diff options
Diffstat (limited to 'tensorflow/python/kernel_tests/losses_test.py')
-rw-r--r-- | tensorflow/python/kernel_tests/losses_test.py | 216 |
1 files changed, 108 insertions, 108 deletions
diff --git a/tensorflow/python/kernel_tests/losses_test.py b/tensorflow/python/kernel_tests/losses_test.py index 87fc715783..3ce0b74263 100644 --- a/tensorflow/python/kernel_tests/losses_test.py +++ b/tensorflow/python/kernel_tests/losses_test.py @@ -61,62 +61,62 @@ class AbsoluteDifferenceLossTest(test.TestCase): self._labels = constant_op.constant([1, 9, 2, -5, -2, 6], shape=(2, 3)) def testValueErrorThrownWhenWeightIsNone(self): - with self.test_session(): + with self.cached_session(): with self.assertRaises(ValueError): losses.absolute_difference( self._predictions, self._predictions, weights=None) def testAllCorrectNoLossWeight(self): loss = losses.absolute_difference(self._predictions, self._predictions) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(0.0, loss.eval(), 3) def testNonZeroLoss(self): loss = losses.absolute_difference(self._labels, self._predictions) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(5.5, loss.eval(), 3) def testNonZeroLossWithPythonScalarWeight(self): weights = 2.3 loss = losses.absolute_difference(self._labels, self._predictions, weights) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(5.5 * weights, loss.eval(), 3) def testNonZeroLossWithScalarTensorWeight(self): weights = 2.3 loss = losses.absolute_difference(self._labels, self._predictions, constant_op.constant(weights)) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(5.5 * weights, loss.eval(), 3) def testNonZeroLossWithOneDimBatchSpecificWeights(self): weights = constant_op.constant((1.2, 0.0), shape=(2, 1)) loss = losses.absolute_difference(self._labels, self._predictions, weights) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(5.6, loss.eval(), 3) def testNonZeroLossWithTwoDimBatchSpecificWeights(self): weights = constant_op.constant([1.2, 0.0], shape=[2, 1]) loss = losses.absolute_difference(self._labels, self._predictions, weights) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(5.6, loss.eval(), 3) def testNonZeroLossWithSampleSpecificWeights(self): weights = constant_op.constant([3, 6, 5, 0, 4, 2], shape=[2, 3]) loss = losses.absolute_difference(self._labels, self._predictions, weights) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(16.6, loss.eval(), 3) def testNonZeroLossWithSampleSpecificWeightsMostZero(self): weights = constant_op.constant([0, 0, 0, 0, 0, 2], shape=[2, 3]) loss = losses.absolute_difference(self._labels, self._predictions, weights) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(6.0, loss.eval(), 3) def testLossWithSampleSpecificWeightsAllZero(self): weights = array_ops.zeros((2, 3)) loss = losses.absolute_difference(self._labels, self._predictions, weights) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(0.0, loss.eval(), 3) @test_util.assert_no_new_pyobjects_executing_eagerly @@ -134,12 +134,12 @@ class SoftmaxCrossEntropyLossTest(test.TestCase): logits = constant_op.constant([[10.0, 0.0, 0.0], [0.0, 10.0, 0.0], [0.0, 0.0, 10.0]]) labels = constant_op.constant([[1, 0, 0], [0, 1, 0], [0, 0, 1]]) - with self.test_session(): + with self.cached_session(): with self.assertRaises(ValueError): losses.softmax_cross_entropy(labels, logits, weights=None) def testAllCorrect(self): - with self.test_session(): + with self.cached_session(): logits = constant_op.constant([[10.0, 0.0, 0.0], [0.0, 10.0, 0.0], [0.0, 0.0, 10.0]]) labels = constant_op.constant([[1, 0, 0], [0, 1, 0], [0, 0, 1]]) @@ -152,7 +152,7 @@ class SoftmaxCrossEntropyLossTest(test.TestCase): [0.0, 0.0, 10.0]]) labels = constant_op.constant([[0, 0, 1], [1, 0, 0], [0, 1, 0]]) - with self.test_session(): + with self.cached_session(): loss = losses.softmax_cross_entropy(labels, logits) self.assertEquals(loss.op.name, 'softmax_cross_entropy_loss/value') self.assertAlmostEqual(loss.eval(), 10.0, 3) @@ -162,7 +162,7 @@ class SoftmaxCrossEntropyLossTest(test.TestCase): [0.0, 0.0, 10.0]]) labels = constant_op.constant([[0, 0, 1], [1, 0, 0], [0, 1, 0]]) weights = 2.3 - with self.test_session(): + with self.cached_session(): loss = losses.softmax_cross_entropy(labels, logits, weights) self.assertAlmostEqual(weights * 10.0, loss.eval(), 3) @@ -171,7 +171,7 @@ class SoftmaxCrossEntropyLossTest(test.TestCase): [0.0, 0.0, 10.0]]) labels = constant_op.constant([[0, 0, 1], [1, 0, 0], [0, 1, 0]]) weights = 2.3 - with self.test_session(): + with self.cached_session(): loss = losses.softmax_cross_entropy(labels, logits, constant_op.constant(weights)) self.assertAlmostEqual(weights * 10.0, loss.eval(), 3) @@ -181,7 +181,7 @@ class SoftmaxCrossEntropyLossTest(test.TestCase): [0.0, 0.0, 10.0]]) labels = constant_op.constant([[0, 0, 1], [1, 0, 0], [0, 1, 0]]) weights = constant_op.constant((1.2, 3.4, 5.6)) - with self.test_session(): + with self.cached_session(): loss = losses.softmax_cross_entropy(labels, logits, weights) self.assertAlmostEqual((1.2 + 3.4 + 5.6) * 10.0 / 3.0, loss.eval(), 3) @@ -190,7 +190,7 @@ class SoftmaxCrossEntropyLossTest(test.TestCase): [0.0, 0.0, 10.0]]) labels = constant_op.constant([[0, 0, 1], [1, 0, 0], [0, 1, 0]]) weights = constant_op.constant([0, 0, 0], shape=[3]) - with self.test_session(): + with self.cached_session(): loss = losses.softmax_cross_entropy(labels, logits, weights) self.assertAlmostEqual(0.0, loss.eval(), 3) @@ -199,12 +199,12 @@ class SoftmaxCrossEntropyLossTest(test.TestCase): [0.0, 0.0, 10.0]]) labels = constant_op.constant([[0, 0, 1], [1, 0, 0], [0, 1, 0]]) weights = constant_op.constant([1.2, 0, 0], shape=[3]) - with self.test_session(): + with self.cached_session(): loss = losses.softmax_cross_entropy(labels, logits, weights) self.assertAlmostEqual(12.0, loss.eval(), 3) def testSoftmaxWithMeasurementSpecificWeightsRaisesException(self): - with self.test_session(): + with self.cached_session(): logits = constant_op.constant([[100.0, -100.0, -100.0], [-100.0, 100.0, -100.0], [-100.0, -100.0, 100.0]]) @@ -215,7 +215,7 @@ class SoftmaxCrossEntropyLossTest(test.TestCase): losses.softmax_cross_entropy(labels, logits, weights=weights).eval() def testSoftmaxLabelSmoothing(self): - with self.test_session(): + with self.cached_session(): # Softmax Cross Entropy Loss is: # -\sum_i p_i \log q_i # where for a softmax activation @@ -242,12 +242,12 @@ class SparseSoftmaxCrossEntropyLossTest(test.TestCase): logits = constant_op.constant([[10.0, 0.0, 0.0], [0.0, 10.0, 0.0], [0.0, 0.0, 10.0]]) labels = constant_op.constant([[0], [1], [2]]) - with self.test_session(): + with self.cached_session(): with self.assertRaises(ValueError): losses.sparse_softmax_cross_entropy(labels, logits, weights=None) def testAllCorrectInt32Labels(self): - with self.test_session(): + with self.cached_session(): logits = constant_op.constant([[10.0, 0.0, 0.0], [0.0, 10.0, 0.0], [0.0, 0.0, 10.0]]) labels = constant_op.constant([[0], [1], [2]], dtype=dtypes.int32) @@ -263,7 +263,7 @@ class SparseSoftmaxCrossEntropyLossTest(test.TestCase): losses.sparse_softmax_cross_entropy(labels, logits) def testAllCorrectInt64Labels(self): - with self.test_session(): + with self.cached_session(): logits = constant_op.constant([[10.0, 0.0, 0.0], [0.0, 10.0, 0.0], [0.0, 0.0, 10.0]]) labels = constant_op.constant([[0], [1], [2]], dtype=dtypes.int64) @@ -272,7 +272,7 @@ class SparseSoftmaxCrossEntropyLossTest(test.TestCase): self.assertAlmostEqual(loss.eval(), 0.0, 3) def testAllCorrectNonColumnLabels(self): - with self.test_session(): + with self.cached_session(): logits = constant_op.constant([[10.0, 0.0, 0.0], [0.0, 10.0, 0.0], [0.0, 0.0, 10.0]]) labels = constant_op.constant([0, 1, 2]) @@ -285,7 +285,7 @@ class SparseSoftmaxCrossEntropyLossTest(test.TestCase): [0.0, 0.0, 10.0]]) labels = constant_op.constant([[2], [0], [1]], dtype=dtypes.int32) - with self.test_session(): + with self.cached_session(): loss = losses.sparse_softmax_cross_entropy(labels, logits) self.assertEquals(loss.op.name, 'sparse_softmax_cross_entropy_loss/value') self.assertAlmostEqual(loss.eval(), 10.0, 3) @@ -295,7 +295,7 @@ class SparseSoftmaxCrossEntropyLossTest(test.TestCase): [0.0, 0.0, 10.0]]) labels = constant_op.constant([[2], [0], [1]], dtype=dtypes.int64) - with self.test_session(): + with self.cached_session(): loss = losses.sparse_softmax_cross_entropy(labels, logits) self.assertEquals(loss.op.name, 'sparse_softmax_cross_entropy_loss/value') self.assertAlmostEqual(loss.eval(), 10.0, 3) @@ -305,7 +305,7 @@ class