def __init__(self, unique_section_names, target_dir, mode):
super().__init__()
n_files_ea_section = [] # number of files for each section
n_vectors_ea_file = [] # number of vectors for each file
data = numpy.empty(
(0, CONFIG["feature"]["n_frames"] * CONFIG["feature"]["n_mels"]),
dtype=float,
)
for section_name in unique_section_names:
# get file list for each section
# all values of y_true are zero in training
print("target_dir : %s" % (target_dir + "_" + section_name))
files, _ = util.file_list_generator(
target_dir=target_dir,
section_name=section_name,
dir_name="train",
mode=mode,
)
print("number of files : %s" % (str(len(files))))
n_files_ea_section.append(len(files))
# extract features from audio files and
# concatenate them into Numpy array.
features, n_features = concat_features(files)
data = numpy.append(data, features, axis=0)
n_vectors_ea_file.append(n_features)
n_vectors_ea_file = flatten(n_vectors_ea_file)
# make target labels for conditioning
# they are not one-hot vector!
labels = numpy.zeros((data.shape[0]), dtype=int)
start_index = 0
for section_index in range(unique_section_names.shape[0]):
for file_id in range(n_files_ea_section[section_index]):
labels[
start_index : start_index + n_vectors_ea_file[file_id]
] = section_index
start_index += n_vectors_ea_file[file_id]
# 1D vector to 2D image (1ch)
self.data = data.reshape(
(
data.shape[0],
1, # number of channels
CONFIG["feature"]["n_frames"],
CONFIG["feature"]["n_mels"],
)
)
self.labels = labels
def fit_gamma_dist(model, target_dir, mode):
"""
- Calculate anomaly scores over sections.
- Fit gamma distribution for anomaly scores.
- Save the parameters of the distribution.
"""
section_names = util.get_section_names(target_dir, dir_name="train")
dataset_scores = numpy.array([], dtype=numpy.float64)
# calculate anomaly scores over sections
for section_index, section_name in enumerate(section_names):
section_files, _ = util.file_list_generator(
target_dir=target_dir,
section_name=section_name,
dir_name="train",
mode=mode,
)
section_scores = [0.0 for k in section_files]
for file_idx, file_path in enumerate(section_files):
section_scores[file_idx] = calc_anomaly_score(
model, file_path=file_path, section_index=section_index
)
section_scores = numpy.array(section_scores)
dataset_scores = numpy.append(dataset_scores, section_scores)
dataset_scores = numpy.array(dataset_scores)
# fit gamma distribution for anomaly scores
gamma_params = scipy.stats.gamma.fit(dataset_scores)
gamma_params = list(gamma_params)
# save the parameters of the distribution
score_file_path = "{model}/score_distr_{machine_type}.pkl".format(
model=CONFIG["model_directory"], machine_type=os.path.split(target_dir)[1]
)
joblib.dump(gamma_params, score_file_path)
def main():
"""
Perform model evaluation.
"""
# check mode
# "development": mode == True
# "evaluation": mode == False
mode = util.command_line_chk() # constant: True or False
if mode is None:
sys.exit(-1)
# make result directory
os.makedirs(CONFIG["result_directory"], exist_ok=True)
# load base_directory list
dir_list = util.select_dirs(config=CONFIG, mode=mode)
# initialize lines in csv for AUC and pAUC
csv_lines = []
performance = {"section": None, "all": None}
# anomaly scores and decision results
score_list = {"anomaly": None, "decision": None}
if mode:
performance["all"] = []
for idx, target_dir in enumerate(dir_list):
print("===============================================")
print("[%d/%d] %s" % (idx + 1, len(dir_list), target_dir))
print("================ MODEL LOAD =================")
model = load_model(config=CONFIG,
machine_type=os.path.split(target_dir)[1])
decision_threshold = calc_decision_threshold(target_dir)
if mode:
# results for each machine type
csv_lines.append([os.path.split(target_dir)[1]
]) # append machine type
csv_lines.append([
"section", "domain", "AUC", "pAUC", "precision", "recall",
"F1 score"
])
performance["section"] = []
for dir_name in ["source_test", "target_test"]:
for section_name in util.get_section_names(target_dir,
dir_name=dir_name):
# load test file
test_files, y_true = util.file_list_generator(
target_dir=target_dir,
section_name=section_name,
dir_name=dir_name,
mode=mode,
)
print(
"============== BEGIN TEST FOR A SECTION %s OF %s =============="
% (section_name, dir_name))
# - perform test for a section
# - anomaly scores and decision results are saved in score_list
y_pred = test_section(model, test_files, decision_threshold,
score_list)
# save anomaly scores and decision results
save_anomaly_score(score_list, target_dir, section_name,
dir_name)
if mode:
# evaluation_scores (tuple): auc, p_auc, prec, recall, f1_score
eval_scores = calc_evaluation_scores(
y_true, y_pred, decision_threshold)
csv_lines.append([
section_name.split("_", 1)[1],
dir_name.split("_", 1)[0],
*eval_scores, # unpack
])
performance["section"].append(eval_scores)
performance["all"].append(eval_scores)
print(
"============ END OF TEST FOR A SECTION %s OF %s ============\n"
% (section_name, dir_name))
if mode:
# calculate averages for AUCs and pAUCs
csv_lines = calc_performance_section(performance["section"],
csv_lines)
del model
if mode:
# calculate averages for AUCs and pAUCs over all sections
csv_lines = calc_performance_all(performance["all"], csv_lines)
# output results
save_result(csv_lines)
def main():
"""
Perform model training and validation.
"""
# check mode
# "development": mode == True
# "evaluation": mode == False
mode = util.command_line_chk() # constant: True or False
if mode is None:
sys.exit(-1)
# make output directory
os.makedirs(CONFIG["model_directory"], exist_ok=True)
# load base_directory list
dir_list = util.select_dirs(config=CONFIG, mode=mode)
for idx, target_dir in enumerate(dir_list):
print("===============================================")
print("[%d/%d] %s" % (idx + 1, len(dir_list), target_dir))
print("\n============== DATASET_GENERATOR ==============")
# generate file list under "target_dir" directory.
files, _ = util.file_list_generator(
target_dir=target_dir,
section_name="*",
dir_name="train",
mode=mode,
)
dcase_dataset = DcaseDataset(files) # generate dataset from file list.
print("===============================================")
print("\n=========== DATALOADER_GENERATOR ==============")
data_loader = {"train": None, "val": None}
data_loader["train"], data_loader["val"] = get_dataloader(
dcase_dataset)
print("===============================================")
print("\n================ MODEL TRAINING ===============")
model = get_model().to(DEVICE)
optimizer, _ = get_optimizer(model)
# optimizer, scheduler = get_optimizer(model) # optional
# display summary of model through torchinfo
summary(
model,
input_size=(
CONFIG["training"]["batch_size"],
CONFIG["feature"]["n_mels"] * CONFIG["feature"]["n_frames"],
),
)
# training loop
for epoch in range(1, CONFIG["training"]["epochs"] + 1):
print("Epoch {:2d}: ".format(epoch), end="")
training(
model=model,
data_loader=data_loader["train"],
optimizer=optimizer,
# scheduler=scheduler # optional
)
validation(model=model, data_loader=data_loader["val"])
del data_loader # delete the dataset for training.
# fit gamma distribution for anomaly scores
# and save the parameters of the distribution
fit_gamma_dist(dcase_dataset, model, target_dir)
print("============== SAVE MODEL ==============")
save_model(
model,
model_dir=CONFIG["model_directory"],
machine_type=os.path.split(target_dir)[1],
)
print("============== END TRAINING ==============")