1 # Copyright 2015 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 at 6 # 7 # http://www.apache.org/licenses/LICENSE-2.0 8 # 9 # Unless required by applicable law or agreed to in writing, software 10 # 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 and 13 # limitations under the License. 14 # ============================================================================== 15 """Tests for Uniform distribution.""" 16 17 from __future__ import absolute_import 18 from __future__ import division 19 from __future__ import print_function 20 21 import importlib 22 23 import numpy as np 24 25 from tensorflow.python.framework import constant_op 26 from tensorflow.python.framework import errors 27 from tensorflow.python.framework import tensor_shape 28 from tensorflow.python.ops import array_ops 29 from tensorflow.python.ops import math_ops 30 from tensorflow.python.ops.distributions import uniform as uniform_lib 31 from tensorflow.python.platform import test 32 from tensorflow.python.platform import tf_logging 33 34 35 def try_import(name): # pylint: disable=invalid-name 36 module = None 37 try: 38 module = importlib.import_module(name) 39 except ImportError as e: 40 tf_logging.warning("Could not import %s: %s" % (name, str(e))) 41 return module 42 43 44 stats = try_import("scipy.stats") 45 46 47 class UniformTest(test.TestCase): 48 49 def testUniformRange(self): 50 with self.test_session(): 51 a = 3.0 52 b = 10.0 53 uniform = uniform_lib.Uniform(low=a, high=b) 54 self.assertAllClose(a, uniform.low.eval()) 55 self.assertAllClose(b, uniform.high.eval()) 56 self.assertAllClose(b - a, uniform.range().eval()) 57 58 def testUniformPDF(self): 59 with self.test_session(): 60 a = constant_op.constant([-3.0] * 5 + [15.0]) 61 b = constant_op.constant([11.0] * 5 + [20.0]) 62 uniform = uniform_lib.Uniform(low=a, high=b) 63 64 a_v = -3.0 65 b_v = 11.0 66 x = np.array([-10.5, 4.0, 0.0, 10.99, 11.3, 17.0], dtype=np.float32) 67 68 def _expected_pdf(): 69 pdf = np.zeros_like(x) + 1.0 / (b_v - a_v) 70 pdf[x > b_v] = 0.0 71 pdf[x < a_v] = 0.0 72 pdf[5] = 1.0 / (20.0 - 15.0) 73 return pdf 74 75 expected_pdf = _expected_pdf() 76 77 pdf = uniform.prob(x) 78 self.assertAllClose(expected_pdf, pdf.eval()) 79 80 log_pdf = uniform.log_prob(x) 81 self.assertAllClose(np.log(expected_pdf), log_pdf.eval()) 82 83 def testUniformShape(self): 84 with self.test_session(): 85 a = constant_op.constant([-3.0] * 5) 86 b = constant_op.constant(11.0) 87 uniform = uniform_lib.Uniform(low=a, high=b) 88 89 self.assertEqual(uniform.batch_shape_tensor().eval(), (5,)) 90 self.assertEqual(uniform.batch_shape, tensor_shape.TensorShape([5])) 91 self.assertAllEqual(uniform.event_shape_tensor().eval(), []) 92 self.assertEqual(uniform.event_shape, tensor_shape.TensorShape([])) 93 94 def testUniformPDFWithScalarEndpoint(self): 95 with self.test_session(): 96 a = constant_op.constant([0.0, 5.0]) 97 b = constant_op.constant(10.0) 98 uniform = uniform_lib.Uniform(low=a, high=b) 99 100 x = np.array([0.0, 8.0], dtype=np.float32) 101 expected_pdf = np.array([1.0 / (10.0 - 0.0), 1.0 / (10.0 - 5.0)]) 102 103 pdf = uniform.prob(x) 104 self.assertAllClose(expected_pdf, pdf.eval()) 105 106 def testUniformCDF(self): 107 with self.test_session(): 108 batch_size = 6 109 a = constant_op.constant([1.0] * batch_size) 110 b = constant_op.constant([11.0] * batch_size) 111 a_v = 1.0 112 b_v = 11.0 113 x = np.array([-2.5, 2.5, 4.0, 0.0, 10.99, 12.0], dtype=np.float32) 114 115 uniform = uniform_lib.Uniform(low=a, high=b) 116 117 def _expected_cdf(): 118 cdf = (x - a_v) / (b_v - a_v) 119 cdf[x >= b_v] = 1 120 cdf[x < a_v] = 0 121 return cdf 122 123 cdf = uniform.cdf(x) 124 self.assertAllClose(_expected_cdf(), cdf.eval()) 125 126 log_cdf = uniform.log_cdf(x) 127 self.assertAllClose(np.log(_expected_cdf()), log_cdf.eval()) 128 129 def testUniformEntropy(self): 130 with self.test_session(): 131 a_v = np.array([1.0, 1.0, 1.0]) 132 b_v = np.array([[1.5, 2.0, 3.0]]) 133 uniform = uniform_lib.Uniform(low=a_v, high=b_v) 134 135 expected_entropy = np.log(b_v - a_v) 136 self.assertAllClose(expected_entropy, uniform.entropy().eval()) 137 138 def testUniformAssertMaxGtMin(self): 139 with self.test_session(): 140 a_v = np.array([1.0, 1.0, 1.0], dtype=np.float32) 141 b_v = np.array([1.0, 2.0, 3.0], dtype=np.float32) 142 uniform = uniform_lib.Uniform(low=a_v, high=b_v, validate_args=True) 143 144 with