How to use scale method in ATX

Best Python code snippet using ATX

affine.py

Source:affine.py

...266                array_ops.zeros([], identity_multiplier.dtype),267                ["identity_multiplier should be non-zero."])],268            identity_multiplier)269      return identity_multiplier270    scale = distribution_util.make_tril_scale(271        loc=shift,272        scale_tril=tril,273        scale_diag=diag,274        scale_identity_multiplier=identity_multiplier,275        validate_args=validate_args,276        assert_positive=False,277        shape_hint=shape_hint)278    if perturb_factor is not None:279      return linalg.LinearOperatorLowRankUpdate(280          scale,281          u=perturb_factor,282          diag_update=perturb_diag,283          is_diag_update_positive=perturb_diag is None,284          is_non_singular=True,  # Implied by is_positive_definite=True.285          is_self_adjoint=True,286          is_positive_definite=True,287          is_square=True)288    return scale289  @property290  def shift(self):291    """The `shift` `Tensor` in `Y = scale @ X + shift`."""292    return self._shift293  @property294  def scale(self):295    """The `scale` `LinearOperator` in `Y = scale @ X + shift`."""296    return self._scale297  def _forward(self, x):298    y = x299    if self._is_only_identity_multiplier:300      y *= self._scale301      if self.shift is not None:302        return y + self.shift303      return y304    y, sample_shape = self._shaper.make_batch_of_event_sample_matrices(305        y, expand_batch_dim=False)306    with ops.control_dependencies(self._maybe_check_scale() if307                                  self.validate_args else []):308      y = self.scale.matmul(y)309    y = self._shaper.undo_make_batch_of_event_sample_matrices(310        y, sample_shape, expand_batch_dim=False)311    if self.shift is not None:312      y += self.shift313    return y314  def _inverse(self, y):315    x = y316    if self.shift is not None:317      x -= self.shift318    if self._is_only_identity_multiplier:319      return x / self._scale320    x, sample_shape = self._shaper.make_batch_of_event_sample_matrices(321        x, expand_batch_dim=False)322    # Solve fails if the op is singular so we may safely skip this assertion.323    x = self.scale.solve(x)324    x = self._shaper.undo_make_batch_of_event_sample_matrices(325        x, sample_shape, expand_batch_dim=False)326    return x327  def _forward_log_det_jacobian(self, x):328    # is_constant_jacobian = True for this bijector, hence the329    # `log_det_jacobian` need only be specified for a single input, as this will330    # be tiled to match `event_ndims`.331    if self._is_only_identity_multiplier:332      # We don't pad in this case and instead let the fldj be applied333      # via broadcast.334      event_size = array_ops.shape(x)[-1]335      event_size = math_ops.cast(event_size, dtype=self._scale.dtype)336      return math_ops.log(math_ops.abs(self._scale)) * event_size337    return self.scale.log_abs_determinant()338  def _maybe_check_scale(self):339    try:340      return [self.scale.assert_non_singular()]341    except NotImplementedError:342      pass...

