def test_sortquantile_shape(self):
axis = 1
x = tf.random.uniform((3, 20, 4), dtype=tf.float32)
soft_q = ops.softquantiles(x, 0.3, 0.03, axis=axis)
self.assertEqual(soft_q.shape, (3, 4))
soft_q2 = ops.softquantiles(x, 0.3, 0.03, axis=axis, may_squeeze=False)
self.assertEqual(soft_q2.shape, (3, 1, 4))
def call(self, inputs):
# For some reason, in graph mode the reshape is necessary.
return tf.reshape(
ops.softquantiles(inputs,
self._quantiles,
axis=self._axis,
**self._kwargs),
self.get_output_shape(tf.shape(inputs)))
def call(self, y_true, y_pred):
error = tf.pow(tf.abs(tf.squeeze(y_pred) - y_true), self._power)
width = self._end_quantile - self._start_quantile
quantile = 0.5 * (self._end_quantile + self._start_quantile)
return ops.softquantiles(error,
quantile,
quantile_width=width,
axis=0,
**self._kwargs)
def test_softquantiles(self):
num_points = 19
sorted_values = tf.range(0, num_points, dtype=tf.float32)
x = tf.random.shuffle(sorted_values)
target_quantiles = [0.25, 0.50, 0.75]
target_indices = [4, 9, 14]
soft_q = ops.softquantiles(x, target_quantiles, epsilon=1e-3)
hard_q = tf.gather(sorted_values, target_indices)
self.assertAllClose(soft_q, hard_q, 0.2, 0.2)
def test_softquantile(self, quantile):
# Builds the input vector so that the desired quantile always corresponds to
# an exact integer index.
num_points_before_quantile = 10
step = quantile / num_points_before_quantile
num_points = int(1.0 / step + 1.0)
quantile_width = step
axis = 1
x = tf.random.uniform((3, num_points, 4), dtype=tf.float32)
soft_q = ops.softquantiles(
x, quantile, quantile_width, axis=axis, epsilon=1e-3)
# Compare to getting the exact quantile.
hard_q = tf.gather(
tf.sort(x, direction='ASCENDING', axis=axis),
int(quantile * num_points), axis=1)
self.assertAllClose(soft_q, hard_q, 0.2, 0.2)
def call(self, y_true, y_pred):
error = tf.pow(tf.abs(tf.squeeze(y_pred) - y_true), self._power)
return ops.softquantiles(error, self._quantile, axis=0, **self._kwargs)
def test_softquantile_errors(self, q1, q2, width):
x = tf.random.uniform((3, 10))
with self.assertRaises(tf.errors.InvalidArgumentError):
ops.softquantiles(x, [q1, q2], quantile_width=width, axis=-1)