self.assertRaisesWithPredicateMatch(errors.InvalidArgumentError, 145 "x < y"): 146 uniform.low.eval() 147 148 def testUniformSample(self): 149 with self.test_session(): 150 a = constant_op.constant([3.0, 4.0]) 151 b = constant_op.constant(13.0) 152 a1_v = 3.0 153 a2_v = 4.0 154 b_v = 13.0 155 n = constant_op.constant(100000) 156 uniform = uniform_lib.Uniform(low=a, high=b) 157 158 samples = uniform.sample(n, seed=137) 159 sample_values = samples.eval() 160 self.assertEqual(sample_values.shape, (100000, 2)) 161 self.assertAllClose( 162 sample_values[::, 0].mean(), (b_v + a1_v) / 2, atol=1e-2) 163 self.assertAllClose( 164 sample_values[::, 1].mean(), (b_v + a2_v) / 2, atol=1e-2) 165 self.assertFalse( 166 np.any(sample_values[::, 0] < a1_v) or np.any(sample_values >= b_v)) 167 self.assertFalse( 168 np.any(sample_values[::, 1] < a2_v) or np.any(sample_values >= b_v)) 169 170 def _testUniformSampleMultiDimensional(self): 171 # DISABLED: Please enable this test once b/issues/30149644 is resolved. 172 with self.test_session(): 173 batch_size = 2 174 a_v = [3.0, 22.0] 175 b_v = [13.0, 35.0] 176 a = constant_op.constant([a_v] * batch_size) 177 b = constant_op.constant([b_v] * batch_size) 178 179 uniform = uniform_lib.Uniform(low=a, high=b) 180 181 n_v = 100000 182 n = constant_op.constant(n_v) 183 samples = uniform.sample(n) 184 self.assertEqual(samples.get_shape(), (n_v, batch_size, 2)) 185 186 sample_values = samples.eval() 187 188 self.assertFalse( 189 np.any(sample_values[:, 0, 0] < a_v[0]) or 190 np.any(sample_values[:, 0, 0] >= b_v[0])) 191 self.assertFalse( 192 np.any(sample_values[:, 0, 1] < a_v[1]) or 193 np.any(sample_values[:, 0, 1] >= b_v[1])) 194 195 self.assertAllClose( 196 sample_values[:, 0, 0].mean(), (a_v[0] + b_v[0]) / 2, atol=1e-2) 197 self.assertAllClose( 198 sample_values[:, 0, 1].mean(), (a_v[1] + b_v[1]) / 2, atol=1e-2) 199 200 def testUniformMean(self): 201 with self.test_session(): 202 a = 10.0 203 b = 100.0 204 uniform = uniform_lib.Uniform(low=a, high=b) 205 if not stats: 206 return 207 s_uniform = stats.uniform(loc=a, scale=b - a) 208 self.assertAllClose(uniform.mean().eval(), s_uniform.mean()) 209 210 def testUniformVariance(self): 211 with self.test_session(): 212 a = 10.0 213 b = 100.0 214 uniform = uniform_lib.Uniform(low=a, high=b) 215 if not stats: 216 return 217 s_uniform = stats.uniform(loc=a, scale=b - a) 218 self.assertAllClose(uniform.variance().eval(), s_uniform.var()) 219 220 def testUniformStd(self): 221 with self.test_session(): 222 a = 10.0 223 b = 100.0 224 uniform = uniform_lib.Uniform(low=a, high=b) 225 if not stats: 226 return 227 s_uniform = stats.uniform(loc=a, scale=b - a) 228 self.assertAllClose(uniform.stddev().eval(), s_uniform.std()) 229 230 def testUniformNans(self): 231 with self.test_session(): 232 a = 10.0 233 b = [11.0, 100.0] 234 uniform = uniform_lib.Uniform(low=a, high=b) 235 236 no_nans = constant_op.constant(1.0) 237 nans = constant_op.constant(0.0) / constant_op.constant(0.0) 238 self.assertTrue(math_ops.is_nan(nans).eval()) 239 with_nans = array_ops.stack([no_nans, nans]) 240 241 pdf = uniform.prob(with_nans) 242 243 is_nan = math_ops.is_nan(pdf).eval() 244 self.assertFalse(is_nan[0]) 245 self.assertTrue(is_nan[1]) 246 247 def testUniformSamplePdf(self): 248 with self.test_session(): 249 a = 10.0 250 b = [11.0, 100.0] 251 uniform = uniform_lib.Uniform(a, b) 252 self.assertTrue( 253 math_ops.reduce_all(uniform.prob(uniform.sample(10)) > 0).eval()) 254 255 def testUniformBroadcasting(self): 256 with self.test_session(): 257 a = 10.0 258 b = [11.0, 20.0] 259 uniform = uniform_lib.Uniform(a, b) 260 261 pdf = uniform.prob([[10.5, 11.5], [9.0, 19.0], [10.5, 21.0]]) 262 expected_pdf = np.array([[1.0, 0.1], [0.0, 0.1], [1.0, 0.0]]) 263 self.assertAllClose(expected_pdf, pdf.eval()) 264 265 def testUniformSampleWithShape(self): 266 with self.test_session(): 267 a = 10.0 268 b = [11.0, 20.0] 269 uniform = uniform_lib.Uniform(a, b) 270 271 pdf = uniform.prob(uniform.sample((2, 3))) 272 # pylint: disable=bad-continuation 273 expected_pdf = [ 274 [[1.0, 0.1], [1.0, 0.1], [1.0, 0.1]], 275 [[1.0, 0.1], [1.0, 0.1], [1.0, 0.1]], 276 ] 277 # pylint: enable=bad-continuation 278 self.assertAllClose(expected_pdf, pdf.eval()) 279 280 pdf = uniform.prob(uniform.sample()) 281 expected_pdf = [1.0, 0.1] 282 self.assertAllClose(expected_pdf, pdf.eval()) 283 284 285 if __name__ == "__main__": 286 test.main() 287
