARS.MODEL_NAME), NN_PARAMETER_SEARCH_PREFIX.format(classif_type, classif_level, NN_BATCH_SIZE))
index = param_results_path.rfind('/')
fh.create_folder(fh.link_paths(fh.join_paths(nn_parameter_search_location, GLOBAL_VARS.MODEL_NAME), NN_PARAMETER_SEARCH_PREFIX.format(classif_type, classif_level, NN_BATCH_SIZE))[:index])
###########
# print(X_data.shape)
# input_size, sequence_size, output_size = X_data.shape[2], X_data.shape[0], n_classes
# NN_BATCH_SIZE, PARTS_LEVEL, NN_MAX_EPOCHS, QUEUE_SIZE = 1, 2, 200, 100
# lstm_output_sizes = [500, 1000]
# w_dropout_options = [0., 0.5]
# u_dropout_options = [0., 0.5]
# stack_layers_options = [1, 2, 3]
# lstm_output_size, w_dropout_do, u_dropout_do, stack_layers = 500, 0.0, 0.0, 1
# conv_size, conv_filter_length, max_pooling_length = 128, 2, 2
# EARLY_STOPPER_MIN_DELTA, EARLY_STOPPER_PATIENCE = 0.00001, 15
# import tensorflow as tf
# from tensorflow import keras
# from keras.layers import Input, Dense, Dropout, Activation
# from keras.models import Model, Sequential
# from keras.layers.convolutional import MaxPooling1D, Convolution1D
# from keras.layers.recurrent import LSTM
# model = Sequential()
# # model.add(Convolution1D(nb_filter=conv_size, input_shape=(sequence_size, input_size),
# # filter_length=conv_filter_length, border_mode='same', activation='relu'))
# # model.add(MaxPooling1D(pool_length=max_pooling_length))
# model.add(LSTM(lstm_output_size, activation='tanh', recurrent_activation='hard_sigmoid', use_bias=False,
# input_dim=input_size, dropout_W=w_dropout_do, dropout_U=u_dropout_do,
# implementation=1,
# return_sequences=False if 1 == stack_layers else True,
# go_backwards=False, stateful=False, unroll=False,
# name='lstm_{}_w-drop_{}_u-drop_{}_layer_{}'.format(lstm_output_size, str(u_dropout_do),
# str(w_dropout_do), str(1))))
# # model.add(LSTM(lstm_output_size, activation='tanh', recurrent_activation='hard_sigmoid', use_bias=False,
# # input_dim=input_size, dropout_W=w_dropout_do, dropout_U=u_dropout_do,
# # implementation=1,
# # return_sequences=False if 2 == stack_layers else True,
# # go_backwards=False, stateful=False, unroll=False,
# # name='lstm_{}_w-drop_{}_u-drop_{}_layer_{}'.format(lstm_output_size, str(u_dropout_do),
# # str(w_dropout_do), str(2))))
# # model.add(LSTM(lstm_output_size, activation='tanh', recurrent_activation='hard_sigmoid', use_bias=False,
# # input_dim=input_size, dropout_W=w_dropout_do, dropout_U=u_dropout_do,
# # implementation=1,
# # return_sequences=False if 3 == stack_layers else True,
# # go_backwards=False, stateful=False, unroll=False,
# # name='lstm_{}_w-drop_{}_u-drop_{}_layer_{}'.format(lstm_output_size, str(u_dropout_do),
# # str(w_dropout_do), str(3))))
# model.add(Dense(output_size, activation='sigmoid', name='sigmoid_output'))
# model.compile(optimizer='rmsprop', loss='binary_crossentropy')
# input_matrix = fh.join_paths(matrices_save_location, GLOBAL_VARS.MODEL_NAME)
# early_stopper = keras.callbacks.EarlyStopping(monitor='val_loss', min_delta=EARLY_STOPPER_MIN_DELTA,
# patience=EARLY_STOPPER_PATIENCE, verbose=1, mode='auto')
# metrics_callback = mh.MetricsCallback(input_matrix, classif_type, PARTS_LEVEL, NN_BATCH_SIZE, is_mlp=False)
# history, yvp, yvp_binary = pmh.fit_predict_generator(model, X_data, y_train, Xv_data, y_val, training_docs_list, val_docs_list, early_stopper, metrics_callback, NN_BATCH_SIZE, NN_MAX_EPOCHS, QUEUE_SIZE)
# validation_metrics = mh.get_sequential_metrics(y_val, yvp, yvp_binary)
# mh.display_sequential_metrics(validation_metrics)
# ###########
# useful to skip all the already tested models
if load_existing_results:
