How to use test_session method in Airtest

Best Python code snippet using Airtest

ops_test.py

Source:ops_test.py

...24from inception.slim import variables25class ConvTest(tf.test.TestCase):26    def testCreateConv(self):27        height, width = 3, 328        with self.test_session():29            images = tf.random_uniform((5, height, width, 3), seed=1)30            output = ops.conv2d(images, 32, [3, 3])31            self.assertEquals(output.op.name, 'Conv/Relu')32            self.assertListEqual(33                output.get_shape().as_list(), [34                    5, height, width, 32])35    def testCreateSquareConv(self):36        height, width = 3, 337        with self.test_session():38            images = tf.random_uniform((5, height, width, 3), seed=1)39            output = ops.conv2d(images, 32, 3)40            self.assertEquals(output.op.name, 'Conv/Relu')41            self.assertListEqual(42                output.get_shape().as_list(), [43                    5, height, width, 32])44    def testCreateConvWithTensorShape(self):45        height, width = 3, 346        with self.test_session():47            images = tf.random_uniform((5, height, width, 3), seed=1)48            output = ops.conv2d(images, 32, images.get_shape()[1:3])49            self.assertEquals(output.op.name, 'Conv/Relu')50            self.assertListEqual(51                output.get_shape().as_list(), [52                    5, height, width, 32])53    def testCreateFullyConv(self):54        height, width = 6, 655        with self.test_session():56            images = tf.random_uniform((5, height, width, 32), seed=1)57            output = ops.conv2d(58                images, 64, images.get_shape()[59                    1:3], padding='VALID')60            self.assertEquals(output.op.name, 'Conv/Relu')61            self.assertListEqual(output.get_shape().as_list(), [5, 1, 1, 64])62    def testCreateVerticalConv(self):63        height, width = 3, 364        with self.test_session():65            images = tf.random_uniform((5, height, width, 3), seed=1)66            output = ops.conv2d(images, 32, [3, 1])67            self.assertEquals(output.op.name, 'Conv/Relu')68            self.assertListEqual(output.get_shape().as_list(),69                                 [5, height, width, 32])70    def testCreateHorizontalConv(self):71        height, width = 3, 372        with self.test_session():73            images = tf.random_uniform((5, height, width, 3), seed=1)74            output = ops.conv2d(images, 32, [1, 3])75            self.assertEquals(output.op.name, 'Conv/Relu')76            self.assertListEqual(output.get_shape().as_list(),77                                 [5, height, width, 32])78    def testCreateConvWithStride(self):79        height, width = 6, 680        with self.test_session():81            images = tf.random_uniform((5, height, width, 3), seed=1)82            output = ops.conv2d(images, 32, [3, 3], stride=2)83            self.assertEquals(output.op.name, 'Conv/Relu')84            self.assertListEqual(output.get_shape().as_list(),85                                 [5, height / 2, width / 2, 32])86    def testCreateConvCreatesWeightsAndBiasesVars(self):87        height, width = 3, 388        images = tf.random_uniform((5, height, width, 3), seed=1)89        with self.test_session():90            self.assertFalse(variables.get_variables('conv1/weights'))91            self.assertFalse(variables.get_variables('conv1/biases'))92            ops.conv2d(images, 32, [3, 3], scope='conv1')93            self.assertTrue(variables.get_variables('conv1/weights'))94            self.assertTrue(variables.get_variables('conv1/biases'))95    def testCreateConvWithScope(self):96        height, width = 3, 397        with self.test_session():98            images = tf.random_uniform((5, height, width, 3), seed=1)99            output = ops.conv2d(images, 32, [3, 3], scope='conv1')100            self.assertEquals(output.op.name, 'conv1/Relu')101    def testCreateConvWithoutActivation(self):102        height, width = 3, 3103        with self.test_session():104            images = tf.random_uniform((5, height, width, 3), seed=1)105            output = ops.conv2d(images, 32, [3, 3], activation=None)106            self.assertEquals(output.op.name, 'Conv/BiasAdd')107    def testCreateConvValid(self):108        height, width = 3, 3109        with self.test_session():110            images = tf.random_uniform((5, height, width, 3), seed=1)111            output = ops.conv2d(images, 32, [3, 3], padding='VALID')112            self.assertListEqual(output.get_shape().as_list(), [5, 1, 1, 32])113    