def predict(model_file):
import keras.models
_, languages = common.build_label_binarizer()
model = keras.models.load_model(model_file)
results = model.predict(samples)
scores = np.zeros(len(languages))
for result in results:
scores[np.argmax(result)] += 1
return scores, languages
parser.add_argument(
'--test',
dest='test',
action='store_true',
help='test the previously trained model against the test set')
parser.set_defaults(test=False)
args = parser.parse_args()
input_shape = (FB_HEIGHT, WIDTH, COLOR_DEPTH)
if args.test:
model = load_model(modelFileName)
input_shape = (FB_HEIGHT, WIDTH, COLOR_DEPTH)
label_binarizer, clazzes = common.build_label_binarizer()
test_labels, test_features, test_metadata = common.load_data(
label_binarizer, foldsFolder, 'test', [1], input_shape)
common.test(test_labels, test_features, test_metadata, model, clazzes)
else:
accuracies = []
numFolds = len(
glob(os.path.join(foldsFolder, "train_metadata.fold*.npy")))
generator = common.train_generator(numFolds,
foldsFolder,
input_shape,
max_iterations=1)
first = True