laplace_test.py

Source:laplace_test.py

1# Copyright 2016 The TensorFlow Authors. All Rights Reserved.2#3# Licensed under the Apache License, Version 2.0 (the "License");4# you may not use this file except in compliance with the License.5# You may obtain a copy of the License at6#7#     http://www.apache.org/licenses/LICENSE-2.08#9# Unless required by applicable law or agreed to in writing, software10# distributed under the License is distributed on an "AS IS" BASIS,11# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.12# See the License for the specific language governing permissions and13# limitations under the License.14# ==============================================================================15from __future__ import absolute_import16from __future__ import division17from __future__ import print_function18import numpy as np19from scipy import stats20import tensorflow as tf21class LaplaceTest(tf.test.TestCase):22  def testLaplaceShape(self):23    with self.test_session():24      loc = tf.constant([3.0] * 5)25      scale = tf.constant(11.0)26      laplace = tf.contrib.distributions.Laplace(loc=loc, scale=scale)27      self.assertEqual(laplace.batch_shape().eval(), (5,))28      self.assertEqual(laplace.get_batch_shape(), tf.TensorShape([5]))29      self.assertAllEqual(laplace.event_shape().eval(), [])30      self.assertEqual(laplace.get_event_shape(), tf.TensorShape([]))31  def testLaplaceLogPDF(self):32    with self.test_session():33      batch_size = 634      loc = tf.constant([2.0] * batch_size)35      scale = tf.constant([3.0] * batch_size)36      loc_v = 2.037      scale_v = 3.038      x = np.array([2.5, 2.5, 4.0, 0.1, 1.0, 2.0], dtype=np.float32)39      laplace = tf.contrib.distributions.Laplace(loc=loc, scale=scale)40      expected_log_pdf = stats.laplace.logpdf(x, loc_v, scale=scale_v)41      log_pdf = laplace.log_pdf(x)42      self.assertEqual(log_pdf.get_shape(), (6,))43      self.assertAllClose(log_pdf.eval(), expected_log_pdf)44      pdf = laplace.pdf(x)45      self.assertEqual(pdf.get_shape(), (6,))46      self.assertAllClose(pdf.eval(), np.exp(expected_log_pdf))47  def testLaplaceLogPDFMultidimensional(self):48    with self.test_session():49      batch_size = 650      loc = tf.constant([[2.0, 4.0]] * batch_size)51      scale = tf.constant([[3.0, 4.0]] * batch_size)52      loc_v = np.array([2.0, 4.0])53      scale_v = np.array([3.0, 4.0])54      x = np.array([[2.5, 2.5, 4.0, 0.1, 1.0, 2.0]], dtype=np.float32).T55      laplace = tf.contrib.distributions.Laplace(loc=loc, scale=scale)56      expected_log_pdf = stats.laplace.logpdf(x, loc_v, scale=scale_v)57      log_pdf = laplace.log_pdf(x)58      log_pdf_values = log_pdf.eval()59      self.assertEqual(log_pdf.get_shape(), (6, 2))60      self.assertAllClose(log_pdf_values, expected_log_pdf)61      pdf = laplace.pdf(x)62      pdf_values = pdf.eval()63      self.assertEqual(pdf.get_shape(), (6, 2))64      self.assertAllClose(pdf_values, np.exp(expected_log_pdf))65  def testLaplaceLogPDFMultidimensionalBroadcasting(self):66    with self.test_session():67      batch_size = 668      loc = tf.constant([[2.0, 4.0]] * batch_size)69      scale = tf.constant(3.0)70      loc_v = np.array([2.0, 4.0])71      scale_v = 3.072      x = np.array([[2.5, 2.5, 4.0, 0.1, 1.0, 2.0]], dtype=np.float32).T73      laplace = tf.contrib.distributions.Laplace(loc=loc, scale=scale)74      expected_log_pdf = stats.laplace.logpdf(x, loc_v, scale=scale_v)75      log_pdf = laplace.log_pdf(x)76      log_pdf_values = log_pdf.eval()77      self.assertEqual(log_pdf.get_shape(), (6, 2))78      self.assertAllClose(log_pdf_values, expected_log_pdf)79      pdf = laplace.pdf(x)80      pdf_values = pdf.eval()81      self.assertEqual(pdf.get_shape(), (6, 2))82      self.assertAllClose(pdf_values, np.exp(expected_log_pdf))83  def testLaplaceCDF(self):84    with self.test_session():85      batch_size = 686      loc = tf.constant([2.0] * batch_size)87      scale = tf.constant([3.0] * batch_size)88      loc_v = 2.089      scale_v = 3.090      x = np.array([2.5, 2.5, 4.0, 0.1, 1.0, 2.0], dtype=np.float32)91      laplace = tf.contrib.distributions.Laplace(loc=loc, scale=scale)92      expected_cdf = stats.laplace.cdf(x, loc_v, scale=scale_v)93      cdf = laplace.cdf(x)94      