def testCreateConvWithWD(self):114        height, width = 3, 3115        with self.test_session() as sess:116            images = tf.random_uniform((5, height, width, 3), seed=1)117            ops.conv2d(images, 32, [3, 3], weight_decay=0.01)118            wd = tf.get_collection(tf.GraphKeys.REGULARIZATION_LOSSES)[0]119            self.assertEquals(wd.op.name,120                              'Conv/weights/Regularizer/L2Regularizer/value')121            sess.run(tf.initialize_all_variables())122            self.assertTrue(sess.run(wd) <= 0.01)123    def testCreateConvWithoutWD(self):124        height, width = 3, 3125        with self.test_session():126            images = tf.random_uniform((5, height, width, 3), seed=1)127            ops.conv2d(images, 32, [3, 3], weight_decay=0)128            self.assertEquals(129                tf.get_collection(tf.GraphKeys.REGULARIZATION_LOSSES), [])130    def testReuseVars(self):131        height, width = 3, 3132        with self.test_session():133            images = tf.random_uniform((5, height, width, 3), seed=1)134            ops.conv2d(images, 32, [3, 3], scope='conv1')135            self.assertEquals(len(variables.get_variables()), 2)136            ops.conv2d(images, 32, [3, 3], scope='conv1', reuse=True)137            self.assertEquals(len(variables.get_variables()), 2)138    def testNonReuseVars(self):139        height, width = 3, 3140        with self.test_session():141            images = tf.random_uniform((5, height, width, 3), seed=1)142            ops.conv2d(images, 32, [3, 3])143            self.assertEquals(len(variables.get_variables()), 2)144            ops.conv2d(images, 32, [3, 3])145            self.assertEquals(len(variables.get_variables()), 4)146    def testReuseConvWithWD(self):147        height, width = 3, 3148        with self.test_session():149            images = tf.random_uniform((5, height, width, 3), seed=1)150            ops.conv2d(images, 32, [3, 3], weight_decay=0.01, scope='conv1')151            self.assertEquals(len(variables.get_variables()), 2)152            self.assertEquals(153                len(tf.get_collection(tf.GraphKeys.REGULARIZATION_LOSSES)), 1)154            ops.conv2d(images, 32, [3, 3], weight_decay=0.01, scope='conv1',155                       reuse=True)156            self.assertEquals(len(variables.get_variables()), 2)157            self.assertEquals(158                len(tf.get_collection(tf.GraphKeys.REGULARIZATION_LOSSES)), 1)159    def testConvWithBatchNorm(self):160        height, width = 3, 3161        with self.test_session():162            images = tf.random_uniform((5, height, width, 32), seed=1)163            with scopes.arg_scope([ops.conv2d], batch_norm_params={'decay': 0.9}):164                net = ops.conv2d(images, 32, [3, 3])165                net = ops.conv2d(net, 32, [3, 3])166            self.assertEquals(len(variables.get_variables()), 8)167            self.assertEquals(168                len(variables.get_variables('Conv/BatchNorm')), 3)169            self.assertEquals(170                len(variables.get_variables('Conv_1/BatchNorm')), 3)171    def testReuseConvWithBatchNorm(self):172        height, width = 3, 3173        with self.test_session():174            images = tf.random_uniform((5, height, width, 32), seed=1)175            with scopes.arg_scope([ops.conv2d], batch_norm_params={'decay': 0.9}):176                net = ops.conv2d(images, 32, [3, 3], scope='Conv')177                net = ops.conv2d(net, 32, [3, 3], scope='Conv', reuse=True)178            self.assertEquals(len(variables.get_variables()), 4)179            self.assertEquals(180                len(variables.get_variables('Conv/BatchNorm')), 3)181            self.assertEquals(182                len(variables.get_variables('Conv_1/BatchNorm')), 0)183class FCTest(tf.test.TestCase):184    def testCreateFC(self):185        height, width = 3, 3186        with self.test_session():187            inputs = tf.random_uniform((5, height * width * 3), seed=1)188            output = ops.fc(inputs, 32)189            self.assertEquals(output.op.name, 'FC/Relu')190            self.assertListEqual(output.get_shape().as_list(), [5, 32])191    def testCreateFCWithScope(self):192        height, width = 3, 3193        with self.test_session():194            inputs = tf.random_uniform((5, height * width * 3), seed=1)195            output = ops.fc(inputs, 32, scope='fc1')196            self.assertEquals(output.op.name, 'fc1/Relu')197    def testCreateFcCreatesWeightsAndBiasesVars(self):198        height, width = 3, 3199        inputs = tf.random_uniform((5, height * width * 3), seed=1)200        