def predict(self, Ximg, yimg):
if self.verbose:
print('[AttentionTrend] Predicting')
model = keras.attention_models.load_model(self.output_directory + 'best_model.h5')
model_metrics, conf_mat, y_true, y_pred = predict_model_deep_learning(model, Ximg, yimg, self.output_directory)
save_logs(self.output_directory, self.hist, y_pred, y_true, self.duration)
keras.backend.clear_session()
if self.verbose:
print('[AttentionTrend] Prediction done!')
return model_metrics, conf_mat
def predict(self, Ximg, yimg):
if self.verbose:
print('[' + self.classifier_name + '] Predicting')
model = keras.models.load_model(
self.output_directory + 'best_model.h5',
custom_objects={
'MultiHeadAttention': MultiHeadAttention,
'SeqSelfAttention': SeqSelfAttention,
# 'PreProcessingLayer': PreProcessingLayer
})
model_metrics, conf_mat, y_true, y_pred = predict_model_deep_learning(
model, Ximg, yimg, self.output_directory)
save_logs(self.output_directory, self.hist, y_pred, y_true,
self.duration)
keras.backend.clear_session()
if self.verbose:
print('[' + self.classifier_name + '] Prediction done!')
return model_metrics, conf_mat
copy.deepcopy(server_part),
epoch)
client_part.load_state_dict(extractor_w)
server_part.load_state_dict(classifer_w)
if args.cifar100:
test_acc1, test_acc5, test_loss = test_inference4split4cifar100(
args, client_part, server_part, test_dataset)
print("|---- Test Accuracy1: {:.2f}%, Test Accuracy5: {:.2f}%".
format(100 * test_acc1, 100 * test_acc5))
log_obj = {
'test_acc1': "{:.2f}%".format(100 * test_acc1),
'test_acc5': "{:.2f}%".format(100 * test_acc5),
'loss': test_loss,
'epoch': epoch
}
else:
test_acc, test_loss = test_inference4split(args, client_part,
server_part,
test_dataset)
print("|---- Test Accuracy: {:.2f}%".format(100 * test_acc))
log_obj = {
'test_acc': "{:.2f}%".format(100 * test_acc),
'loss': test_loss,
'epoch': epoch
}
logs.append(log_obj)
if args.cifar100:
save_cifar100_logs(logs, TAG, args)
else:
save_logs(logs, TAG, args)