param_results_path = fh.link_paths(fh.join_paths(nn_parameter_search_location, GLOBAL_VARS.MODEL_NAME),
NN_PARAMETER_SEARCH_PREFIX.format(classif_type, classif_level, NN_BATCH_SIZE))
if fh.ensure_exists_path_location(param_results_path):
print('Loading Previous results in {}'.format(param_results_path))
param_results_dict = pickle.load(open(param_results_path, 'rb'))
else:
print('No Previous results exist in {}'.format(param_results_path))
for params in param_sampler:
start_time = time.time()
lstm_output_size = params['lstm_output_size']
w_dropout_do = params['w_dropout']
u_dropout_do = params['u_dropout']
stack_layers = params['stack_layers']
conv_size = params['conv_size']
conv_filter_length = params['conv_filter_length']
conv_max_pooling_length = params['conv_max_pooling_length']
GLOBAL_VARS.NN_MODEL_NAME = 'lstm_size_{}_w-drop_{}_u-drop_{}_stack_{}_conv_{}'.format(
lstm_output_size, w_dropout_do, u_dropout_do, stack_layers, str(conv_size))
if conv_size:
GLOBAL_VARS.NN_MODEL_NAME += '_conv-filter-length_{}_max-pooling-size_{}'.format(conv_filter_length,
conv_max_pooling_length)
if GLOBAL_VARS.NN_MODEL_NAME in param_results_dict.keys():
print("skipping: {}".format(GLOBAL_VARS.NN_MODEL_NAME))
continue
# creating the actual keras model
model = pmh.get_keras_rnn_model(NN_INPUT_NEURONS, NN_SEQUENCE_SIZE, NN_OUTPUT_NEURONS,
lstm_output_size, w_dropout_do, u_dropout_do, stack_layers, conv_size,
conv_filter_length, conv_max_pooling_length)
classifier_name, parameters = ch.get_sequential_classifier_information(model)
model_name = text_vectorizer+'/'+class_vectorizer+'/'+classifier_name
input_matrix = fh.join_paths(matrices_save_location, GLOBAL_VARS.MODEL_NAME)
early_stopper = keras.callbacks.EarlyStopping(monitor='val_loss', min_delta=EARLY_STOPPER_MIN_DELTA,
patience=EARLY_STOPPER_PATIENCE, verbose=1, mode='auto')
metrics_callback = mh.MetricsCallback(input_matrix, classif_type, PARTS_LEVEL, NN_BATCH_SIZE, is_mlp=False)
history, yvp, yvp_binary = pmh.fit_predict_generator(model, X_data, y_train, Xv_data, y_val, training_docs_list, val_docs_list, early_stopper, metrics_callback, NN_BATCH_SIZE, NN_MAX_EPOCHS, QUEUE_SIZE)
print('\nGenerating Validation Metrics')
validation_metrics = mh.get_sequential_metrics(y_val, yvp, yvp_binary)
mh.display_sequential_metrics(classifier_name, validation_metrics)
param_results_dict[GLOBAL_VARS.NN_MODEL_NAME] = dict()
# param_results_dict[GLOBAL_VARS.NN_MODEL_NAME]['best_validation_metrics'] = best_validation_metrics
param_results_dict[GLOBAL_VARS.NN_MODEL_NAME]['epochs'] = len(history.history['val_loss'])
param_results_dict[GLOBAL_VARS.NN_MODEL_NAME]['best_weights'] = metrics_callback.best_weights
param_results_dict[GLOBAL_VARS.NN_MODEL_NAME]['best_val_loss'] = metrics_callback.best_val_loss
param_results_dict[GLOBAL_VARS.NN_MODEL_NAME]['training_loss'] = metrics_callback.losses
param_results_dict[GLOBAL_VARS.NN_MODEL_NAME]['validation_loss'] = metrics_callback.val_losses
duration = time.time() - start_time
param_results_dict[GLOBAL_VARS.NN_MODEL_NAME]['duration'] = duration
ch.delete_variables(history, metrics_callback)
ch.save_results(classifier_name+'_LSTM', validation_metrics, parameters, model_name, classif_level, classif_type, dataset_location)
if save_results:
file = open(param_results_path, 'wb')
pickle.dump(param_results_dict, file)
