def validate(crf, validate_data_true_label, validate_loader, val_dataset_size,
device):
all_predictions = []
for image_batch, data_ids, _ in validate_loader:
all_predictions.extend(crf.make_predictions(image_batch).cpu().numpy())
all_predictions = np.squeeze(np.array(all_predictions))
return mapk(actual=validate_data_true_label,
predicted=parse_validation_data_labels(all_predictions),
k=3)
def get_val_acc(self):
predictions = []
for i, val_data in self.cur_val_data_fold.iterrows():
val_data = build_feature.set_feature_mat(val_data['drawing'], 256)
prediction = np.exp(self.weights @ val_data)
prediction = np.argsort(prediction)
predictions.append(prediction[-3:])
return mapk(actual=np.matrix(
np.ones(self.cur_val_data_size, dtype=np.int8) * self.cur_cat),
predicted=np.array(predictions),
k=3)
def validate(resnet, validate_data_true_label, validate_loader,
val_dataset_size, device):
all_predictions = []
resnet.eval()
for image_batch, _ in validate_loader:
image_batch = image_batch.to(device)
outputs = resnet(image_batch)
_, predicted = torch.topk(outputs.data, 3)
all_predictions.extend(predicted.cpu().numpy())
all_predictions = np.squeeze(np.array(all_predictions))
return mapk(actual=validate_data_true_label,
predicted=parse_validation_data_labels(all_predictions),
k=3)
def get_val_acc(self, val_sess, predictions_i, feat_i_ph, weights_ph,
weights):
predictions = []
for data_id in self.cur_val_fold_seq:
val_data = build_feature.set_feature_mat(
self.cur_val_data_fold.loc[data_id, 'drawing'])
predictions.append(
val_sess.run(predictions_i,
feed_dict={
feat_i_ph: val_data,
weights_ph: weights
}))
return mapk(actual=np.matrix(
np.ones((self.num_val_data), dtype=np.int8) * self.cur_cat),
predicted=np.array(predictions),
k=3)