self.assertEqual(cdf.get_shape(), (6,))95      self.assertAllClose(cdf.eval(), expected_cdf)96  def testLaplaceMean(self):97    with self.test_session():98      loc_v = np.array([1.0, 3.0, 2.5])99      scale_v = np.array([1.0, 4.0, 5.0])100      laplace = tf.contrib.distributions.Laplace(loc=loc_v, scale=scale_v)101      expected_means = stats.laplace.mean(loc_v, scale=scale_v)102      self.assertEqual(laplace.mean().get_shape(), (3,))103      self.assertAllClose(laplace.mean().eval(), expected_means)104  def testLaplaceMode(self):105    with self.test_session():106      loc_v = np.array([0.5, 3.0, 2.5])107      scale_v = np.array([1.0, 4.0, 5.0])108      laplace = tf.contrib.distributions.Laplace(loc=loc_v, scale=scale_v)109      self.assertEqual(laplace.mode().get_shape(), (3,))110      self.assertAllClose(laplace.mode().eval(), loc_v)111  def testLaplaceVariance(self):112    with self.test_session():113      loc_v = np.array([1.0, 3.0, 2.5])114      scale_v = np.array([1.0, 4.0, 5.0])115      laplace = tf.contrib.distributions.Laplace(loc=loc_v, scale=scale_v)116      expected_variances = stats.laplace.var(loc_v, scale=scale_v)117      self.assertEqual(laplace.variance().get_shape(), (3,))118      self.assertAllClose(laplace.variance().eval(), expected_variances)119  def testLaplaceStd(self):120    with self.test_session():121      loc_v = np.array([1.0, 3.0, 2.5])122      scale_v = np.array([1.0, 4.0, 5.0])123      laplace = tf.contrib.distributions.Laplace(loc=loc_v, scale=scale_v)124      expected_std = stats.laplace.std(loc_v, scale=scale_v)125      self.assertEqual(laplace.std().get_shape(), (3,))126      self.assertAllClose(laplace.std().eval(), expected_std)127  def testLaplaceEntropy(self):128    with self.test_session():129      loc_v = np.array([1.0, 3.0, 2.5])130      scale_v = np.array([1.0, 4.0, 5.0])131      expected_entropy = stats.laplace.entropy(loc_v, scale=scale_v)132      laplace = tf.contrib.distributions.Laplace(loc=loc_v, scale=scale_v)133      self.assertEqual(laplace.entropy().get_shape(), (3,))134      self.assertAllClose(laplace.entropy().eval(), expected_entropy)135  def testLaplaceSample(self):136    with tf.Session():137      loc_v = 4.0138      scale_v = 3.0139      loc = tf.constant(loc_v)140      scale = tf.constant(scale_v)141      n = 100000142      laplace = tf.contrib.distributions.Laplace(loc=loc, scale=scale)143      samples = laplace.sample(n, seed=137)144      sample_values = samples.eval()145      self.assertEqual(samples.get_shape(), (n,))146      self.assertEqual(sample_values.shape, (n,))147      self.assertAllClose(sample_values.mean(),148                          stats.laplace.mean(loc_v, scale=scale_v),149                          rtol=0.05, atol=0.)150      self.assertAllClose(sample_values.var(),151                          stats.laplace.var(loc_v, scale=scale_v),152                          rtol=0.05, atol=0.)153      self.assertTrue(self._kstest(loc_v, scale_v, sample_values))154  def testLaplaceSampleMultiDimensional(self):155    with tf.Session():156      loc_v = np.array([np.arange(1, 101, dtype=np.float32)])  # 1 x 100157      scale_v = np.array([np.arange(1, 11, dtype=np.float32)]).T  # 10 x 1158      laplace = tf.contrib.distributions.Laplace(loc=loc_v, scale=scale_v)159      n = 10000160      samples = laplace.sample(n, seed=137)161      sample_values = samples.eval()162      self.assertEqual(samples.get_shape(), (n, 10, 100))163      self.assertEqual(sample_values.shape, (n, 10, 100))164      zeros = np.zeros_like(loc_v + scale_v)  # 10 x 100165      loc_bc = loc_v + zeros166      scale_bc = scale_v + zeros167      self.assertAllClose(168          sample_values.mean(axis=0),169          stats.laplace.mean(loc_bc, scale=scale_bc),170          rtol=0.35, atol=0.)171      self.assertAllClose(172          sample_values.var(axis=0),173          stats.laplace.var(loc_bc, scale=scale_bc),174          rtol=0.10, atol=0.)175      fails = 0176      trials = 0177      for ai, a in enumerate(np.reshape(loc_v, [-1])):178        for bi, b in enumerate(np.reshape(scale_v, [-1])):179          s = sample_values[:, bi, ai]180          trials += 1181          fails += 0 if self._kstest(a, b, s) else 1182      self.assertLess(fails, trials * 0.03)183  def _kstest(self, loc, scale, samples):184    # Uses the Kolmogorov-Smirnov test for goodness of fit.185    ks, _ = stats.kstest(samples, stats.laplace(loc, scale=scale).cdf)186    # Return True when the test passes.187    return ks < 0.02188  def testLaplacePdfOfSampleMultiDims(self):189    with tf.Session() as sess:190      laplace = tf.contrib.distributions.Laplace(191          loc=[7., 11.], scale=[[5.], [6.]])192      num = 50000193      samples = laplace.sample(num, seed=137)194      pdfs = laplace.pdf(samples)195      sample_vals, pdf_vals = sess.run([samples, pdfs])196      self.assertEqual(samples.get_shape(), (num, 2, 2))197      self.assertEqual(pdfs.get_shape(), (num, 2, 2))198      self.assertAllClose(199          stats.laplace.mean([[7., 11.], [7., 11.]],200                             scale=np.array([[5., 5.], [6., 6.]])),201          sample_vals.mean(axis=0),202          rtol=0.05, atol=0.)203      self.assertAllClose(204          stats.laplace.var([[7., 11.], [7., 11.]],205                            scale=np.array([[5., 5.], [6., 6.]])),206          sample_vals.var(axis=0),207          rtol=0.05, atol=0.)208      self._assertIntegral(sample_vals[:, 0, 0], pdf_vals[:, 0, 0], err=0.02)209      self._assertIntegral(sample_vals[:, 0, 1], pdf_vals[:, 0, 1], err=0.02)210      self._assertIntegral(sample_vals[:, 1, 0], pdf_vals[:, 1, 0], err=0.02)211      self._assertIntegral(sample_vals[:, 1, 1], pdf_vals[:, 1, 1], err=0.02)212  def _assertIntegral(self, sample_vals, pdf_vals, err=1e-3):213    s_p = zip(sample_vals, pdf_vals)214    prev = (0, 0)215    total = 0216    for k in sorted(s_p, key=lambda x: x[0]):217      pair_pdf = (k[1] + prev[1]) / 2218      total += (k[0] - prev[0]) * pair_pdf219      prev = k220    self.assertNear(1., total, err=err)221  def testLaplaceNonPositiveInitializationParamsRaises(self):222    with self.test_session():223      loc_v = tf.constant(0.0, name="loc")224      scale_v = tf.constant(-1.0, name="scale")225      laplace = tf.contrib.distributions.Laplace(226          loc=loc_v, scale=scale_v, validate_args=True)227      with self.assertRaisesOpError("scale"):228        laplace.mean().eval()229      loc_v = tf.constant(1.0, name="loc")230      scale_v = tf.constant(0.0, name="scale")231      laplace = tf.contrib.distributions.Laplace(232          loc=loc_v, scale=scale_v, validate_args=True)233      with self.assertRaisesOpError("scale"):234        laplace.mean().eval()235  def testLaplaceWithSoftplusScale(self):236    with self.test_session():237      loc_v = tf.constant([0.0, 1.0], name="loc")238      scale_v = tf.constant([-1.0, 2.0], name="scale")239      laplace = tf.contrib.distributions.LaplaceWithSoftplusScale(240          loc=loc_v, scale=scale_v)241      self.assertAllClose(tf.nn.softplus(scale_v).eval(), laplace.scale.eval())242      self.assertAllClose(loc_v.eval(), laplace.loc.eval())243if __name__ == "__main__":...

mvn_diag_plus_low_rank.py

Source:mvn_diag_plus_low_rank.py

...192          loc, scale_diag, scale_identity_multiplier, scale_perturb_factor,193          scale_perturb_diag]):194        has_low_rank = (scale_perturb_factor is not None or195                        scale_perturb_diag is not None)196        scale = distribution_util.make_diag_scale(197            loc=loc,198            scale_diag=scale_diag,199            scale_identity_multiplier=scale_identity_multiplier,200            validate_args=validate_args,201            assert_positive=has_low_rank)202        scale_perturb_factor = _convert_to_tensor(203            scale_perturb_factor,204            name="scale_perturb_factor")205        scale_perturb_diag = _convert_to_tensor(206            scale_perturb_diag,207            name="scale_perturb_diag")208        if has_low_rank:209          scale = linalg.LinearOperatorLowRankUpdate(210              scale,...

Automation Testing Tutorials

Learn to execute automation testing from scratch with LambdaTest Learning Hub. Right from setting up the prerequisites to run your first automation test, to following best practices and diving deeper into advanced test scenarios. LambdaTest Learning Hubs compile a list of step-by-step guides to help you be proficient with different test automation frameworks i.e. Selenium, Cypress, TestNG etc.