SparseSoftmaxCrossEntropyLossTest(test.TestCase): [0.0, 0.0, 10.0]]) labels = constant_op.constant([2, 0, 1]) - with self.test_session(): + with self.cached_session(): loss = losses.sparse_softmax_cross_entropy(labels, logits) self.assertEquals(loss.op.name, 'sparse_softmax_cross_entropy_loss/value') self.assertAlmostEqual(loss.eval(), 10.0, 3) @@ -315,7 +315,7 @@ class SparseSoftmaxCrossEntropyLossTest(test.TestCase): [0.0, 0.0, 10.0]]) labels = constant_op.constant([[2], [0], [1]]) weights = 2.3 - with self.test_session(): + with self.cached_session(): loss = losses.sparse_softmax_cross_entropy(labels, logits, weights) self.assertAlmostEqual(weights * 10.0, loss.eval(), 3) @@ -324,7 +324,7 @@ class SparseSoftmaxCrossEntropyLossTest(test.TestCase): [0.0, 0.0, 10.0]]) labels = constant_op.constant([[2], [0], [1]]) weights = 2.3 - with self.test_session(): + with self.cached_session(): loss = losses.sparse_softmax_cross_entropy(labels, logits, constant_op.constant(weights)) self.assertAlmostEqual(weights * 10.0, loss.eval(), 3) @@ -334,7 +334,7 @@ class SparseSoftmaxCrossEntropyLossTest(test.TestCase): [0.0, 0.0, 10.0]]) labels = constant_op.constant([[2], [0], [1]]) weights = 2.3 - with self.test_session(): + with self.cached_session(): loss = losses.sparse_softmax_cross_entropy( labels, logits, constant_op.constant((weights,))) self.assertAlmostEqual(weights * 10.0, loss.eval(), 3) @@ -345,7 +345,7 @@ class SparseSoftmaxCrossEntropyLossTest(test.TestCase): [0.0, 0.0, 10.0]]) labels = constant_op.constant([[2], [0], [1]]) weights = array_ops.placeholder(dtypes.float32) - with self.test_session() as sess: + with self.cached_session() as sess: loss = losses.sparse_softmax_cross_entropy(labels, logits, weights) loss_val = sess.run(loss, feed_dict={weights: ((1.2,), (3.4,), (5.6,))}) @@ -355,7 +355,7 @@ class SparseSoftmaxCrossEntropyLossTest(test.TestCase): logits = array_ops.placeholder(dtypes.float32) labels = array_ops.placeholder(dtypes.int32) weights = 1.0 - with self.test_session() as sess: + with self.cached_session() as sess: loss = losses.sparse_softmax_cross_entropy(labels, logits, weights) loss_val = sess.run(loss, feed_dict={ @@ -370,7 +370,7 @@ class SparseSoftmaxCrossEntropyLossTest(test.TestCase): logits = array_ops.placeholder(dtypes.float32, shape=(None, 3)) labels = array_ops.placeholder(dtypes.int32, shape=(None, 1)) weights = array_ops.placeholder(dtypes.float32) - with self.test_session() as sess: + with self.cached_session() as sess: loss = losses.sparse_softmax_cross_entropy(labels, logits, weights) loss_val = sess.run(loss, feed_dict={ @@ -387,7 +387,7 @@ class SparseSoftmaxCrossEntropyLossTest(test.TestCase): [0.0, 0.0, 10.0]]) labels = constant_op.constant([[2], [0], [1]]) weights = constant_op.constant([1.2, 3.4, 5.6], shape=(3, 1)) - with self.test_session(): + with self.cached_session(): loss = losses.sparse_softmax_cross_entropy(labels, logits, weights) self.assertAlmostEqual((1.2 + 3.4 + 5.6) * 10.0 / 3.0, loss.eval(), 3) @@ -396,7 +396,7 @@ class SparseSoftmaxCrossEntropyLossTest(test.TestCase): [0.0, 0.0, 10.0]]) labels = constant_op.constant([[2], [0], [1]]) weights = constant_op.constant([[1.2], [3.4], [5.6]]) - with self.test_session(): + with self.cached_session(): loss = losses.sparse_softmax_cross_entropy(labels, logits, weights) self.assertAlmostEqual((1.2 + 3.4 + 5.6) * 10.0 / 3.0, loss.eval(), 3) @@ -405,7 +405,7 @@ class SparseSoftmaxCrossEntropyLossTest(test.TestCase): [0.0, 0.0, 10.0]]) labels = constant_op.constant([[2], [0], [1]]) weights = constant_op.constant([0, 0, 0], shape=(3, 1)) - with self.test_session(): + with self.cached_session(): loss = losses.sparse_softmax_cross_entropy(labels, logits, weights) self.assertAlmostEqual(0.0, loss.eval(), 3) @@ -414,12 +414,12 @@ class SparseSoftmaxCrossEntropyLossTest(test.TestCase): [0.0, 0.0, 10.0]]) labels = constant_op.constant([[2], [0], [1]]) weights = constant_op.constant([1.2, 0, 0], shape=(3, 1)) - with self.test_session(): + with self.cached_session(): loss = losses.sparse_softmax_cross_entropy(labels, logits, weights) self.assertAlmostEqual(12.0, loss.eval(), 3) def testMeasurementSpecificWeightsRaisesException(self): - with self.test_session(): + with self.cached_session(): logits = constant_op.constant([[100.0, -100.0, -100.0], [-100.0, 