with self.test_session():201            self.assertFalse(variables.get_variables('fc1/weights'))202            self.assertFalse(variables.get_variables('fc1/biases'))203            ops.fc(inputs, 32, scope='fc1')204            self.assertTrue(variables.get_variables('fc1/weights'))205            self.assertTrue(variables.get_variables('fc1/biases'))206    def testReuseVars(self):207        height, width = 3, 3208        inputs = tf.random_uniform((5, height * width * 3), seed=1)209        with self.test_session():210            ops.fc(inputs, 32, scope='fc1')211            self.assertEquals(len(variables.get_variables('fc1')), 2)212            ops.fc(inputs, 32, scope='fc1', reuse=True)213            self.assertEquals(len(variables.get_variables('fc1')), 2)214    def testNonReuseVars(self):215        height, width = 3, 3216        inputs = tf.random_uniform((5, height * width * 3), seed=1)217        with self.test_session():218            ops.fc(inputs, 32)219            self.assertEquals(len(variables.get_variables('FC')), 2)220            ops.fc(inputs, 32)221            self.assertEquals(len(variables.get_variables('FC')), 4)222    def testCreateFCWithoutActivation(self):223        height, width = 3, 3224        with self.test_session():225            inputs = tf.random_uniform((5, height * width * 3), seed=1)226            output = ops.fc(inputs, 32, activation=None)227            self.assertEquals(output.op.name, 'FC/xw_plus_b')228    def testCreateFCWithWD(self):229        height, width = 3, 3230        with self.test_session() as sess:231            inputs = tf.random_uniform((5, height * width * 3), seed=1)232            ops.fc(inputs, 32, weight_decay=0.01)233            wd = tf.get_collection(tf.GraphKeys.REGULARIZATION_LOSSES)[0]234            self.assertEquals(wd.op.name,235                              'FC/weights/Regularizer/L2Regularizer/value')236            sess.run(tf.initialize_all_variables())237            self.assertTrue(sess.run(wd) <= 0.01)238    def testCreateFCWithoutWD(self):239        height, width = 3, 3240        with self.test_session():241            inputs = tf.random_uniform((5, height * width * 3), seed=1)242            ops.fc(inputs, 32, weight_decay=0)243            self.assertEquals(244                tf.get_collection(tf.GraphKeys.REGULARIZATION_LOSSES), [])245    def testReuseFCWithWD(self):246        height, width = 3, 3247        with self.test_session():248            inputs = tf.random_uniform((5, height * width * 3), seed=1)249            ops.fc(inputs, 32, weight_decay=0.01, scope='fc')250            self.assertEquals(len(variables.get_variables()), 2)251            self.assertEquals(252                len(tf.get_collection(tf.GraphKeys.REGULARIZATION_LOSSES)), 1)253            ops.fc(inputs, 32, weight_decay=0.01, scope='fc', reuse=True)254            self.assertEquals(len(variables.get_variables()), 2)255            self.assertEquals(256                len(tf.get_collection(tf.GraphKeys.REGULARIZATION_LOSSES)), 1)257    def testFCWithBatchNorm(self):258        height, width = 3, 3259        with self.test_session():260            images = tf.random_uniform((5, height * width * 3), seed=1)261            with scopes.arg_scope([ops.fc], batch_norm_params={}):262                net = ops.fc(images, 27)263                net = ops.fc(net, 27)264            self.assertEquals(len(variables.get_variables()), 8)265            self.assertEquals(len(variables.get_variables('FC/BatchNorm')), 3)266            self.assertEquals(267                len(variables.get_variables('FC_1/BatchNorm')), 3)268    def testReuseFCWithBatchNorm(self):269        height, width = 3, 3270        with self.test_session():271            images = tf.random_uniform((5, height * width * 3), seed=1)272            with scopes.arg_scope([ops.fc], batch_norm_params={'decay': 0.9}):273                net = ops.fc(images, 27, scope='fc1')274                net = ops.fc(net, 27, scope='fc1', reuse=True)275            self.assertEquals(len(variables.get_variables()), 4)276            self.assertEquals(len(variables.get_variables('fc1/BatchNorm')), 3)277class MaxPoolTest(tf.test.TestCase):278    def testCreateMaxPool(self):279        height, width = 3, 3280        with self.test_session():281            images = tf.random_uniform((5, height, width, 3), seed=1)282            output = ops.max_pool(images, [3, 3])283            self.assertEquals(output.op.name, 'MaxPool/MaxPool')284            self.assertListEqual(output.get_shape().as_list(), [5, 1, 1, 3])285    def testCreateSquareMaxPool(self):286        height, width = 3, 3287        with