def preprocessing_data_for_fasttext(data_frame, text_vectorizer,
class_vectorizer):
root_location = fh.get_root_location('data/fasttext_outcome/')
data_frame['text'] = data_frame['text'].replace(
'\n', ' ', regex=True).replace('\t', ' ', regex=True)
model_name = text_vectorizer + '/' + class_vectorizer + '/FastText'
try:
X_train, X_test, Y_train, Y_test, _, _, _, _ = ch.apply_df_vectorizer(
data_frame, text_vectorizer, class_vectorizer, model_name)
X_train, X_val, Y_train, Y_val = ch.get_train_test_from_data(
X_train, Y_train)
# self.data_vectors = pd.DataFrame(columns=range(vectors_size), index=range(corpus_size))
if not isinstance(X_train, pd.DataFrame):
train = pd.DataFrame(data=X_train)
test = pd.DataFrame(data=X_test)
val = pd.DataFrame(data=X_val)
# test_labels = pd.DataFrame(columns=[''])
else:
train = X_train
test = X_test
val = X_val
train.loc[:, 1] = Y_train
test.loc[:, 1] = Y_test
val.loc[:, 1] = Y_val
train.drop(columns=['patent_id'], inplace=True)
test.drop(columns=['patent_id'], inplace=True)
val.drop(columns=['patent_id'], inplace=True)
data_frame.to_csv(fh.link_paths(root_location, 'dataframe.csv'),
index=False,
sep=' ',
header=False,
quoting=csv.QUOTE_NONE,
quotechar="",
escapechar=" ")
train.to_csv(fh.link_paths(root_location, 'training set.csv'),
index=False,
sep=' ',
header=False,
quoting=csv.QUOTE_NONE,
quotechar="",
escapechar=" ")
test.to_csv(fh.link_paths(root_location, 'testing set.csv'),
index=False,
sep=',',
header=False,
quoting=csv.QUOTE_NONE,
quotechar="",
escapechar=" ")
except:
print('a problem occurred while trying to store the dataframes')
X_train, X_test, Y_train, Y_test, _, _, _, _ = ch.apply_df_vectorizer(
data_frame, text_vectorizer, class_vectorizer, model_name)
X_train, X_val, Y_train, Y_val = ch.get_train_test_from_data(
X_train, Y_train)
val = pd.DataFrame({'text': X_val, 'classification': Y_val})
train = pd.DataFrame({'text': X_train, 'classification': Y_train})
test = pd.DataFrame({'text': X_test, 'classification': Y_test})
data_frame.to_csv(fh.link_paths(root_location, 'dataframe.csv'),
index=False,
sep=' ',
header=False,
quoting=csv.QUOTE_NONE,
quotechar="",
escapechar=" ")
val.to_csv(fh.link_paths(root_location, 'validating set.csv'),
index=False,
sep=' ',
header=False,
quoting=csv.QUOTE_NONE,
quotechar="",
escapechar=" ")
train.to_csv(fh.link_paths(root_location, 'training set.csv'),
index=False,
sep=' ',
header=False,
quoting=csv.QUOTE_NONE,
quotechar="",
escapechar=" ")
test.to_csv(fh.link_paths(root_location, 'testing set.csv'),
index=False,
sep=',',
header=False,
quoting=csv.QUOTE_NONE,
quotechar="",
escapechar=" ")
def apply_multi_label_classification_without_pipeline(
data_frame, text_vectorizer, class_vectorizer, classif_level,
classif_type, source_path):
baseline_name = text_vectorizer + '/' + class_vectorizer + '/onevsrest/'
vectorizer_results = ch.apply_df_vectorizer(data_frame, text_vectorizer,
class_vectorizer,
'[both]' + baseline_name)
X_train, X_test, y_train, y_test, classes, n_classes, vocab_processor, len_vocabulary = vectorizer_results
# len_vocabulary = 57335
# len_vocabulary = 34736
print(X_train[0])
print('len_vocabulary: ', len_vocabulary, ' num_classes: ', n_classes)
# Run classifier
classifier = OneVsRestClassifier(pmh.get_logistic()) # here
