def predict(args):
"""Loads model from file, makes predictions and computes metrics.
All created files are saved to args.out_dir directory if provided,
or to results_<task> otherwise.
Creates files:
conf_matrix.png file with confusion matrix,
report.txt with various metrics,
preds_{task}.npy with raw predictions.
"""
if args.task.startswith('4'):
test_features = np.load('test_features_3b.npy')
test_labels = get_labels(split='test')
y_true = np.array([CLASSES.index(l) for l in test_labels])
out_dir = args.out_dir or f'results_{args.task}'
for c in [0.001, 0.01, 0.1, 1.0, 10]:
svc = SVC()
svc.load_from_file(f'svc_{args.task}_C_{c}')
y_pred = svc.predict(test_features)
evaluate(y_true, y_pred, None, CLASSES,
os.path.join(out_dir, f'C_{c}'))
else:
model: Model = load_model(
f'model_fc_{args.task}.h5',
custom_objects={'top_5_accuracy': top_5_accuracy})
test_generator = create_data_generator(split='test',
target_size=args.target_size,
batch_size=args.batch_size,
shuffle=False)
# get predictions
preds = model.predict_generator(test_generator, verbose=1)
# create output directory
out_dir = args.out_dir or f'results_{args.task}'
os.makedirs(out_dir, exist_ok=True)
# save numpy array with predictions
save_file = os.path.join(out_dir, f'preds_{args.task}.npy')
np.save(save_file, preds)
print(f'Predictions saved to: {save_file}')
# first, prepare y_pred, y_true and class names
# y_pred are classes predicted with the highest probability
y_pred = np.array([np.argmax(x) for x in preds])
# since we did not shuffle data in data generator,
# classes attribute of the generator contains true labels for each sample
y_true = np.array(test_generator.classes)
# class_names = list(test_generator.class_indices.keys())
# class_names.sort(key=lambda x: test_generator.class_indices[x])
evaluate(y_true, y_pred, preds, CLASSES, out_dir)
def predict(y, z):
clf = SVC(kernel="linear")
clf.load_from_file("./model")
clf.predict([[y, z]])[0]