self.test_session():288            images = tf.random_uniform((5, height, width, 3), seed=1)289            output = ops.max_pool(images, 3)290            self.assertEquals(output.op.name, 'MaxPool/MaxPool')291            self.assertListEqual(output.get_shape().as_list(), [5, 1, 1, 3])292    def testCreateMaxPoolWithScope(self):293        height, width = 3, 3294        with self.test_session():295            images = tf.random_uniform((5, height, width, 3), seed=1)296            output = ops.max_pool(images, [3, 3], scope='pool1')297            self.assertEquals(output.op.name, 'pool1/MaxPool')298    def testCreateMaxPoolSAME(self):299        height, width = 3, 3300        with self.test_session():301            images = tf.random_uniform((5, height, width, 3), seed=1)302            output = ops.max_pool(images, [3, 3], padding='SAME')303            self.assertListEqual(output.get_shape().as_list(), [5, 2, 2, 3])304    def testCreateMaxPoolStrideSAME(self):305        height, width = 3, 3306        with self.test_session():307            images = tf.random_uniform((5, height, width, 3), seed=1)308            output = ops.max_pool(images, [3, 3], stride=1, padding='SAME')309            self.assertListEqual(310                output.get_shape().as_list(), [311                    5, height, width, 3])312    def testGlobalMaxPool(self):313        height, width = 3, 3314        with self.test_session():315            images = tf.random_uniform((5, height, width, 3), seed=1)316            output = ops.max_pool(images, images.get_shape()[1:3], stride=1)317            self.assertListEqual(output.get_shape().as_list(), [5, 1, 1, 3])318class AvgPoolTest(tf.test.TestCase):319    def testCreateAvgPool(self):320        height, width = 3, 3321        with self.test_session():322            images = tf.random_uniform((5, height, width, 3), seed=1)323            output = ops.avg_pool(images, [3, 3])324            self.assertEquals(output.op.name, 'AvgPool/AvgPool')325            self.assertListEqual(output.get_shape().as_list(), [5, 1, 1, 3])326    def testCreateSquareAvgPool(self):327        height, width = 3, 3328        with self.test_session():329            images = tf.random_uniform((5, height, width, 3), seed=1)330            output = ops.avg_pool(images, 3)331            self.assertEquals(output.op.name, 'AvgPool/AvgPool')332            self.assertListEqual(output.get_shape().as_list(), [5, 1, 1, 3])333    def testCreateAvgPoolWithScope(self):334        height, width = 3, 3335        with self.test_session():336            images = tf.random_uniform((5, height, width, 3), seed=1)337            output = ops.avg_pool(images, [3, 3], scope='pool1')338            self.assertEquals(output.op.name, 'pool1/AvgPool')339    def testCreateAvgPoolSAME(self):340        height, width = 3, 3341        with self.test_session():342            images = tf.random_uniform((5, height, width, 3), seed=1)343            output = ops.avg_pool(images, [3, 3], padding='SAME')344            self.assertListEqual(output.get_shape().as_list(), [5, 2, 2, 3])345    def testCreateAvgPoolStrideSAME(self):346        height, width = 3, 3347        with self.test_session():348            images = tf.random_uniform((5, height, width, 3), seed=1)349            output = ops.avg_pool(images, [3, 3], stride=1, padding='SAME')350            self.assertListEqual(351                output.get_shape().as_list(), [352                    5, height, width, 3])353    def testGlobalAvgPool(self):354        height, width = 3, 3355        with self.test_session():356            images = tf.random_uniform((5, height, width, 3), seed=1)357            output = ops.avg_pool(images, images.get_shape()[1:3], stride=1)358            self.assertListEqual(output.get_shape().as_list(), [5, 1, 1, 3])359class OneHotEncodingTest(tf.test.TestCase):360    def testOneHotEncodingCreate(self):361        with self.test_session():362            labels = tf.constant([0, 1, 2])363            output = ops.one_hot_encoding(labels, num_classes=3)364            self.assertEquals(output.op.name, 'OneHotEncoding/SparseToDense')365            self.assertListEqual(output.get_shape().as_list(), [3, 3])366    def testOneHotEncoding(self):367        with self.test_session():368            labels = tf.constant([0, 1, 2])369            one_hot_labels = tf.constant([[1, 0, 0],370                                          [0, 1, 0],371                                          [0, 0, 1]])372            output = ops.one_hot_encoding(labels, num_classes=3)373            self.assertAllClose(output.eval(), one_hot_labels.eval())374class