100.0, -100.0], [-100.0, -100.0, 100.0]]) @@ -432,7 +432,7 @@ class SparseSoftmaxCrossEntropyLossTest(test.TestCase): def testInconsistentWeightSizeRaisesException(self): """The weight tensor has incorrect number of elements.""" - with self.test_session(): + with self.cached_session(): logits = constant_op.constant([[100.0, -100.0, -100.0], [-100.0, 100.0, -100.0], [-100.0, -100.0, 100.0]]) @@ -445,7 +445,7 @@ class SparseSoftmaxCrossEntropyLossTest(test.TestCase): def testInconsistentLabelSizeRaisesException(self): """The label tensor has incorrect number of elements.""" - with self.test_session(): + with self.cached_session(): logits = constant_op.constant([[100.0, -100.0, -100.0], [-100.0, 100.0, -100.0], [-100.0, -100.0, 100.0]]) @@ -458,7 +458,7 @@ class SparseSoftmaxCrossEntropyLossTest(test.TestCase): def testInconsistentWeightShapeRaisesException(self): """The weight tensor has incorrect shape.""" - with self.test_session(): + with self.cached_session(): logits = constant_op.constant([[100.0, -100.0, -100.0, -100.0], [-100.0, 100.0, -100.0, -100.0], [-100.0, -100.0, 100.0, -100.0], @@ -472,7 +472,7 @@ class SparseSoftmaxCrossEntropyLossTest(test.TestCase): def testInconsistentLabelShapeRaisesException(self): """The label tensor has incorrect shape.""" - with self.test_session(): + with self.cached_session(): logits = constant_op.constant([[100.0, -100.0, -100.0, -100.0], [-100.0, 100.0, -100.0, -100.0], [-100.0, -100.0, 100.0, -100.0], @@ -488,7 +488,7 @@ class SparseSoftmaxCrossEntropyLossTest(test.TestCase): class SigmoidCrossEntropyLossTest(test.TestCase): def testAllCorrectSigmoid(self): - with self.test_session(): + with self.cached_session(): logits = constant_op.constant([[100.0, -100.0, -100.0], [-100.0, 100.0, -100.0], [-100.0, -100.0, 100.0]]) @@ -506,7 +506,7 @@ class SigmoidCrossEntropyLossTest(test.TestCase): loss = losses.sigmoid_cross_entropy(labels, logits, weights) self.assertEquals(logits.dtype, loss.dtype) - with self.test_session() as sess: + with self.cached_session() as sess: loss = sess.run(loss, feed_dict={ logits: np.ones((32, 1)), @@ -522,7 +522,7 @@ class SigmoidCrossEntropyLossTest(test.TestCase): loss = losses.sigmoid_cross_entropy(labels, logits, weights) self.assertEquals(logits.dtype, loss.dtype) - with self.test_session() as sess: + with self.cached_session() as sess: loss = sess.run(loss, feed_dict={ logits: np.ones((32, 2)), @@ -531,7 +531,7 @@ class SigmoidCrossEntropyLossTest(test.TestCase): self.assertAlmostEqual(0.313, loss, 3) def testAllWrongSigmoid(self): - with self.test_session(): + with self.cached_session(): logits = constant_op.constant([[100.0, -100.0, -100.0], [-100.0, 100.0, -100.0], [-100.0, -100.0, 100.0]]) @@ -542,7 +542,7 @@ class SigmoidCrossEntropyLossTest(test.TestCase): self.assertAlmostEqual(loss.eval(), 600.0 / 9.0, 3) def testAllWrongSigmoidWithMeasurementSpecificWeights(self): - with self.test_session(): + with self.cached_session(): logits = constant_op.constant([[100.0, -100.0, -100.0], [-100.0, 100.0, -100.0], [-100.0, -100.0, 100.0]]) @@ -562,7 +562,7 @@ class SigmoidCrossEntropyLossTest(test.TestCase): self.assertEquals(logits.dtype, loss.dtype) self.assertEquals('sigmoid_cross_entropy_loss/value', loss.op.name) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(0.0, loss.eval(), 3) def testSigmoidFloat64(self): @@ -577,7 +577,7 @@ class SigmoidCrossEntropyLossTest(test.TestCase): loss = losses.sigmoid_cross_entropy(labels, logits) self.assertEquals(logits.dtype, loss.dtype) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(44.444, loss.eval(), 3) def testSigmoidNoReduction(self): @@ -590,7 +590,7 @@ class SigmoidCrossEntropyLossTest(test.TestCase): labels, logits, reduction=losses.Reduction.NONE) self.assertEquals(logits.dtype, loss.dtype) - with self.test_session(): + with self.cached_session(): self.assertAllClose(( (0., 0., 0.), (0., 100., 100.), @@ -598,7 +598,7 @@ class SigmoidCrossEntropyLossTest(test.TestCase): ), loss.eval(), 3) def testSigmoidLabelSmoothingCorrect(self): - with self.test_session(): + with self.cached_session(): logits = constant_op.constant([[100.0, -100.0, -100.0]]) labels = constant_op.constant([[1, 0, 1]]) # Sigmoid cross entropy loss is: @@ -621,7 +621,7 @@ class