# classifier = OneVsRestClassifier(pmh.get_SVC()) # here
# classifier = OneVsRestClassifier(pmh.get_multinomialNB()) # ERROR WORD2VEC/DOC2VEC (X has negative value)
# classifier = OneVsRestClassifier(pmh.get_decision_tree()) # here
# classifier = OneVsRestClassifier(pmh.get_kneighbors()) # here
# classifier = OneVsRestClassifier(pmh.get_linear_SVC())
# classifier = OneVsRestClassifier(pmh.get_random_forest_classifier()) # here
# classifier = OneVsRestClassifier(pmh.get_SGD_classifier())
train_predictions = np.ndarray(shape=(n_classes, y_train.shape[0]),
dtype=int)
predictions = np.ndarray(shape=(n_classes, y_test.shape[0]), dtype=int)
###
precision = dict()
recall = dict()
average_precision = dict()
classifier_name, parameters = ch.get_complex_classifier_information(
str(classifier), 1, 1, 2, 0)
second_training = False
just_once = True
another_try = False # single train and estimation instead of multi-train-estimation steps (it performs better with svm and logistic)
if not another_try:
for _ in range(1):
for i in range(n_classes):
if second_training:
if classifier_name in [
'DecisionTreeClassifier', 'KNeighborsClassifier',
'MultinomialNB', 'RandomForestClassifier'
]:
# do not provide the second metrics
break
elif just_once:
# fit again with the whole set of classes - should be better
classifier.fit(X_train, y_train)
y_score = classifier.decision_function(X_test)
precision[i], recall[i], average_precision[
i] = mh.calculate_recall_curve_precision_score(
y_test[:, i], y_score[:, i], None,
y_test[:, i], y_score[:, i])
just_once = False
else:
predictions[i] = pmh.fit_predict_functions(
classifier, X_train, y_train[:, i], X_test)
train_predictions[i] = classifier.predict(X_train)
print('**Processing classes {0:0.2f} % ...**'.format(
((i + 1) / n_classes) * 100))
second_training = True
break
predictions = predictions.transpose()
print('transposed')
else:
predictions = pmh.fit_predict_functions(classifier, X_train, y_train,
X_test)
train_predictions = classifier.predict(X_train)
# metrics
model_name = '[each class predictions]' + baseline_name + classifier_name
manual_metrics = mh.calculate_manual_metrics(model_name, y_test,
predictions)
none_average, binary_average, micro_average, macro_average = manual_metrics
# metrics
list_metrics = mh.calculate_metrics(model_name, y_test, predictions)
none_average, binary_average, micro_average, macro_average = list_metrics
ch.save_results(classifier_name, list_metrics, parameters, model_name,
classif_level, classif_type, source_path)
if not just_once:
print('not just once')
train_predictions = classifier.predict(X_train)
predictions = classifier.predict(X_test)
# metrics
mh.calculate_metrics_with_recall_curve(y_score, y_train, y_test,
train_predictions, predictions)
# metrics
model_name = '[all classes predictions]' + baseline_name + classifier_name
list_metrics = mh.calculate_metrics(model_name, y_test, predictions)
none_average, binary_average, micro_average, macro_average = list_metrics
ch.save_results(classifier_name, list_metrics, parameters, model_name,
classif_level, classif_type, source_path)
def first_attempt_based_on_text_classification_paper(data_frame, text_vectorizer, class_vectorizer):
# Parameters - can be placed at the beginning of the script
# ==================================================
# Data loading params
tf.flags.DEFINE_float("dev_sample_percentage", .2, "Percentage of the training data to use for validation")
tf.flags.DEFINE_string("positive_data_file", "./data/rt-polaritydata/rt-polarity.pos", "Data source for the positive data.")
tf.flags.DEFINE_string("negative_data_file", "./data/rt-polaritydata/rt-polarity.neg", "Data source for the negative data.")