DropoutTest(tf.test.TestCase):375    def testCreateDropout(self):376        height, width = 3, 3377        with self.test_session():378            images = tf.random_uniform((5, height, width, 3), seed=1)379            output = ops.dropout(images)380            self.assertEquals(output.op.name, 'Dropout/dropout/mul_1')381            output.get_shape().assert_is_compatible_with(images.get_shape())382    def testCreateDropoutNoTraining(self):383        height, width = 3, 3384        with self.test_session():385            images = tf.random_uniform(386                (5, height, width, 3), seed=1, name='images')387            output = ops.dropout(images, is_training=False)388            self.assertEquals(output, images)389class FlattenTest(tf.test.TestCase):390    def testFlatten4D(self):391        height, width = 3, 3392        with self.test_session():393            images = tf.random_uniform(394                (5, height, width, 3), seed=1, name='images')395            output = ops.flatten(images)396            self.assertEquals(output.get_shape().num_elements(),397                              images.get_shape().num_elements())398            self.assertEqual(output.get_shape()[0], images.get_shape()[0])399    def testFlatten3D(self):400        height, width = 3, 3401        with self.test_session():402            images = tf.random_uniform(403                (5, height, width), seed=1, name='images')404            output = ops.flatten(images)405            self.assertEquals(output.get_shape().num_elements(),406                              images.get_shape().num_elements())407            self.assertEqual(output.get_shape()[0], images.get_shape()[0])408    def testFlattenBatchSize(self):409        height, width = 3, 3410        with self.test_session() as sess:411            images = tf.random_uniform(412                (5, height, width, 3), seed=1, name='images')413            inputs = tf.placeholder(tf.int32, (None, height, width, 3))414            output = ops.flatten(inputs)415            self.assertEquals(output.get_shape().as_list(),416                              [None, height * width * 3])417            output = sess.run(output, {inputs: images.eval()})418            self.assertEquals(output.size,419                              images.get_shape().num_elements())420            self.assertEqual(output.shape[0], images.get_shape()[0])421class BatchNormTest(tf.test.TestCase):422    def testCreateOp(self):423        height, width = 3, 3424        with self.test_session():425            images = tf.random_uniform((5, height, width, 3), seed=1)426            output = ops.batch_norm(images)427            self.assertTrue(output.op.name.startswith('BatchNorm/batchnorm'))428            self.assertListEqual(429                output.get_shape().as_list(), [430                    5, height, width, 3])431    def testCreateVariables(self):432        height, width = 3, 3433        with self.test_session():434            images = tf.random_uniform((5, height, width, 3), seed=1)435            ops.batch_norm(images)436            beta = variables.get_variables_by_name('beta')[0]437            self.assertEquals(beta.op.name, 'BatchNorm/beta')438            gamma = variables.get_variables_by_name('gamma')439            self.assertEquals(gamma, [])440            moving_mean = tf.moving_average_variables()[0]441            moving_variance = tf.moving_average_variables()[1]442            self.assertEquals(moving_mean.op.name, 'BatchNorm/moving_mean')443            self.assertEquals(444                moving_variance.op.name,445                'BatchNorm/moving_variance')446    def testCreateVariablesWithScale(self):447        height, width = 3, 3448        with self.test_session():449            images = tf.random_uniform((5, height, width, 3), seed=1)450            ops.batch_norm(images, scale=True)451            beta = variables.get_variables_by_name('beta')[0]452            gamma = variables.get_variables_by_name('gamma')[0]453            self.assertEquals(beta.op.name, 'BatchNorm/beta')454            self.assertEquals(gamma.op.name, 'BatchNorm/gamma')455            moving_mean = tf.moving_average_variables()[0]456            moving_variance = tf.moving_average_variables()[1]457            self.assertEquals(moving_mean.op.name, 'BatchNorm/moving_mean')458            self.assertEquals(459                moving_variance.op.name,460                'BatchNorm/moving_variance')461    def testCreateVariablesWithoutCenterWithScale(self):462        height, width = 3, 3463        with self.test_session():464            images = tf.random_uniform((5, height, width, 3), seed=1)465            