def run_for_epoch(self, epoch, x, y, type):
self.network.eval()
predictions_dict = {"tp": [], "fp": [], "tn": [], "fn": []}
predictions = []
self.test_batch_loss, self.test_batch_accuracy, self.test_batch_uar, self.test_batch_ua, audio_for_tensorboard_test = [], [], [], [], None
with torch.no_grad():
for i, (audio_data, label) in enumerate(zip(x, y)):
label = tensor(label).float()
test_predictions = self.network(audio_data).squeeze(1)
test_loss = self.loss_function(test_predictions, label)
test_predictions = nn.Sigmoid()(test_predictions)
predictions.append(test_predictions.numpy())
test_accuracy, test_uar = accuracy_fn(test_predictions, label,
self.threshold)
self.test_batch_loss.append(test_loss.numpy())
self.test_batch_accuracy.append(test_accuracy.numpy())
self.test_batch_uar.append(test_uar)
tp, fp, tn, fn = custom_confusion_matrix(
test_predictions, label, threshold=self.threshold)
predictions_dict['tp'].extend(tp)
predictions_dict['fp'].extend(fp)
predictions_dict['tn'].extend(tn)
predictions_dict['fn'].extend(fn)
print(f'***** {type} Metrics ***** ')
print(f'***** {type} Metrics ***** ', file=self.log_file)
print(
f"Loss: {np.mean(self.test_batch_loss)} | Accuracy: {np.mean(self.test_batch_accuracy)} | UAR: {np.mean(self.test_batch_uar)}"
)
print(
f"Loss: {np.mean(self.test_batch_loss)} | Accuracy: {np.mean(self.test_batch_accuracy)} | UAR: {np.mean(self.test_batch_uar)}",
file=self.log_file)
log_summary(self.writer,
epoch,
accuracy=np.mean(self.test_batch_accuracy),
loss=np.mean(self.test_batch_loss),
uar=np.mean(self.test_batch_uar),
lr=self.optimiser.state_dict()['param_groups'][0]['lr'],
type=type)
log_conf_matrix(self.writer,
epoch,
predictions_dict=predictions_dict,
type=type)
y = [element for sublist in y for element in sublist]
predictions = [
element for sublist in predictions for element in sublist
]
write_to_npy(filename=self.debug_filename,
predictions=predictions,
labels=y,
epoch=epoch,
accuracy=np.mean(self.test_batch_accuracy),
loss=np.mean(self.test_batch_loss),
uar=np.mean(self.test_batch_uar),
lr=self.optimiser.state_dict()['param_groups'][0]['lr'],
predictions_dict=predictions_dict,
type=type)
def run_for_epoch(self, epoch, x, y, type):
self.network.eval()
for m in self.network.modules():
if isinstance(m, nn.BatchNorm2d):
m.track_running_stats = False
predictions_dict = {"tp": [], "fp": [], "tn": [], "fn": []}
overall_predictions = []
self.test_batch_loss, self.test_batch_accuracy, self.test_batch_uar, self.test_batch_ua, self.test_batch_f1, self.test_batch_precision, self.test_batch_recall, audio_for_tensorboard_test = [], [], [], [], [], [], [], None
with torch.no_grad():
for i, (audio_data, label) in enumerate(zip(x, y)):
label = to_tensor(label, device=self.device).float()
audio_data = to_tensor(audio_data, device=self.device)
test_predictions = self.network(audio_data).squeeze(1)
test_loss = self.loss_function(test_predictions, label)
test_predictions = nn.Sigmoid()(test_predictions)
overall_predictions.extend(to_numpy(test_predictions))
test_accuracy, test_uar, test_precision, test_recall, test_f1 = accuracy_fn(
test_predictions, label, self.threshold)
self.test_batch_loss.append(to_numpy(test_loss))
self.test_batch_accuracy.append(to_numpy(test_accuracy))
self.test_batch_uar.append(test_uar)
self.test_batch_f1.append(test_f1)
self.test_batch_precision.append(test_precision)
self.test_batch_recall.append(test_recall)
tp, fp, tn, fn = custom_confusion_matrix(
test_predictions, label, threshold=self.threshold)
predictions_dict['tp'].extend(tp)
predictions_dict['fp'].extend(fp)
predictions_dict['tn'].extend(tn)
predictions_dict['fn'].extend(fn)
print(f'***** {type} Metrics ***** ')
print(f'***** {type} Metrics ***** ', file=self.log_file)
print(
f"Loss: {'%.3f' % np.mean(self.test_batch_loss)} | Accuracy: {'%.3f' % np.mean(self.test_batch_accuracy)} | UAR: {'%.3f' % np.mean(self.test_batch_uar)}| F1:{'%.3f' % np.mean(self.test_batch_f1)} | Precision:{'%.3f' % np.mean(self.test_batch_precision)} | Recall:{'%.3f' % np.mean(self.test_batch_recall)}"
)
print(
f"Loss: {'%.3f' % np.mean(self.test_batch_loss)} | Accuracy: {'%.3f' % np.mean(self.test_batch_accuracy)} | UAR: {'%.3f' % np.mean(self.test_batch_uar)}| F1:{'%.3f' % np.mean(self.test_batch_f1)} | Precision:{'%.3f' % np.mean(self.test_batch_precision)} | Recall:{'%.3f' % np.mean(self.test_batch_recall)}",
file=self.log_file)
print(
"****************************************************************")
print(np.array(overall_predictions).shape)
