com.logger.info("model exists")
continue
history_img = "{model}/history_{machine_type}.png".format(
model=param["model_directory"], machine_type=machine_type)
# pickle file for storing anomaly score distribution
score_distr_file_path = "{model}/score_distr_{machine_type}.pkl".format(
model=param["model_directory"], machine_type=machine_type)
# generate dataset
print("============== DATASET_GENERATOR ==============")
# get file list for all sections
# all values of y_true are zero in training
files, y_true = com.file_list_generator(target_dir=target_dir,
section_name="*",
dir_name="train",
mode=mode)
dataset = DCASE_Dataset(files, param)
# data loader and split train and validation
train_size = int(len(dataset) * (1 - param["fit"]["validation_split"]))
train_loader = DataLoader(Subset(dataset, list(range(0, train_size))),
batch_size=param["fit"]["batch_size"],
shuffle=param["fit"]["shuffle"],
drop_last=True,
num_workers=os.cpu_count(),
pin_memory=True)
val_loader = DataLoader(Subset(dataset,
])
performance = []
dir_names = ["source_test", "target_test"]
for dir_name in dir_names:
#list machine id
section_names = com.get_section_names(target_dir,
dir_name=dir_name)
for section_name in section_names:
# load test file
files, y_true = com.file_list_generator(
target_dir=target_dir,
section_name=section_name,
dir_name=dir_name,
mode=mode)
# setup anomaly score file path
anomaly_score_csv = "{result}/anomaly_score_{machine_type}_{section_name}_{dir_name}.csv".format(
result=param["result_directory"],
machine_type=machine_type,
section_name=section_name,
dir_name=dir_name)
anomaly_score_list = []
# setup decision result file path
decision_result_csv = "{result}/decision_result_{machine_type}_{section_name}_{dir_name}.csv".format(
result=param["result_directory"],
machine_type=machine_type,
tflite_model = converter.convert()
tflite_file = "{model}/model_{machine_type}.tflite".format(
model=param["model_directory"], machine_type=machine_type)
with tf.io.gfile.GFile(tflite_file, 'wb') as f:
f.write(tflite_model)
# Quantization of weights (but not the activations)
converter.optimizations = [tf.lite.Optimize.DEFAULT]
tflite_model = converter.convert()
tflite_file = "{model}/model_{machine_type}_quant.tflite".format(
model=param["model_directory"], machine_type=machine_type)
with tf.io.gfile.GFile(tflite_file, 'wb') as f:
f.write(tflite_model)
print("============== DATASET_GENERATOR ==============")
files = com.file_list_generator(target_dir)
train_data = com.list_to_vector_array(
files,
msg="generate train_dataset",
n_mels=param["feature"]["n_mels"],
frames=param["feature"]["frames"],
n_fft=param["feature"]["n_fft"],
hop_length=param["feature"]["hop_length"],
power=param["feature"]["power"])
def representative_dataset_gen():
for sample in train_data[::5]:
sample = numpy.expand_dims(sample.astype(numpy.float32),
axis=0)
yield [sample]