def test_weights_in_out_none(self):
"""Test case with no weights at all."""
scope = 'StreamingLossScope'
targets = tf.constant([[0, 1, 2], [0, 9, 23]], dtype=tf.int32)
predictions = tf.constant([[0, 1, 2], [0, 9, 23]], dtype=tf.int32)
values = tf.constant([5, 6, 7], dtype=tf.float32)
func = mock.Mock()
func.side_effect = [(values, None)]
avg = streaming.StreamingAverage()
avg.compute = mock.MagicMock()
loss = losses.StreamingLoss(func, avg)
loss.compute(targets, predictions, scope=scope)
func.assert_called_once_with(targets, predictions, None)
avg.compute.assert_called_once()
args, kwargs = avg.compute.call_args
act_values, = args
self.assertEqual(act_values, values)
self.assertIn('weights', kwargs)
self.assertEqual(kwargs.pop('weights'), None)
self.assertIn('scope', kwargs)
self.assertEqual(kwargs.pop('scope').name, scope)
def test_weights_out_none(self):
"""Test case with no weights returned by the wrapped function."""
scope = 'MyScope'
targets = tf.constant([[0, 1, 2], [0, 9, 23]], dtype=tf.int32)
predictions = tf.constant([[0, 1, 2], [0, 9, 23]], dtype=tf.int32)
weights = tf.constant([[1, 1, 1], [0, 0, 1]], dtype=tf.float32)
values = tf.constant([5, 6, 7], dtype=tf.float32)
func = mock.Mock()
func.side_effect = [(values, None)]
avg = streaming.StreamingAverage()
avg.compute = mock.MagicMock()
metric = metrics.StreamingMetric(func, avg)
metric.compute(targets, predictions, weights, scope=scope)
func.assert_called_once_with(targets, predictions, weights)
avg.compute.assert_called_once()
args, kwargs = avg.compute.call_args
act_values, act_weights_out = args
self.assertEqual(act_values, values)
self.assertEqual(act_weights_out, None)
self.assertIn('scope', kwargs)
self.assertEqual(kwargs.pop('scope').name, scope)
def test_default(self):
"""Default test for the StreamingAvrage class."""
# Set up the data.
values_01 = np.asarray([[1, 2, 3], [4, 5, 1000]], dtype=np.float32) # pylint: disable=I0011,E1101
weights_01 = np.asarray([[1, 1, 1], [1, 1, 0]], dtype=np.float32) # pylint: disable=I0011,E1101
avg_01 = 3.0
values_02 = np.asarray([[1000, 1000, 8], [9, 10, 1000]],
dtype=np.float32) # pylint: disable=I0011,E1101
weights_02 = np.asarray([[0, 0, 1], [1, 1, 0]], dtype=np.float32) # pylint: disable=I0011,E1101
avg_02 = 9.0
avg = 5.25
# Build the graph.
values = tf.placeholder(dtype=tf.float32, shape=[2, 3], name='values')
weights = tf.placeholder(dtype=tf.float32,
shape=[2, 3],
name='weights')
streaming_avg = streaming.StreamingAverage(name='StreamingAvg')
streaming_avg.compute(values, weights)
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
sess.run(tf.local_variables_initializer())
# First batch.
feed_dict = {values: values_01, weights: weights_01}
sess.run(streaming_avg.update_op, feed_dict)
self.assertEqual(avg_01,
sess.run(streaming_avg.batch_value, feed_dict))
self.assertEqual(avg_01, sess.run(streaming_avg.value, feed_dict))
# Second batch.
feed_dict = {
values: values_02,
weights: weights_02,
}
sess.run(streaming_avg.update_op, feed_dict)
self.assertEqual(avg_02,
sess.run(streaming_avg.batch_value, feed_dict))
self.assertEqual(avg, sess.run(streaming_avg.value, feed_dict))
# Reset.
sess.run(streaming_avg.reset_op)
self.assertEqual(0.0, sess.run(streaming_avg.value))
self.assertEqual(0, sess.run(streaming_avg.count))
# Second batch as first.
feed_dict = {
values: values_02,
weights: weights_02,
}
sess.run(streaming_avg.update_op, feed_dict)
self.assertEqual(avg_02,
sess.run(streaming_avg.batch_value, feed_dict))
self.assertEqual(avg_02, sess.run(streaming_avg.value, feed_dict))
def __init__(self, func, average=None, name=None):
super(StreamingMetric, self).__init__(name=name)
self._func = func
self._avg = average or streaming.StreamingAverage()