def _testXent(self, np_features, np_labels, use_gpu=False):
np_loss, np_backprop = self._npXent(np_features, np_labels)
with self.test_session(use_gpu=use_gpu) as sess:
loss, backprop = gen_nn_ops.softmax_cross_entropy_with_logits(
np_features, np_labels)
tf_loss, tf_backprop = sess.run([loss, backprop])
self.assertAllCloseAccordingToType(np_loss, tf_loss)
self.assertAllCloseAccordingToType(np_backprop, tf_backprop)
def _testSingleClass(self, use_gpu=False):
for dtype in np.float16, np.float32:
with self.test_session(use_gpu=use_gpu) as sess:
loss, backprop = gen_nn_ops.softmax_cross_entropy_with_logits(
np.array([[1.], [-1.], [0.]]).astype(dtype),
np.array([[-1.], [0.], [1.]]).astype(dtype))
tf_loss, tf_backprop = sess.run([loss, backprop])
self.assertAllClose([0.0, 0.0, 0.0], tf_loss)
self.assertAllClose([[2.0], [1.0], [0.0]], tf_backprop)
def testShapeBroadcast(self):
np_f = np.array([[1., 2., 3., 4.], [1., 2., 3.,
4.]]).astype(np.float32)
np_l = np.array([[0., 0., 0., 1.], [0., .5, .5,
0.]]).astype(np.float32)
np_loss, np_backprop = self._npXent(np_f, np_l)
tf_f = constant_op.constant(
np.array([[1., 2., 3., 4.]]).astype(np.float32))
tf_l = constant_op.constant(
np.array([[0., 0., 0., 1.], [0., .5, .5, 0.]]).astype(np.float32))
for use_gpu in [False, True]:
with self.cached_session(use_gpu=use_gpu) as sess:
loss, backprop = gen_nn_ops.softmax_cross_entropy_with_logits(
tf_f, tf_l)
tf_loss, tf_backprop = self.evaluate([loss, backprop])
self.assertAllCloseAccordingToType(np_loss, tf_loss)
self.assertAllCloseAccordingToType(np_backprop, tf_backprop)
def testShapeBroadcast(self):
np_f = np.array([[1., 2., 3., 4.],
[1., 2., 3., 4.]]).astype(np.float32)
np_l = np.array([[0., 0., 0., 1.],
[0., .5, .5, 0.]]).astype(np.float32)
np_loss, np_backprop = self._npXent(np_f, np_l)
tf_f = constant_op.constant(
np.array([[1., 2., 3., 4.]]).astype(np.float32))
tf_l = constant_op.constant(
np.array([[0., 0., 0., 1.], [0., .5, .5, 0.]]).astype(np.float32))
for use_gpu in [False, True]:
with self.test_session(use_gpu=use_gpu) as sess:
loss, backprop = gen_nn_ops.softmax_cross_entropy_with_logits(
tf_f, tf_l)
tf_loss, tf_backprop = sess.run([loss, backprop])
self.assertAllCloseAccordingToType(np_loss, tf_loss)
self.assertAllCloseAccordingToType(np_backprop, tf_backprop)
def testExceptionThrowing(self):
with self.session(), test_util.force_gpu():
for dtype in [dtypes.float16, dtypes.float32, dtypes.float64]:
features = constant_op.constant([[0.3, 0.5], [0.5, 0.6]],
dtype=dtype)
labels = constant_op.constant([[0.2, 0.4], [0.1, 0.2]],
dtype=dtype)
with self.assertRaisesRegex(
errors_impl.UnimplementedError,
"The GPU implementation of SoftmaxCrossEntropyWithLogits that "
+
"would have been executed is not deterministic. Note that the "
+
"Python API uses an alternative, deterministic, GPU-accelerated "
+ "path when determinism is enabled."):
result = gen_nn_ops.softmax_cross_entropy_with_logits(
features=features, labels=labels)
self.evaluate(result)
def test_softmax_cross_entropy_with_logits_1d2d(self):
num_classes = 10
batch_size = 100
total_size = batch_size * num_classes
# labels should be a valid prob distribution across num_classes for each training sample
# Compute the softmax transformation along the second axis (i.e. along the rows summing over num_classes)
labels = tf.nn.softmax(np.random.rand(batch_size, num_classes), axis=1)
# features/Logits could be any real number
features = 200 * np.random.rand(1, num_classes)
out = softmax_cross_entropy_with_logits(features, labels)
sess_fn = lambda sess: sess.run(out)
expected = self.without_ngraph(sess_fn)
result = self.with_ngraph(sess_fn)
assert np.allclose(result[0], expected[0], rtol=0, atol=1e-02) # loss
assert np.allclose(result[1], expected[1], rtol=0,
atol=1e-02) # backprop
def testNotMatrix(self):
with self.cached_session():
with self.assertRaises(ValueError):
gen_nn_ops.softmax_cross_entropy_with_logits([0., 1., 2., 3.],
[0., 1., 0., 1.])
def testShapeMismatch(self):
with self.cached_session():
with self.assertRaises(ValueError):
gen_nn_ops.softmax_cross_entropy_with_logits(
[[0., 1.], [2., 3.]], [[0., 1., 0.], [1., 0., 0.]])
def testNotMatrix(self):
with self.test_session():
with self.assertRaises(ValueError):
gen_nn_ops.softmax_cross_entropy_with_logits([0., 1., 2., 3.],
[0., 1., 0., 1.])
def testShapeMismatch(self):
with self.test_session():
with self.assertRaises(ValueError):
gen_nn_ops.softmax_cross_entropy_with_logits(
[[0., 1.], [2., 3.]], [[0., 1., 0.], [1., 0., 0.]])