SigmoidCrossEntropyLossTest(test.TestCase): self.assertAlmostEqual(loss.eval(), expected_value, 3) def testSigmoidLabelSmoothingEqualsSoftmaxTwoLabel(self): - with self.test_session(): + with self.cached_session(): label_smoothing = 0.1 sigmoid_logits = constant_op.constant([[100.0, -100.0, -100.0]]) sigmoid_labels = constant_op.constant([[1, 0, 1]]) @@ -656,33 +656,33 @@ class LogLossTest(test.TestCase): self._labels = constant_op.constant(labels) def testValueErrorThrownWhenWeightIsNone(self): - with self.test_session(): + with self.cached_session(): with self.assertRaises(ValueError): losses.log_loss(self._labels, self._labels, weights=None) def testAllCorrectNoLossWeight(self): loss = losses.log_loss(self._labels, self._labels) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(0.0, loss.eval(), 3) def testAllCorrectNoLossWeightWithPlaceholder(self): tf_predictions = array_ops.placeholder( dtypes.float32, shape=self._np_labels.shape) loss = losses.log_loss(self._labels, tf_predictions) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual( 0.0, loss.eval(feed_dict={tf_predictions: self._np_labels}), 3) def testNonZeroLoss(self): loss = losses.log_loss(self._labels, self._predictions) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(-np.sum(self._expected_losses) / 6.0, loss.eval(), 3) def testNonZeroLossWithPythonScalarWeight(self): weights = 2.3 loss = losses.log_loss(self._labels, self._predictions, weights) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(weights * -np.sum(self._expected_losses) / 6.0, loss.eval(), 3) @@ -690,7 +690,7 @@ class LogLossTest(test.TestCase): weights = 2.3 loss = losses.log_loss(self._labels, self._predictions, constant_op.constant(weights)) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(weights * -np.sum(self._expected_losses) / 6.0, loss.eval(), 3) @@ -700,7 +700,7 @@ class LogLossTest(test.TestCase): weights = 2.3 loss = losses.log_loss(self._labels, tf_predictions, constant_op.constant(weights)) - with self.test_session() as sess: + with self.cached_session() as sess: loss = sess.run(loss, feed_dict={tf_predictions: self._np_predictions}) self.assertAlmostEqual(weights * -np.sum(self._expected_losses) / 6.0, loss, 3) @@ -710,7 +710,7 @@ class LogLossTest(test.TestCase): weights = 2.3 loss = losses.log_loss(self._labels, tf_predictions, constant_op.constant(weights)) - with self.test_session() as sess: + with self.cached_session() as sess: loss = sess.run(loss, feed_dict={tf_predictions: self._np_predictions}) self.assertAlmostEqual(weights * -np.sum(self._expected_losses) / 6.0, loss, 3) @@ -721,7 +721,7 @@ class LogLossTest(test.TestCase): self._expected_losses, np.asarray([1.2, 1.2, 1.2, 3.4, 3.4, 3.4]).reshape((2, 3))) loss = losses.log_loss(self._labels, self._predictions, weights) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(-np.sum(expected_losses) / 6.0, loss.eval(), 3) def testNonZeroLossWithOneDimBatchSpecificWeightsSomeZero(self): @@ -730,7 +730,7 @@ class LogLossTest(test.TestCase): np.asarray([1.2, 1.2, 1.2, 0, 0, 0]).reshape( (2, 3))) loss = losses.log_loss(self._labels, self._predictions, weights) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(-np.sum(expected_losses) / 3.0, loss.eval(), 3) def testNonZeroLossWithTwoDimBatchSpecificWeightsSomeZero(self): @@ -739,12 +739,12 @@ class LogLossTest(test.TestCase): np.asarray([1.2, 1.2, 1.2, 0, 0, 0]).reshape( (2, 3))) loss = losses.log_loss(self._labels, self._predictions, weights) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(-np.sum(expected_losses) / 3.0, loss.eval(), 3) def testWeightsWithSameNumDimsButWrongShapeThrowsException(self): weights = constant_op.constant(np.random.normal(size=(2, 4)), shape=[2, 4]) - with self.test_session(): + with self.cached_session(): with self.assertRaises(ValueError): losses.log_loss(self._labels, self._predictions, weights) @@ -757,7 +757,7 @@ class LogLossTest(test.TestCase): self._predictions, constant_op.constant( weights, shape=(2, 3))) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(-np.sum(expected_losses) / 5.0, loss.eval(), 3) def testNonZeroLossWithMeasurementSpecificWeightsWithPlaceholder(self): @@ -771,7 +771,7 @@ class LogLossTest(test.TestCase): constant_op.constant( weights, shape=(2, 3))) - with