ops.batch_norm(images, center=False, scale=True)466            beta = variables.get_variables_by_name('beta')467            self.assertEquals(beta, [])468            gamma = variables.get_variables_by_name('gamma')[0]469            self.assertEquals(gamma.op.name, 'BatchNorm/gamma')470            moving_mean = tf.moving_average_variables()[0]471            moving_variance = tf.moving_average_variables()[1]472            self.assertEquals(moving_mean.op.name, 'BatchNorm/moving_mean')473            self.assertEquals(474                moving_variance.op.name,475                'BatchNorm/moving_variance')476    def testCreateVariablesWithoutCenterWithoutScale(self):477        height, width = 3, 3478        with self.test_session():479            images = tf.random_uniform((5, height, width, 3), seed=1)480            ops.batch_norm(images, center=False, scale=False)481            beta = variables.get_variables_by_name('beta')482            self.assertEquals(beta, [])483            gamma = variables.get_variables_by_name('gamma')484            self.assertEquals(gamma, [])485            moving_mean = tf.moving_average_variables()[0]486            moving_variance = tf.moving_average_variables()[1]487            self.assertEquals(moving_mean.op.name, 'BatchNorm/moving_mean')488            self.assertEquals(489                moving_variance.op.name,490                'BatchNorm/moving_variance')491    def testMovingAverageVariables(self):492        height, width = 3, 3493        with self.test_session():494            images = tf.random_uniform((5, height, width, 3), seed=1)495            ops.batch_norm(images, scale=True)496            moving_mean = tf.moving_average_variables()[0]497            moving_variance = tf.moving_average_variables()[1]498            self.assertEquals(moving_mean.op.name, 'BatchNorm/moving_mean')499            self.assertEquals(500                moving_variance.op.name,501                'BatchNorm/moving_variance')502    def testUpdateOps(self):503        height, width = 3, 3504        with self.test_session():505            images = tf.random_uniform((5, height, width, 3), seed=1)506            ops.batch_norm(images)507            update_ops = tf.get_collection(ops.UPDATE_OPS_COLLECTION)508            update_moving_mean = update_ops[0]509            update_moving_variance = update_ops[1]510            self.assertEquals(update_moving_mean.op.name,511                              'BatchNorm/AssignMovingAvg')512            self.assertEquals(update_moving_variance.op.name,513                              'BatchNorm/AssignMovingAvg_1')514    def testReuseVariables(self):515        height, width = 3, 3516        with self.test_session():517            images = tf.random_uniform((5, height, width, 3), seed=1)518            ops.batch_norm(images, scale=True, scope='bn')519            ops.batch_norm(images, scale=True, scope='bn', reuse=True)520            beta = variables.get_variables_by_name('beta')521            gamma = variables.get_variables_by_name('gamma')522            self.assertEquals(len(beta), 1)523            self.assertEquals(len(gamma), 1)524            moving_vars = tf.get_collection('moving_vars')525            self.assertEquals(len(moving_vars), 2)526    def testReuseUpdateOps(self):527        height, width = 3, 3528        with self.test_session():529            images = tf.random_uniform((5, height, width, 3), seed=1)530            ops.batch_norm(images, scope='bn')531            self.assertEquals(532                len(tf.get_collection(ops.UPDATE_OPS_COLLECTION)), 2)533            ops.batch_norm(images, scope='bn', reuse=True)534            self.assertEquals(535                len(tf.get_collection(ops.UPDATE_OPS_COLLECTION)), 4)536    def testCreateMovingVars(self):537        height, width = 3, 3538        with self.test_session():539            images = tf.random_uniform((5, height, width, 3), seed=1)540            _ = ops.batch_norm(images, moving_vars='moving_vars')541            moving_mean = tf.get_collection('moving_vars',542                                            'BatchNorm/moving_mean')543            self.assertEquals(len(moving_mean), 1)544            self.assertEquals(moving_mean[0].op.name, 'BatchNorm/moving_mean')545            moving_variance = tf.get_collection('moving_vars',546                                                'BatchNorm/moving_variance')547            self.assertEquals(len(moving_variance), 1)548            self.assertEquals(549                moving_variance[0].op.name,550                'BatchNorm/moving_variance')551    def testComputeMovingVars(self):552        