def test_scalar_collect_batch_weight_broadcast(self):
collector = StatisticsCollector()
collector.collect([2, 1, 7, 6], weight=1.)
self.assertTrue(collector.has_value)
self.assertEqual(collector.counter, 4)
self.assertAlmostEqual(collector.mean, 4.)
self.assertAlmostEqual(collector.square, 22.5)
self.assertAlmostEqual(collector.weight_sum, 4.)
def test_collect_empty_weight(self):
collector = StatisticsCollector()
collector.collect([2, 1, 7, 6], weight=[])
self.assertTrue(collector.has_value)
self.assertEqual(collector.counter, 4)
self.assertAlmostEqual(collector.mean, 4.)
self.assertAlmostEqual(collector.square, 22.5)
self.assertAlmostEqual(collector.weight_sum, 4.)
def test_scalar_collect_batch(self):
collector = StatisticsCollector()
collector.collect([2, 1, 7], weight=[1, 3, 6])
self.assertTrue(collector.has_value)
self.assertEqual(collector.counter, 3)
self.assertAlmostEqual(collector.mean, 4.7)
self.assertAlmostEqual(collector.square, 30.1)
self.assertAlmostEqual(collector.weight_sum, 10.)
def test_vector_collect(self):
collector = StatisticsCollector(shape=(3, 2))
arr = np.arange(6).reshape([3, 2])
collector.collect(arr)
self.assertTrue(collector.has_value)
self.assertEqual(collector.counter, 1)
np.testing.assert_almost_equal(collector.mean, arr)
np.testing.assert_almost_equal(collector.square, arr**2)
self.assertAlmostEqual(collector.weight_sum, 1.)
def test_collect_empty(self):
collector = StatisticsCollector()
collector.collect([])
self.assertEqual(collector.shape, ())
self.assertFalse(collector.has_value)
self.assertEqual(collector.counter, 0)
self.assertAlmostEqual(collector.mean, 0.)
self.assertAlmostEqual(collector.square, 0.)
self.assertEqual(collector.weight_sum, 0.)
def test_vector_collect_batch_weight_broadcast(self):
collector = StatisticsCollector(shape=(3, 2))
arr = np.arange(24).reshape([4, 3, 2])
collector.collect(arr, weight=1.)
self.assertTrue(collector.has_value)
self.assertEqual(collector.counter, 4)
np.testing.assert_almost_equal(collector.mean,
[[9, 10], [11, 12], [13, 14]])
np.testing.assert_almost_equal(
collector.square, [[126., 145.], [166., 189.], [214., 241.]])
self.assertAlmostEqual(collector.weight_sum, 4.)
def test_vector_collect_batch(self):
collector = StatisticsCollector(shape=(3, 2))
arr = np.arange(24).reshape([4, 3, 2])
collector.collect(arr, weight=[1, 2, 3, 4])
self.assertTrue(collector.has_value)
self.assertEqual(collector.counter, 4)
np.testing.assert_almost_equal(collector.mean,
[[12, 13], [14, 15], [16, 17]])
np.testing.assert_almost_equal(
collector.square, [[180., 205.], [232., 261.], [292., 325.]])
self.assertAlmostEqual(collector.weight_sum, 10.)
def test_empty_vector(self):
collector = StatisticsCollector(shape=(3, 2))
self.assertEqual(collector.shape, (3, 2))
self.assertFalse(collector.has_value)
self.assertEqual(collector.counter, 0)
np.testing.assert_almost_equal(collector.mean, np.zeros([3, 2]))
np.testing.assert_almost_equal(collector.square, np.zeros([3, 2]))
self.assertEqual(collector.weight_sum, 0.)
def test_scalar_multi_collect(self):
collector = StatisticsCollector()
collector.collect(2)
collector.collect(1, weight=3)
collector.collect(7, weight=6)
self.assertTrue(collector.has_value)
self.assertEqual(collector.counter, 3)
self.assertAlmostEqual(collector.mean, 4.7)
self.assertAlmostEqual(collector.square, 30.1)
self.assertAlmostEqual(collector.weight_sum, 10.)
def test_reset(self):
collector = StatisticsCollector()
collector.collect([2, 1, 7, 6])
collector.reset()
self.assertFalse(collector.has_value)
self.assertEqual(collector.counter, 0)
self.assertAlmostEqual(collector.mean, 0.)
self.assertAlmostEqual(collector.square, 0.)
self.assertAlmostEqual(collector.var, 0.)
self.assertAlmostEqual(collector.stddev, 0.)
self.assertAlmostEqual(collector.weight_sum, 0.)
def metric_collector(self, metric_name):
"""
Get a :class:`~tfsnippet.utils.StatisticsCollector` for metric.
The mean value of the collected metrics will be added to summary
after exiting the context. Other statistics will be discarded.
Args:
metric_name (str): The name of this metric.
Yields:
StatisticsCollector: The collector for metric values.
"""
self._require_context()
acc = StatisticsCollector()
yield acc
if acc.has_value:
self.collect_metrics(metrics={metric_name: acc.mean})
def test_vector_multi_collect(self):
collector = StatisticsCollector(shape=(3, 2))
arr = np.arange(24).reshape([4, 3, 2])
collector.collect(arr[0])
collector.collect(arr[1], weight=2)
collector.collect(arr[2], weight=3)
collector.collect(arr[3], weight=4)
self.assertTrue(collector.has_value)
self.assertEqual(collector.counter, 4)
np.testing.assert_almost_equal(collector.mean,
[[12, 13], [14, 15], [16, 17]])
np.testing.assert_almost_equal(
collector.square, [[180., 205.], [232., 261.], [292., 325.]])
np.testing.assert_almost_equal(
collector.var, np.maximum(collector.square - collector.mean**2,
0.))
np.testing.assert_almost_equal(
collector.stddev,
np.sqrt(np.maximum(collector.square - collector.mean**2, 0.)))
self.assertAlmostEqual(collector.weight_sum, 10.)
def test_var_std(self):
collector = StatisticsCollector()
collector.collect([2, 1, 7, 6])
self.assertEqual(collector.counter, 4)
self.assertAlmostEqual(collector.var, 6.5)
self.assertAlmostEqual(collector.stddev, 2.549509756796)
def test_shape_mismatch(self):
collector = StatisticsCollector(shape=(3, 2))
with pytest.raises(
ValueError,
match=r'Shape mismatch: \(3,\) not ending with \(3, 2\)'):
collector.collect([1, 2, 3])