# Model Hyperparameters
tf.flags.DEFINE_integer("embedding_dim", 128, "Dimensionality of character embedding (default: 128)")
tf.flags.DEFINE_string("filter_sizes", "3,4,5", "Comma-separated filter sizes (default: '3,4,5')")
tf.flags.DEFINE_integer("num_filters", 128, "Number of filters per filter size (default: 128)")
tf.flags.DEFINE_float("dropout_keep_prob", 0.5, "Dropout keep probability (default: 0.5)")
tf.flags.DEFINE_float("l2_reg_lambda", 0.0, "L2 regularization lambda (default: 0.0)")
# Training parameters
tf.flags.DEFINE_integer("batch_size", 64, "Batch Size (default: 64)")
tf.flags.DEFINE_integer("num_epochs", 200, "Number of training epochs (default: 200)")
tf.flags.DEFINE_integer("evaluate_every", 100, "Evaluate model on dev set after this many steps (default: 100)")
tf.flags.DEFINE_integer("checkpoint_every", 100, "Save model after this many steps (default: 100)")
tf.flags.DEFINE_integer("num_checkpoints", 5, "Number of checkpoints to store (default: 5)")
# Misc Parameters
tf.flags.DEFINE_boolean("allow_soft_placement", True, "Allow device soft device placement")
tf.flags.DEFINE_boolean("log_device_placement", False, "Log placement of ops on devices")
FLAGS = tf.flags.FLAGS
# FLAGS._parse_flags()
# print("\nParameters:")
# for attr, value in sorted(FLAGS.__flags.items()):
# print("{}={}".format(attr.upper(), value))
# print("")
dev_sample_percentage = FLAGS.dev_sample_percentage
# x_train, y_train, vocab_processor, len_vocabulary, x_dev, y_dev, classe, n_classes = preprocess(data_frame, dev_sample_percentage)
model_name = text_vectorizer+'/'+class_vectorizer+'/TextCNN'
standard_results = ch.apply_df_vectorizer(data_frame, 'standard', 'multi_label', model_name)
x_train, x_dev, y_train, y_dev, classes, n_classes, vocab_processor, len_vocabulary = standard_results
with tf.Graph().as_default():
session_conf = tf.ConfigProto(
allow_soft_placement = FLAGS.allow_soft_placement,
log_device_placement = FLAGS.log_device_placement)
sess = tf.Session(config = session_conf)
with sess.as_default():
# Code that operates on the default graph and session comes here…
cnn = pmh.TextCNN(
sequence_length = x_train.shape[1],
num_classes = n_classes,
vocab_size = (len_vocabulary),
embedding_size = FLAGS.embedding_dim,
filter_sizes = list(map(int, FLAGS.filter_sizes.split(","))),
num_filters = FLAGS.num_filters)
# define training procedure
global_step = tf.Variable(0, name="global_step", trainable=False)
optimizer = tf.train.AdamOptimizer(1e-4)
grads_and_vars = optimizer.compute_gradients(cnn.loss)
train_op = optimizer.apply_gradients(grads_and_vars, global_step=global_step)
# output directory for models and summaries
root_location = fh.get_root_location('data/convolutional_runs/')
timestamp = datetime.datetime.now().isoformat()
out_dir = fh.link_paths(root_location, timestamp)
print("Writing to {}\n".format(out_dir))
# summaries for loss and accuracy
loss_summary = tf.summary.scalar("loss", cnn.loss)
acc_summary = tf.summary.scalar("accuracy", cnn.accuracy)
# train summaries
train_summary_op = tf.summary.merge([loss_summary, acc_summary])
train_summary_dir = fh.join_paths(fh.link_paths(out_dir, 'summaries'), 'train')
train_summary_writer = tf.summary.FileWriter(train_summary_dir, sess.graph_def)
# dev summaries
dev_summary_op = tf.summary.merge([loss_summary, acc_summary])
dev_summary_dir = fh.join_paths(fh.link_paths(out_dir, 'summaries'), 'dev')
dev_summary_writer = tf.summary.FileWriter(dev_summary_dir, sess.graph_def)
# checkpointing
checkpoint_dir = fh.link_paths(out_dir, 'checkpoints')
checkpoint_prefix = fh.join_paths(checkpoint_dir, 'model')
saver = tf.train.Saver(tf.all_variables())
# write vocabulary
try:
vocab_processor.save(os.path.join(out_dir, "vocab"))
except:
pass