height, width = 3, 3553        with self.test_session() as sess:554            image_shape = (10, height, width, 3)555            image_values = np.random.rand(*image_shape)556            expected_mean = np.mean(image_values, axis=(0, 1, 2))557            expected_var = np.var(image_values, axis=(0, 1, 2))558            images = tf.constant(559                image_values,560                shape=image_shape,561                dtype=tf.float32)562            output = ops.batch_norm(images, decay=0.1)563            update_ops = tf.get_collection(ops.UPDATE_OPS_COLLECTION)564            with tf.control_dependencies(update_ops):565                barrier = tf.no_op(name='gradient_barrier')566                output = control_flow_ops.with_dependencies([barrier], output)567            # Initialize all variables568            sess.run(tf.initialize_all_variables())569            moving_mean = variables.get_variables('BatchNorm/moving_mean')[0]570            moving_variance = variables.get_variables(571                'BatchNorm/moving_variance')[0]572            mean, variance = sess.run([moving_mean, moving_variance])573            # After initialization moving_mean == 0 and moving_variance == 1.574            self.assertAllClose(mean, [0] * 3)575            self.assertAllClose(variance, [1] * 3)576            for _ in range(10):577                sess.run([output])578            mean = moving_mean.eval()579            variance = moving_variance.eval()580            # After 10 updates with decay 0.1 moving_mean == expected_mean and581            # moving_variance == expected_var.582            self.assertAllClose(mean, expected_mean)583            self.assertAllClose(variance, expected_var)584    def testEvalMovingVars(self):585        height, width = 3, 3586        with self.test_session() as sess:587            image_shape = (10, height, width, 3)588            image_values = np.random.rand(*image_shape)589            expected_mean = np.mean(image_values, axis=(0, 1, 2))590            expected_var = np.var(image_values, axis=(0, 1, 2))591            images = tf.constant(592                image_values,593                shape=image_shape,594                dtype=tf.float32)595            output = ops.batch_norm(images, decay=0.1, is_training=False)596            update_ops = tf.get_collection(ops.UPDATE_OPS_COLLECTION)597            with tf.control_dependencies(update_ops):598                barrier = tf.no_op(name='gradient_barrier')599                output = control_flow_ops.with_dependencies([barrier], output)600            # Initialize all variables601            sess.run(tf.initialize_all_variables())602            moving_mean = variables.get_variables('BatchNorm/moving_mean')[0]603            moving_variance = variables.get_variables(604                'BatchNorm/moving_variance')[0]605            mean, variance = sess.run([moving_mean, moving_variance])606            # After initialization moving_mean == 0 and moving_variance == 1.607            self.assertAllClose(mean, [0] * 3)608            self.assertAllClose(variance, [1] * 3)609            # Simulate assigment from saver restore.610            init_assigns = [tf.assign(moving_mean, expected_mean),611                            tf.assign(moving_variance, expected_var)]612            sess.run(init_assigns)613            for _ in range(10):614                sess.run([output], {images: np.random.rand(*image_shape)})615            mean = moving_mean.eval()616            variance = moving_variance.eval()617            # Although we feed different images, the moving_mean and moving_variance618            # shouldn't change.619            self.assertAllClose(mean, expected_mean)620            self.assertAllClose(variance, expected_var)621    def testReuseVars(self):622        height, width = 3, 3623        with self.test_session() as sess:624            image_shape = (10, height, width, 3)625            image_values = np.random.rand(*image_shape)626            expected_mean = np.mean(image_values, axis=(0, 1, 2))627            expected_var = np.var(image_values, axis=(0, 1, 2))628            images = tf.constant(629                image_values,630                shape=image_shape,631                dtype=tf.float32)632            output = ops.batch_norm(images, decay=0.1, is_training=False)633            update_ops = tf.get_collection(ops.UPDATE_OPS_COLLECTION)634            with tf.control_dependencies(update_ops):635                barrier = tf.no_op(name='gradient_barrier')636                output = control_flow_ops.with_dependencies([barrier], output)637            # Initialize all variables...

