def test_to_categorical():
# Test without n_classes
one_hot = to_categorical(numpy.asarray([1, 2], dtype="i"))
assert one_hot.shape == (2, 3)
# From keras
# https://github.com/keras-team/keras/blob/master/tests/keras/utils/np_utils_test.py
nc = 5
shapes = [(1, ), (3, ), (4, 3), (5, 4, 3), (3, 1), (3, 2, 1)]
expected_shapes = [
(1, nc),
(3, nc),
(4, 3, nc),
(5, 4, 3, nc),
(3, 1, nc),
(3, 2, 1, nc),
]
labels = [numpy.random.randint(0, nc, shape) for shape in shapes]
one_hots = [to_categorical(label, nc) for label in labels]
for label, one_hot, expected_shape in zip(labels, one_hots,
expected_shapes):
assert one_hot.shape == expected_shape
assert numpy.array_equal(one_hot, one_hot.astype(bool))
assert numpy.all(one_hot.sum(axis=-1) == 1)
assert numpy.all(
numpy.argmax(one_hot, -1).reshape(label.shape) == label)
def Y(answer: int, n_classes: int) -> Array2d:
ops: Ops = get_current_ops()
return cast(
Array2d,
to_categorical(cast(IntsXd, ops.asarray([answer])),
n_classes=n_classes),
)
def test_tensorflow_wrapper_serialize_model_subclass(
X, Y, input_size, n_classes, answer
):
import tensorflow as tf
input_shape = (1, input_size)
ops = get_current_ops()
@keras_subclass(
"foo.v1",
X=ops.alloc2f(*input_shape),
Y=to_categorical(ops.asarray1i([1]), n_classes=n_classes),
input_shape=input_shape,
)
class CustomKerasModel(tf.keras.Model):
def __init__(self, **kwargs):
super(CustomKerasModel, self).__init__(**kwargs)
self.in_dense = tf.keras.layers.Dense(
12, name="in_dense", input_shape=input_shape
)
self.out_dense = tf.keras.layers.Dense(
n_classes, name="out_dense", activation="softmax"
)
def call(self, inputs) -> tf.Tensor:
x = self.in_dense(inputs)
return self.out_dense(x)
model = TensorFlowWrapper(CustomKerasModel())
# Train the model to predict the right single answer
optimizer = Adam()
for i in range(50):
guesses, backprop = model(X, is_train=True)
d_guesses = (guesses - Y) / guesses.shape[0]
backprop(d_guesses)
model.finish_update(optimizer)
predicted = model.predict(X).argmax()
assert predicted == answer
# Save then Load the model from bytes
model.from_bytes(model.to_bytes())
# The from_bytes model gets the same answer
assert model.predict(X).argmax() == answer
def Y(answer, n_classes):
ops = get_current_ops()
return to_categorical(ops.asarray1i([answer]), n_classes=n_classes)
def test_to_categorical(label_smoothing):
# Test without n_classes
one_hot = to_categorical(numpy.asarray([1, 2], dtype="i"))
assert one_hot.shape == (2, 3)
# From keras
# https://github.com/keras-team/keras/blob/master/tests/keras/utils/np_utils_test.py
nc = 5
shapes = [(1, ), (3, ), (4, 3), (5, 4, 3), (3, 1), (3, 2, 1)]
expected_shapes = [
(1, nc),
(3, nc),
(4, 3, nc),
(5, 4, 3, nc),
(3, 1, nc),
(3, 2, 1, nc),
]
labels = [numpy.random.randint(0, nc, shape) for shape in shapes]
one_hots = [to_categorical(label, nc) for label in labels]
smooths = [
to_categorical(label, nc, label_smoothing=label_smoothing)
for label in labels
]
for i in range(len(expected_shapes)):
label = labels[i]
one_hot = one_hots[i]
expected_shape = expected_shapes[i]
smooth = smooths[i]
assert one_hot.shape == expected_shape
assert smooth.shape == expected_shape
assert numpy.array_equal(one_hot, one_hot.astype(bool))
assert numpy.all(one_hot.sum(axis=-1) == 1)
assert numpy.all(
numpy.argmax(one_hot, -1).reshape(label.shape) == label)
assert numpy.all(smooth.argmax(axis=-1) == one_hot.argmax(axis=-1))
assert numpy.all(numpy.isclose(numpy.sum(smooth, axis=-1), 1.0))
assert numpy.isclose(numpy.max(smooth), 1 - label_smoothing)
assert numpy.isclose(numpy.min(smooth),
label_smoothing / (smooth.shape[-1] - 1))
# At least one class is required without label smoothing.
numpy.testing.assert_allclose(to_categorical(numpy.asarray([0, 0, 0]), 1),
[[1.0], [1.0], [1.0]])
numpy.testing.assert_allclose(to_categorical(numpy.asarray([0, 0, 0])),
[[1.0], [1.0], [1.0]])
with pytest.raises(ValueError, match=r"n_classes should be at least 1"):
to_categorical(numpy.asarray([0, 0, 0]), 0)
# At least two classes are required with label smoothing.
numpy.testing.assert_allclose(
to_categorical(numpy.asarray([0, 1, 0]), 2, label_smoothing=0.01),
[[0.99, 0.01], [0.01, 0.99], [0.99, 0.01]],
)
numpy.testing.assert_allclose(
to_categorical(numpy.asarray([0, 1, 0]), label_smoothing=0.01),
[[0.99, 0.01], [0.01, 0.99], [0.99, 0.01]],
)
with pytest.raises(
ValueError,
match=r"n_classes should be greater than 1.*label smoothing.*but 1"
):
to_categorical(numpy.asarray([0, 1, 0]), 1, label_smoothing=0.01),
with pytest.raises(
ValueError,
match=r"n_classes should be greater than 1.*label smoothing.*but 1"
):
to_categorical(numpy.asarray([0, 0, 0]), label_smoothing=0.01),