print(f"{type} predictions mean", np.mean(overall_predictions))
print(f"{type} predictions mean",
np.mean(overall_predictions),
file=self.log_file)
print(f"{type} predictions sum", np.sum(overall_predictions))
print(f"{type} predictions sum",
np.sum(overall_predictions),
file=self.log_file)
print(f"{type} predictions range", np.min(overall_predictions),
np.max(overall_predictions))
print(f"{type} predictions range",
np.min(overall_predictions),
np.max(overall_predictions),
file=self.log_file)
print(f"{type} predictions hist", np.histogram(overall_predictions))
print(f"{type} predictions hist",
np.histogram(overall_predictions),
file=self.log_file)
print(f"{type} predictions variance", np.var(overall_predictions))
print(f"{type} predictions variance",
np.var(overall_predictions),
file=self.log_file)
print(
"****************************************************************")
log_summary(self.writer,
epoch,
accuracy=np.mean(self.test_batch_accuracy),
loss=np.mean(self.test_batch_loss),
uar=np.mean(self.test_batch_uar),
lr=self.optimiser.state_dict()['param_groups'][0]['lr'],
type=type)
log_conf_matrix(self.writer,
epoch,
predictions_dict=predictions_dict,
type=type)
y = [element for sublist in y for element in sublist]
write_to_npy(filename=self.debug_filename,
predictions=overall_predictions,
labels=y,
epoch=epoch,
accuracy=np.mean(self.test_batch_accuracy),
loss=np.mean(self.test_batch_loss),
uar=np.mean(self.test_batch_uar),
lr=self.optimiser.state_dict()['param_groups'][0]['lr'],
predictions_dict=predictions_dict,
type=type)
def run_for_epoch(self, epoch, x, y, jitterx, type):
# self.network.eval()
# for m in self.network.modules():
# if isinstance(m, nn.BatchNorm2d):
# m.track_running_stats = False
predictions_dict = {"tp": [], "fp": [], "tn": [], "fn": []}
logits, predictions = [], []
self.test_batch_loss, self.test_batch_accuracy, self.test_batch_uar, self.test_batch_ua, self.test_batch_f1, self.test_batch_precision, self.test_batch_recall, audio_for_tensorboard_test = [], [], [], [], [], [], [], None
with torch.no_grad():
for i, (audio_data, label,
jitter_shimmer_data) in enumerate(zip(x, y, jitterx)):
label = to_tensor(label, device=self.device).float()
audio_data = to_tensor(audio_data, device=self.device)
jitter_shimmer_data = to_tensor(jitter_shimmer_data,
device=self.device)
test_predictions = self.network(audio_data,
jitter_shimmer_data).squeeze(1)
logits.extend(to_numpy(test_predictions))
test_loss = self.loss_function(test_predictions, label)
test_predictions = nn.Sigmoid()(test_predictions)
predictions.append(to_numpy(test_predictions))
test_accuracy, test_uar, test_precision, test_recall, test_f1 = accuracy_fn(
test_predictions, label, self.threshold)
self.test_batch_loss.append(to_numpy(test_loss))
self.test_batch_accuracy.append(to_numpy(test_accuracy))
self.test_batch_uar.append(test_uar)
self.test_batch_f1.append(test_f1)
self.test_batch_precision.append(test_precision)
self.test_batch_recall.append(test_recall)
tp, fp, tn, fn = custom_confusion_matrix(
test_predictions, label, threshold=self.threshold)
predictions_dict['tp'].extend(tp)
predictions_dict['fp'].extend(fp)
predictions_dict['tn'].extend(tn)
predictions_dict['fn'].extend(fn)
predictions = [
element for sublist in predictions for element in sublist
]
self.logger.info(f'***** {type} Metrics ***** ')
self.logger.info(
f"Loss: {'%.3f' % np.mean(self.test_batch_loss)} | Accuracy: {'%.3f' % np.mean(self.test_batch_accuracy)} | UAR: {'%.3f' % np.mean(self.test_batch_uar)}| F1:{'%.3f' % np.mean(self.test_batch_f1)} | Precision:{'%.3f' % np.mean(self.test_batch_precision)} | Recall:{'%.3f' % np.mean(self.test_batch_recall)}"
)
log_summary(self.writer,
epoch,
accuracy=np.mean(self.test_batch_accuracy),
loss=np.mean(self.test_batch_loss),
uar=np.mean(self.test_batch_uar),
lr=self.optimiser.state_dict()['param_groups'][0]['lr'],
type=type)
log_conf_matrix(self.writer,
epoch,
predictions_dict=predictions_dict,
type=type)
log_learnable_parameter(self.writer,
epoch,
to_tensor(logits, device=self.device),
name=f'{type}_logits')
log_learnable_parameter(self.writer,
epoch,
to_tensor(predictions, device=self.device),
name=f'{type}_predictions')
write_to_npy(filename=self.debug_filename,
predictions=predictions,
labels=y,
epoch=epoch,
accuracy=np.mean(self.test_batch_accuracy),
loss=np.mean(self.test_batch_loss),
uar=np.mean(self.test_batch_uar),
lr=self.optimiser.state_dict()['param_groups'][0]['lr'],
predictions_dict=predictions_dict,
type=type)