def testAccuracyKEquals4(self):
correct, total, accuracy = computeAccuracy(self.distances,
self.consumer_labels,
self.shop_labels,
k=4)
self.assertEqual((correct, total, accuracy), (3, 3, 1))
print("Loaded model from disk")
DATA_DIR = './img_npy_final_features_only/DRESSES/Skirt/'
consumer_features = np.load(DATA_DIR + 'consumer_ResNet50_features.npy')
consumer_labels = np.load(DATA_DIR + 'consumer_labels.npy')
shop_features = np.load(DATA_DIR + 'shop_ResNet50_features.npy')
shop_labels = np.load(DATA_DIR + 'shop_labels.npy')
print (consumer_features.shape)
print (consumer_labels.shape)
print (shop_features.shape)
print (shop_labels.shape)
metrics = [DistanceMetrics.L1] #, DistanceMetrics.L2
top_k = [3,10,20,30,40,50]
for metric in metrics:
print ("Metric: {}".format(metric))
accuracies = computeAccuracy(consumer_features,
shop_features,
consumer_labels,
shop_labels,
metric = metric,
model = model,
k = top_k)
print(accuracies)
preds = [1 if p > 0.5 else 0 for p in preds]
print("accuracy on batch:", accuracy_score(target, preds))
print(confusion_matrix(target, preds))
print(
precision_recall_fscore_support(target,
preds,
average='weighted'))
print("Finished batch {} of {}".format(batch_iter, num_batches))
batch_iter += 1
print("Printing train set accuracy")
computeAccuracy(consumer_features,
shop_features,
consumer_labels,
shop_labels,
metric=metric,
model=model,
k=[10, 20, 30])
print("Printing test set acuracy")
computeAccuracy(test_consumer_features,
shop_features,
test_consumer_labels,
shop_labels,
metric=metric,
model=model,
k=[10, 20, 30])
if (SAVE_MODEL):
model_json = model.to_json()