self.test_session() as sess: + with self.cached_session() as sess: loss = sess.run(loss, feed_dict={tf_predictions: self._np_predictions}) self.assertAlmostEqual(-np.sum(expected_losses) / 5.0, loss, 3) @@ -784,7 +784,7 @@ class LogLossTest(test.TestCase): self._predictions, constant_op.constant( weights, shape=(2, 3))) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(-np.sum(expected_losses), loss.eval(), 3) def testNonZeroLossWithSampleSpecificWeightsMostZeroWithPlaceholder(self): @@ -795,35 +795,35 @@ class LogLossTest(test.TestCase): tf_weights = constant_op.constant(weights, shape=(2, 3)) loss = losses.log_loss(self._labels, tf_predictions, tf_weights) - with self.test_session() as sess: + with self.cached_session() as sess: loss = sess.run(loss, feed_dict={tf_predictions: self._np_predictions}) self.assertAlmostEqual(-np.sum(expected_losses), loss, 3) def testLossWithSampleSpecificWeightsAllZero(self): tf_weights = array_ops.zeros(shape=(2, 3)) loss = losses.log_loss(self._labels, self._predictions, tf_weights) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(0.0, loss.eval(), 3) class HingeLossTest(test.TestCase): def testIncompatibleShapes(self): - with self.test_session(): + with self.cached_session(): logits = constant_op.constant([[-1.0], [2.1]]) labels = constant_op.constant([0.0, 1.0]) with self.assertRaises(ValueError): _ = losses.hinge_loss(labels, logits).eval() def testAllOutsideMargin(self): - with self.test_session(): + with self.cached_session(): logits = constant_op.constant([1.2, -1.4, -1.0, 2.1]) labels = constant_op.constant([1.0, 0.0, 0.0, 1.0]) loss = losses.hinge_loss(labels, logits) self.assertAllClose(loss.eval(), 0.0, atol=1e-3) def testSomeInsideMargin(self): - with self.test_session(): + with self.cached_session(): logits = constant_op.constant([[-0.7], [-1.4], [1.4], [0.6]]) labels = constant_op.constant([[0.0], [0.0], [1.0], [1.0]]) loss = losses.hinge_loss(labels, logits) @@ -832,7 +832,7 @@ class HingeLossTest(test.TestCase): self.assertAllClose(loss.eval(), 0.175, atol=1e-3) def testSomeMisclassified(self): - with self.test_session(): + with self.cached_session(): logits = constant_op.constant([[[1.2], [0.4], [-1.0], [-1.1]]]) labels = constant_op.constant([[[1.0], [0.0], [0.0], [1.0]]]) loss = losses.hinge_loss(labels, logits) @@ -844,14 +844,14 @@ class HingeLossTest(test.TestCase): class HuberLossTest(test.TestCase): def testIncompatibleShapes(self): - with self.test_session(): + with self.cached_session(): predictions = constant_op.constant([[-1.0], [2.1]]) labels = constant_op.constant([0.0, 1.0]) with self.assertRaises(ValueError): _ = losses.huber_loss(labels, predictions).eval() def testAllQuadratic(self): - with self.test_session(): + with self.cached_session(): predictions = constant_op.constant([1.5, -1.4, -1.0, 0.0]) labels = constant_op.constant([1.0, -1.0, 0.0, 0.5]) loss = losses.huber_loss(labels, predictions) @@ -859,7 +859,7 @@ class HuberLossTest(test.TestCase): 0.5 * (0.25 + 0.16 + 1.0 + 0.25) / 4., atol=1e-5) def testAllLinear(self): - with self.test_session(): + with self.cached_session(): predictions = constant_op.constant([1.5, -1.4, -1.0, 0.0]) labels = constant_op.constant([0.0, 1.0, 0.0, 1.5]) loss = losses.huber_loss(labels, predictions) @@ -867,7 +867,7 @@ class HuberLossTest(test.TestCase): (1.5 + 2.4 + 1.0 + 1.5) / 4. - 0.5, atol=1e-5) def testMixedQuadraticLinear(self): - with self.test_session(): + with self.cached_session(): predictions = constant_op.constant([[1.5, -1.4, -1.0, 0.0], [1.5, -1.4, -1.0, 0.0]]) labels = constant_op.constant([[1.0, -1.0, 0.0, 0.5], @@ -879,7 +879,7 @@ class HuberLossTest(test.TestCase): self.assertAllClose(loss.eval(), expected_loss, atol=1e-5) def testAllQuadraticDelta(self): - with self.test_session(): + with self.cached_session(): delta = 0.5 predictions = constant_op.constant([1.5, -1.4, -0.5, 0.0]) labels = constant_op.constant([1.0, -1.0, 0.0, 0.5]) @@ -894,7 +894,7 @@ class HuberLossTest(test.TestCase): expected = delta * np.array([1.5, 2.4, 1.0, 1.5]).mean() expected -= 0.5 * delta**2 loss = losses.huber_loss(labels, predictions, delta=delta) - with self.test_session(): + with self.cached_session(): self.assertAllClose(expected, loss.eval(), atol=1e-5) @@ -906,13 +906,13 @@ class