def __init__(self,
encoder,
predictor,
name='AverageKL',
prefix='Metrics',
dtype=tf.float32,
batch_size=1,
buffer_size=1000):
super(AverageKLMetric, self).__init__(name=name,
prefix=prefix,
metric_names=('KL', 'MSE'))
self._kl_buffer = tf_metrics.TFDeque(buffer_size, dtype)
self._mse_buffer = tf_metrics.TFDeque(buffer_size, dtype)
self._dtype = dtype
self._encoder = encoder
self._predictor = predictor
def test_mean_not_full(self):
d = tf_metrics.TFDeque(3, tf.int32)
self.evaluate(tf.compat.v1.global_variables_initializer())
self.evaluate(d.add(2))
self.evaluate(d.add(4))
self.assertEqual(3.0, self.evaluate(d.mean()))
def test_extend(self):
d = tf_metrics.TFDeque(3, tf.int32, shape=(2,))
self.evaluate(tf.compat.v1.global_variables_initializer())
self.evaluate(d.extend([[1, 1], [2, 2], [3, 3], [4, 4]]))
self.assertAllEqual([[4, 4], [2, 2], [3, 3]], self.evaluate(d.data))
self.assertAllEqual([3, 3], self.evaluate(d.mean()))
def test_extend(self):
d = tf_metrics.TFDeque(3, tf.float32)
self.evaluate(tf.compat.v1.global_variables_initializer())
self.evaluate(d.extend([1, 2, 3, 4]))
self.assertEqual(3.0, self.evaluate(d.mean()))
self.assertEqual(tf.float32, d.mean().dtype)
def test_mean_not_full(self):
d = tf_metrics.TFDeque(3, tf.int32, shape=(2,))
self.evaluate(tf.compat.v1.global_variables_initializer())
self.evaluate(d.add([1, 1]))
self.evaluate(d.add([3, 3]))
self.assertAllEqual([2, 2], self.evaluate(d.mean()))
self.assertEqual(tf.int32, d.mean().dtype)
def __init__(self,
prefix='Metrics',
dtype=tf.float32,
batch_size=1,
buffer_size=10,
obs_dim=None,
start_index=0,
end_index=None,
name=None):
assert obs_dim is not None
self._start_index = start_index
self._end_index = end_index
self._obs_dim = obs_dim
name = self.NAME if name is None else name
super(BaseDistanceMetric, self).__init__(name=name, prefix=prefix)
self._buffer = tf_metrics.TFDeque(buffer_size, dtype)
self._dist_buffer = tf_metrics.TFDeque(
1000, dtype) # Episodes should have length less than 1k
self.dtype = dtype
def test_queue_empty(self):
d = tf_metrics.TFDeque(3, tf.int32)
self.evaluate(tf.compat.v1.global_variables_initializer())
self.assertEqual(0, self.evaluate(d.mean()))
self.assertEqual(tf.int32.min, self.evaluate(d.max()))
self.assertEqual(tf.int32.max, self.evaluate(d.min()))
self.assertEqual(tf.int32, d.mean().dtype)
self.assertEqual(tf.int32, d.min().dtype)
self.assertEqual(tf.int32, d.max().dtype)
def test_mean_roll_over(self):
d = tf_metrics.TFDeque(3, tf.float32)
self.evaluate(tf.compat.v1.global_variables_initializer())
self.evaluate(d.add(1))
self.evaluate(d.add(2))
self.evaluate(d.add(3))
self.evaluate(d.add(4))
self.assertEqual(3.0, self.evaluate(d.mean()))
self.assertEqual(tf.float32, d.mean().dtype)
def test_mean_roll_over(self):
d = tf_metrics.TFDeque(3, tf.float32, shape=(2,))
self.evaluate(tf.compat.v1.global_variables_initializer())
self.evaluate(d.add([1, 1]))
self.evaluate(d.add([2, 2]))
self.evaluate(d.add([3, 3]))
self.evaluate(d.add([4, 4]))
self.assertAllEqual([3.0, 3.0], self.evaluate(d.mean()))
self.assertEqual(tf.float32, d.mean().dtype)
def test_rolls_over(self):
d = tf_metrics.TFDeque(3, tf.int32)
self.evaluate(tf.compat.v1.global_variables_initializer())
self.evaluate(d.add(1))
self.evaluate(d.add(2))
self.evaluate(d.add(3))
self.assertAllEqual([1, 2, 3], self.evaluate(d.data))
self.evaluate(d.add(4))
self.assertAllEqual([4, 2, 3], self.evaluate(d.data))
def test_rolls_over(self):
d = tf_metrics.TFDeque(3, tf.int32, shape=(2,))
self.evaluate(tf.compat.v1.global_variables_initializer())
self.evaluate(d.add([1, 1]))
self.evaluate(d.add([2, 2]))
self.evaluate(d.add([3, 3]))
self.assertAllEqual([[1, 1], [2, 2], [3, 3]], self.evaluate(d.data))
self.evaluate(d.add([4, 4]))
self.assertAllEqual([[4, 4], [2, 2], [3, 3]], self.evaluate(d.data))
def test_clear(self):
d = tf_metrics.TFDeque(3, tf.int32, shape=(2,))
self.evaluate(tf.compat.v1.global_variables_initializer())
self.evaluate(d.add([1, 1]))
self.evaluate(d.clear())
self.assertAllEqual((0, 2), self.evaluate(d.data).shape)
self.evaluate(d.add([2, 2]))
self.evaluate(d.add([3, 3]))
self.assertAllEqual([[2, 2], [3, 3]], self.evaluate(d.data))
def test_data_is_zero(self): d = tf_metrics.TFDeque(3, tf.int32) self.evaluate(tf.compat.v1.global_variables_initializer()) self.assertAllEqual([], self.evaluate(d.data))
def test_mean_empty(self):
d = tf_metrics.TFDeque(3, tf.int32)
self.evaluate(tf.compat.v1.global_variables_initializer())
self.assertEqual(0, self.evaluate(d.mean()))
def test_mean_empty(self):
d = tf_metrics.TFDeque(3, tf.int32, shape=(2,))
self.evaluate(tf.compat.v1.global_variables_initializer())
self.assertAllEqual([0, 0], self.evaluate(d.mean()))
self.assertEqual(tf.int32, d.mean().dtype)
def test_max(self): d = tf_metrics.TFDeque(4, tf.float32) self.evaluate(tf.compat.v1.global_variables_initializer()) self.evaluate(d.extend([1, 2, 3, 4])) self.assertEqual(4.0, self.evaluate(d.max()))
def test_data_is_zero(self): d = tf_metrics.TFDeque(3, tf.int32, shape=(2,)) self.evaluate(tf.compat.v1.global_variables_initializer()) values = self.evaluate(d.data) self.assertAllEqual((0, 2), values.shape)