def test_evaluate_named_model(data_dir, mocker):
x = numpy.random.randint(256, size=(100, 48, 48, 3)).astype(numpy.float64)
y = numpy.random.randint(4, size=(100, ))
meanscsv = str(data_dir.join("zoolander_means.csv"))
with open(meanscsv, "w") as csvfile:
writer = csv.writer(csvfile)
writer.writerow([125.3, 127.12, 121.9])
expected_samples = x.copy()
expected_samples[:, :, :, 0] = (expected_samples[:, :, :, 0] - 125.3 +
255.0) / (2.0 * 255.0)
expected_samples[:, :, :, 1] = (expected_samples[:, :, :, 1] - 127.12 +
255.0) / (2.0 * 255.0)
expected_samples[:, :, :, 2] = (expected_samples[:, :, :, 2] - 121.9 +
255.0) / (2.0 * 255.0)
expected_targets = keras.utils.to_categorical(y, 4)
with mocker.patch("keras_resnet.models.ResNet50") as model_mock:
keras_resnet.models.ResNet50.return_value = model_mock
resources = mocker.patch("pkg_resources.resource_filename")
resources.side_effect = lambda _, filename: str(
data_dir.join(os.path.basename(filename)))
model = deepometry.model.Model(shape=(48, 48, 3),
units=4,
name="zoolander")
model.compile()
model.evaluate(x, y, batch_size=10, verbose=0)
model_mock.load_weights.assert_called_once_with(
pkg_resources.resource_filename(
"deepometry", os.path.join("data",
"zoolander_checkpoint.hdf5")))
model_mock.evaluate.assert_called_once_with(x=mocker.ANY,
y=mocker.ANY,
batch_size=10,
verbose=0)
_, kwargs = model_mock.evaluate.call_args
samples = kwargs["x"]
assert samples.shape == expected_samples.shape
numpy.testing.assert_array_equal(samples, expected_samples)
targets = kwargs["y"]
assert targets.shape == expected_targets.shape
numpy.testing.assert_array_equal(targets, expected_targets)
def test_evaluate_named_directory(data_dir, mocker):
x = numpy.random.randint(256, size=(100, 48, 48, 3)).astype(numpy.float64)
y = numpy.random.randint(4, size=(100, ))
model_directory = data_dir.mkdir("models")
meanscsv = str(model_directory.join("means.csv"))
with open(meanscsv, "w") as csvfile:
writer = csv.writer(csvfile)
writer.writerow([125.3, 127.12, 121.9])
expected_samples = x.copy()
expected_samples[:, :, :, 0] -= 125.3
expected_samples[:, :, :, 1] -= 127.12
expected_samples[:, :, :, 2] -= 121.9
expected_targets = keras.utils.to_categorical(y, 4)
with mocker.patch("keras_resnet.models.ResNet50") as model_mock:
keras_resnet.models.ResNet50.return_value = model_mock
model = deepometry.model.Model(shape=(48, 48, 3),
units=4,
directory=str(model_directory))
model.compile()
model.evaluate(x, y, batch_size=10, verbose=0)
model_mock.load_weights.assert_called_once_with(
os.path.join(str(model_directory), "checkpoint.hdf5"))
model_mock.evaluate.assert_called_once_with(x=mocker.ANY,
y=mocker.ANY,
batch_size=10,
verbose=0)
_, kwargs = model_mock.evaluate.call_args
samples = kwargs["x"]
assert samples.shape == expected_samples.shape
numpy.testing.assert_array_equal(samples, expected_samples)
targets = kwargs["y"]
assert targets.shape == expected_targets.shape
numpy.testing.assert_array_equal(targets, expected_targets)
def _evaluate(x, y, units, batch_size, directory, name, verbose):
import deepometry.model
model = deepometry.model.Model(directory=directory,
name=name,
shape=x.shape[1:],
units=units)
model.compile()
metrics = model.evaluate(x, y, batch_size=batch_size, verbose=verbose)
return model.model.metrics_names, metrics