MeanSquaredErrorTest(test.TestCase): self._labels = constant_op.constant([1, 9, 2, -5, -2, 6], shape=(2, 3)) def testValueErrorThrownWhenWeightIsNone(self): - with self.test_session(): + with self.cached_session(): with self.assertRaises(ValueError): losses.mean_squared_error( self._predictions, self._predictions, weights=None) def testScalar(self): - with self.test_session(): + with self.cached_session(): self.assertEqual( 0.0, losses.mean_squared_error(predictions=constant_op.constant(0), @@ -920,55 +920,55 @@ class MeanSquaredErrorTest(test.TestCase): def testAllCorrectNoLossWeight(self): loss = losses.mean_squared_error(self._predictions, self._predictions) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(0.0, loss.eval(), 3) def testNonZeroLoss(self): loss = losses.mean_squared_error(self._labels, self._predictions) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(49.5, loss.eval(), 3) def testNonZeroLossWithPythonScalarWeight(self): weights = 2.3 loss = losses.mean_squared_error(self._labels, self._predictions, weights) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(49.5 * weights, loss.eval(), 3) def testNonZeroLossWithScalarTensorWeight(self): weights = 2.3 loss = losses.mean_squared_error(self._labels, self._predictions, constant_op.constant(weights)) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(49.5 * weights, loss.eval(), 3) def testNonZeroLossWithOneDimBatchSpecificWeights(self): weights = constant_op.constant([1.2, 3.4], shape=(2, 1)) loss = losses.mean_squared_error(self._labels, self._predictions, weights) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(767.8 / 6.0, loss.eval(), 3) def testNonZeroLossWithTwoDimBatchSpecificWeights(self): weights = constant_op.constant([1.2, 3.4], shape=[2, 1]) loss = losses.mean_squared_error(self._labels, self._predictions, weights) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(767.8 / 6.0, loss.eval(), 3) def testNonZeroLossWithSampleSpecificWeights(self): weights = constant_op.constant([3, 6, 5, 0, 4, 2], shape=[2, 3]) loss = losses.mean_squared_error(self._labels, self._predictions, weights) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(587 / 5.0, loss.eval(), 3) def testNonZeroLossWithSampleSpecificWeightsMostZero(self): weights = constant_op.constant([0, 0, 0, 0, 0, 2], shape=[2, 3]) loss = losses.mean_squared_error(self._labels, self._predictions, weights) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(18.0, loss.eval(), 3) def testLossWithSampleSpecificWeightsAllZero(self): weights = array_ops.zeros((2, 3)) loss = losses.mean_squared_error(self._labels, self._predictions, weights) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(0.0, loss.eval(), 3) @@ -994,7 +994,7 @@ class MeanPairwiseSquaredErrorTest(test.TestCase): self._expected_losses = np.divide(total, 3.0) def testValueErrorThrownWhenWeightIsNone(self): - with self.test_session(): + with self.cached_session(): with self.assertRaises(ValueError): losses.mean_pairwise_squared_error( predictions=constant_op.constant(self._labels), @@ -1003,7 +1003,7 @@ class MeanPairwiseSquaredErrorTest(test.TestCase): def _test_valid_weights( self, labels, predictions, expected_loss, weights=1.0): - with self.test_session(): + with self.cached_session(): static_inputs_op = losses.mean_pairwise_squared_error( predictions=predictions, labels=labels, weights=weights) self.assertAlmostEqual(expected_loss, static_inputs_op.eval(), places=3) @@ -1054,7 +1054,7 @@ class MeanPairwiseSquaredErrorTest(test.TestCase): init_op = variables.global_variables_initializer() - with self.test_session() as sess: + with self.cached_session() as sess: sess.run(init_op) for grad, _ in gradients_to_variables: np_grad = sess.run(grad) @@ -1073,7 +1073,7 @@ class MeanPairwiseSquaredErrorTest(test.TestCase): predictions=constant_op.constant(self._predictions), labels=constant_op.constant(self._labels), weights=constant_op.constant(weights)) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(weights * np.sum(self._expected_losses), loss.eval(), 3) @@ -1122,7 +1122,7 @@ class MeanPairwiseSquaredErrorTest(test.TestCase): predictions=predictions_placeholder, labels=labels_placeholder, weights=weights_placeholder) - with self.test_session(): + with self.cached_session(): with