def train_model():
emb_matrix, word_map = get_embeddings('datastories.twitter', 300)
pipeline = Pipeline([('preprocessor', tweetsPreprocessor(load=True)),
('extractor',
EmbExtractor(word_idxs=word_map, maxlen=50))])
X_train, X_val, y_train, y_val = load_train_test(pipeline=pipeline,
test_size=0.2)
model = build_attention_rnn(emb_matrix,
classes=3,
maxlen=50,
unit=LSTM,
layers=2,
trainable_emb=False,
bidirectional=True,
attention='simple',
dropout_attention=0.5,
layer_dropout_rnn=0.5,
dropout_rnn=0.5,
rec_dropout_rnn=0.5,
clipnorm=1,
lr=0.01,
loss_l2=0.0001)
print(model.summary())
print('Training model...')
model.fit(X_train,
y_train,
validation_data=(X_val, y_val),
epochs=18,
batch_size=128)
print('Model trained')
print('saving model...')
model.save(
os.path.join(os.path.abspath('data/model_weights'),
'new_bi_model_1.h5'))
print('doone')
continue
index = word_indices[k]
emb_disease = embeddings[index]
emb_period = [sum(x) for x in zip(emb_period, emb_disease)]
emb_list.append(emb_period)
return emb_list
parser = argparse.ArgumentParser()
common_utils.add_common_arguments(parser)
parser.add_argument('--target_repl_coef', type=float, default=0.0)
args = parser.parse_args()
print(args)
target_repl = (args.target_repl_coef > 0.0 and args.mode == 'train')
embeddings, word_indices = get_embeddings(corpus='claims_codes_hs', dim=300)
train_reader = ReadmissionReader(
dataset_dir='/mnt/MIMIC-III-clean/readmission_cv2/data/',
listfile='/mnt/MIMIC-III-clean/readmission_cv2/0_train_listfile801010.csv')
val_reader = ReadmissionReader(
dataset_dir='/mnt/MIMIC-III-clean/readmission_cv2/data/',
listfile='/mnt/MIMIC-III-clean/readmission_cv2/0_val_listfile801010.csv')
discretizer = Discretizer(timestep=float(args.timestep),
store_masks=True,
imput_strategy='previous',
start_time='zero')
N = train_reader.get_number_of_examples()
TASK = "CE" # Specify the Subtask. It is needed to correctly load the data
############################################################################
# PERSISTENCE
############################################################################
# if True save model checkpoints, as well as the corresponding word indices
# you HAVE tp set PERSIST = True, in order to be able to use the trained model later
PERSIST = False
best_model = lambda: "cp_model_task4_sub{}.hdf5".format(TASK)
best_model_word_indices = lambda: "cp_model_task4_sub{}_word_indices.pickle".format(
TASK)
############################################################################
# LOAD DATA
############################################################################
embeddings, word_indices = get_embeddings(corpus=WV_CORPUS, dim=WV_DIM)
if TASK == "BD":
loader = Task4Loader(word_indices,
text_lengths=(target_max_length, text_max_length),
subtask=TASK,
filter_classes={"positive", "negative"},
y_one_hot=False)
classes = ['positive', 'negative']
else:
loader = Task4Loader(word_indices,
text_lengths=(target_max_length, text_max_length),
subtask=TASK)
classes = ["-2", "-1", "0", "1", "2"]
if PERSIST:
neu=opinions[1],
neg=opinions[2])
except:
print('fail')
return render_template(
'analyze.html',
error=f'Remember You can only load .csv file and it has to \
contain one of the followings columns: {approved_col_names}')
if __name__ == '__main__':
url = 'http://localhost:5002/api/'
model_weights = os.path.abspath('data/model_weights/new_bi_model_1.h5')
model = load_model(model_weights,
custom_objects={'Attention': Attention()})
global graph
graph = tf.get_default_graph()
MAXLEN = 50
CORPUS = 'datastories.twitter'
DIM = 300
_, word_map = get_embeddings(CORPUS, DIM)
pipeline = Pipeline([('preprocessor', tweetsPreprocessor(load=False)),
('extractor',
EmbExtractor(word_idxs=word_map, maxlen=MAXLEN))])
app.run(debug=True, host='localhost', port=5002)
#TODO:
'''
- add docs for functions
'''