# initialize all variables
sess.run(tf.global_variables_initializer())
def train_step(x_batch, y_batch):
"""
A single training step
"""
feed_dict = {
cnn.input_x: x_batch,
cnn.input_y: y_batch,
cnn.dropout_keep_prob: FLAGS.dropout_keep_prob
}
_, step, summaries, loss, accuracy = sess.run(
[train_op, global_step, train_summary_op, cnn.loss, cnn.accuracy],
feed_dict)
time_str = datetime.datetime.now().isoformat()
print("{}: step {}, loss {:g}, acc {:g}".format(time_str, step, loss, accuracy))
train_summary_writer.add_summary(summaries, step)
def dev_step(x_batch, y_batch, writer=None):
"""
Evaluates model on a dev set
"""
feed_dict = {
cnn.input_x: x_batch,
cnn.input_y: y_batch,
cnn.dropout_keep_prob: 1.0
}
step, summaries, loss, accuracy = sess.run(
[global_step, dev_summary_op, cnn.loss, cnn.accuracy],
feed_dict)
time_str = datetime.datetime.now().isoformat()
print("{}: step {}, loss {:g}, acc {:g}".format(time_str, step, loss, accuracy))
if writer:
writer.add_summary(summaries, step)
# generate batches
batches = batch_iter(
list(zip(x_train, y_train)), FLAGS.batch_size, FLAGS.num_epochs)
# training loop. For each batch…
for batch in batches:
x_batch, y_batch = zip(*batch)
train_step(x_batch, y_batch)
current_step = tf.train.global_step(sess, global_step)
if current_step % FLAGS.evaluate_every == 0:
print("\nEvaluation:")
dev_step(x_dev, y_dev, writer=dev_summary_writer)
if current_step % FLAGS.checkpoint_every == 0:
path = saver.save(sess, checkpoint_prefix, global_step=current_step)
print("Saved model checkpoint to {}\n".format(path))
def third_attempt_from_web(data_frame, text_vectorizer, class_vectorizer, classif_level, classif_type, dataset_location):
# fashion_mnist = keras.datasets.fashion_mnist
# (train_images, train_labels), (test_images, test_labels) = fashion_mnist.load_data()
# class_names = ['T-shirt/top', 'Trouser', 'Pullover', 'Dress', 'Coat',
# 'Sandal', 'Shirt', 'Sneaker', 'Bag', 'Ankle boot']
# print("### images ###")
# print(test_images)
# print(type(test_images))
# print(test_images.shape)
# print(test_images[0])
# print(test_labels)
# print(type(test_labels))
# print(test_labels.shape)
# print(test_labels[0])
model_name = text_vectorizer+'/'+class_vectorizer+'/CNN'
standard_results = ch.apply_df_vectorizer(data_frame, text_vectorizer, class_vectorizer, model_name)
train_data, test_data, train_labels, test_labels, classes, n_classes, vocab_processor, len_vocabulary = standard_results
# print("### texts ###")
# print(test_data)
# print(type(test_data))
# print(test_data.shape)
# print(test_data[0])
# print(test_labels)
# print(type(test_labels))
# print(test_labels.shape)
# print(test_labels[0])
# preprocess data
# train_images = train_images / 255.0
# test_images = test_images / 255.0
# it reduced all in the range 0--1
# normalize the values
train_data = np.reshape(train_data, [train_data.shape[0], 1, train_data.shape[1]])
test_data = np.reshape(test_data, [test_data.shape[0], 1, test_data.shape[1]])
print(train_labels)
print(train_labels.shape)
for index in range(train_labels.shape[1]):
temp_train_labels = train_labels[:, index] # estimating the first class
temp_test_labels = test_labels[:, index] # estimating the first class
print(temp_train_labels)
model = pmh.get_image_convolutional_from_web(train_data, n_classes)
metrics, predictions = pmh.run_image_cnn_model(model, train_data, temp_train_labels, test_data, temp_test_labels)
print("predictions: ", predictions)
classifier_name, layers = ch.get_sequential_classifier_information(model)
mh.display_convolutional_metrics(classifier_name, metrics[0], metrics[1], metrics[2], temp_test_labels, predictions)
ch.save_results(classifier_name, metrics, layers, model_name, classif_level, classif_type, dataset_location)