self.assertRaisesRegexp(errors_impl.OpError, expected_error_msg): dynamic_inputs_op.eval(feed_dict={ predictions_placeholder: predictions, @@ -1191,7 +1191,7 @@ class MeanPairwiseSquaredErrorTest(test.TestCase): labels=array_ops.concat([labels0, labels1], 0), predictions=array_ops.concat([predictions0, predictions1], 0)) - with self.test_session() as session: + with self.cached_session() as session: loss0, loss1, loss0_1 = session.run([loss0, loss1, loss0_1]) self.assertTrue(loss0 > 0) @@ -1216,7 +1216,7 @@ class CosineDistanceLossTest(test.TestCase): [0, 0, 1], [0, 1, 0]]).reshape((3, 2, 3)) def testValueErrorThrownWhenWeightIsNone(self): - with self.test_session(): + with self.cached_session(): with self.assertRaises(ValueError): losses.cosine_distance( predictions=constant_op.constant(self._labels), @@ -1229,7 +1229,7 @@ class CosineDistanceLossTest(test.TestCase): predictions=constant_op.constant(self._labels), labels=constant_op.constant(self._labels), dim=2) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(0, loss.eval(), 5) def testPartiallyCorrectWithIntegerValues(self): @@ -1237,7 +1237,7 @@ class CosineDistanceLossTest(test.TestCase): predictions=constant_op.constant(self._predictions), labels=constant_op.constant(self._labels), dim=2) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(1, loss.eval(), 5) def testPartiallyCorrectFloatingPointValues(self): @@ -1255,7 +1255,7 @@ class CosineDistanceLossTest(test.TestCase): labels, shape=(3, 1, 3), dtype=dtypes.float32) loss = losses.cosine_distance(tf_labels, tf_preds, dim=2) - with self.test_session(): + with self.cached_session(): self.assertAlmostEqual(1.0, loss.eval(), 5) def testSampleSpecificWeights(self): @@ -1264,7 +1264,7 @@ class CosineDistanceLossTest(test.TestCase): labels=constant_op.constant(self._labels), dim=2, weights=np.asarray((1, 0, 0)).reshape((3, 1, 1))) - with self.test_session(): + with self.cached_session(): self.assertEqual(1.0, loss.eval()) def testMeasurementSpecificWeights(self): @@ -1274,7 +1274,7 @@ class CosineDistanceLossTest(test.TestCase): dim=2, weights=constant_op.constant( [1, 0, 0, 1, 1, 1], shape=(3, 2, 1))) - with self.test_session(): + with self.cached_session(): self.assertEqual(3.0 / 4.0, loss.eval()) def testMeasurementSpecificWeightsWithPlaceholderWithShape(self): @@ -1286,7 +1286,7 @@ class CosineDistanceLossTest(test.TestCase): dim=2, weights=constant_op.constant( [1, 0, 0, 1, 1, 1], shape=(3, 2, 1))) - with self.test_session() as sess: + with self.cached_session() as sess: loss = sess.run(loss, feed_dict={tf_predictions: self._predictions}) self.assertEqual(3.0 / 4.0, loss) @@ -1296,7 +1296,7 @@ class CosineDistanceLossTest(test.TestCase): labels=constant_op.constant(self._labels), dim=2, weights=array_ops.zeros((3, 1, 1))) - with self.test_session(): + with self.cached_session(): self.assertEqual(0, loss.eval()) def testZeroLossWhenAllMeasurementSpecificWeightsAreZero(self): @@ -1305,7 +1305,7 @@ class CosineDistanceLossTest(test.TestCase): labels=constant_op.constant(self._labels), dim=2, weights=array_ops.zeros((3, 2, 1))) - with self.test_session(): + with self.cached_session(): self.assertEqual(0, loss.eval()) @@ -1411,7 +1411,7 @@ class ComputeWeightedLossTest(test.TestCase): weighted_loss = losses.compute_weighted_loss( self._raw_losses, weights=weight) self.assertEqual(1, len(util.get_losses())) - with self.test_session(): + with self.cached_session(): self.assertAllClose( np.mean(weight * self._raw_losses), weighted_loss.eval()) @@ -1429,7 +1429,7 @@ class ComputeWeightedLossTest(test.TestCase): weighted_loss = losses.compute_weighted_loss( self._raw_losses, weights=weights_placeholder) self.assertEqual(1, len(util.get_losses())) - with self.test_session(): + with self.cached_session(): with self.assertRaisesRegexp(errors_impl.OpError, expected_error_msg): weighted_loss.eval(feed_dict={weights_placeholder: weights}) @@ -1452,7 +1452,7 @@ class ComputeWeightedLossTest(test.TestCase): weighted_loss = losses.compute_weighted_loss( raw_losses, weights=weights_placeholder) self.assertEqual(1, len(util.get_losses())) - with self.test_session(): + with self.cached_session(): with self.assertRaisesRegexp(errors_impl.OpError, expected_error_msg): weighted_loss.eval(feed_dict={weights_placeholder: weights}) |