def load_data(window_width, file, sample_rate, rolling_step=None):
if rolling_step is None:
rolling_step = window_width // 2
data = aux_fn.get_data_from_files([file], sample_rate, window_width,
rolling_step)
print(data.shape)
return data
def get_response(model, file, file_meta, sample_rate, rolling_step=None):
window_width = model.layers[0].input_shape[1]
if rolling_step is None:
rolling_step = window_width // 2
data = aux_fn.get_data_from_files(
[file], sample_rate, window_width, rolling_step)
print(data.shape)
yhat = model.predict(np.expand_dims(data, axis=2),
batch_size=data.shape[0], verbose=1)
yhat = np.squeeze(yhat)
response = aux_fn.unroll(yhat, rolling_step)
raw_data = aux_fn.unroll(data, rolling_step)
anomaly_bounds = None
if file_meta['event_present']:
ev_start = round(
file_meta['event_start_in_mixture_seconds'] * sample_rate)
ev_end = round(
ev_start + file_meta['event_length_seconds'] * sample_rate)
anomaly_bounds = (min(ev_start, len(raw_data)),
min(ev_end, len(raw_data)))
return {'data': raw_data, 'resp': response, 'actual': anomaly_bounds}
filemode='w',
level=logging.INFO)
class LogHistory(Callback):
def on_epoch_end(self, batch, logs={}):
logging.info(logs)
model = LSTMAutoencoderV2(batchsize, window_width, encoder_layer_list,
decoder_layer_list)
files = aux_fn.get_all_files(path)
print(['Number of files', len(files)])
data = aux_fn.get_data_from_files(files, sr=sample_rate, ww=window_width)
data = np.expand_dims(data, axis=2)
with tf.Session() as sess:
tf.global_variables_initializer().run()
tf.local_variables_initializer().run()
train_writer = tf.summary.FileWriter(logs_path, sess.graph)
start_time = time.time()
for epoch in range(epochs):
epoch_loss = 0
minibatch = 0
iteration = int((data.shape[0]) / batchsize)
for i in range(iteration):
start = minibatch
end = minibatch + batchsize