variables_test.py

Source:variables_test.py

...20from inception.slim import scopes21from inception.slim import variables22class VariablesTest(tf.test.TestCase):23    def testCreateVariable(self):24        with self.test_session():25            with tf.variable_scope('A'):26                a = variables.variable('a', [5])27                self.assertEquals(a.op.name, 'A/a')28                self.assertListEqual(a.get_shape().as_list(), [5])29    def testGetVariables(self):30        with self.test_session():31            with tf.variable_scope('A'):32                a = variables.variable('a', [5])33            with tf.variable_scope('B'):34                b = variables.variable('a', [5])35            self.assertEquals([a, b], variables.get_variables())36            self.assertEquals([a], variables.get_variables('A'))37            self.assertEquals([b], variables.get_variables('B'))38    def testGetVariablesSuffix(self):39        with self.test_session():40            with tf.variable_scope('A'):41                a = variables.variable('a', [5])42            with tf.variable_scope('A'):43                b = variables.variable('b', [5])44            self.assertEquals([a], variables.get_variables(suffix='a'))45            self.assertEquals([b], variables.get_variables(suffix='b'))46    def testGetVariableWithSingleVar(self):47        with self.test_session():48            with tf.variable_scope('parent'):49                a = variables.variable('child', [5])50            self.assertEquals(a, variables.get_unique_variable('parent/child'))51    def testGetVariableWithDistractors(self):52        with self.test_session():53            with tf.variable_scope('parent'):54                a = variables.variable('child', [5])55                with tf.variable_scope('child'):56                    variables.variable('grandchild1', [7])57                    variables.variable('grandchild2', [9])58            self.assertEquals(a, variables.get_unique_variable('parent/child'))59    def testGetVariableThrowsExceptionWithNoMatch(self):60        var_name = 'cant_find_me'61        with self.test_session():62            with self.assertRaises(ValueError):63                variables.get_unique_variable(var_name)64    def testGetThrowsExceptionWithChildrenButNoMatch(self):65        var_name = 'parent/child'66        with self.test_session():67            with tf.variable_scope(var_name):68                variables.variable('grandchild1', [7])69                variables.variable('grandchild2', [9])70            with self.assertRaises(ValueError):71                variables.get_unique_variable(var_name)72    def testGetVariablesToRestore(self):73        with self.test_session():74            with tf.variable_scope('A'):75                a = variables.variable('a', [5])76            with tf.variable_scope('B'):77                b = variables.variable('a', [5])78            self.assertEquals([a, b], variables.get_variables_to_restore())79    def testNoneGetVariablesToRestore(self):80        with self.test_session():81            with tf.variable_scope('A'):82                a = variables.variable('a', [5], restore=False)83            with tf.variable_scope('B'):84                b = variables.variable('a', [5], restore=False)85            self.assertEquals([], variables.get_variables_to_restore())86            self.assertEquals([a, b], variables.get_variables())87    def testGetMixedVariablesToRestore(self):88        with self.test_session():89            with tf.variable_scope('A'):90                a = variables.variable('a', [5])91                b = variables.variable('b', [5], restore=False)92            with tf.variable_scope('B'):93                c = variables.variable('c', [5])94                d = variables.variable('d', [5], restore=False)95            self.assertEquals([a, b, c, d], variables.get_variables())96            self.assertEquals([a, c], variables.get_variables_to_restore())97    def testReuseVariable(self):98        with self.test_session():99            with tf.variable_scope('A'):100                a = variables.variable('a', [])101            with tf.variable_scope('A', reuse=True):102                b = variables.variable('a', [])103            self.assertEquals(a, b)104            self.assertListEqual([a], variables.get_variables())105    def testVariableWithDevice(self):106        with self.test_session():107            with tf.variable_scope('A'):108                a = variables.variable('a', [], device='cpu:0')109                b = variables.variable('b', [], device='cpu:1')110            self.assertDeviceEqual(a.device, 'cpu:0')111            self.assertDeviceEqual(b.device, 'cpu:1')112    def testVariableWithDeviceFromScope(self):113        with self.test_session():114            with tf.device('/cpu:0'):115                a = variables.variable('a', [])116                b = variables.variable('b', [], device='cpu:1')117            self.assertDeviceEqual(a.device, 'cpu:0')118            self.assertDeviceEqual(b.device, 'cpu:1')119    def testVariableWithDeviceFunction(self):120        class DevFn(object):121            def __init__(self):122                self.counter = -1123            def __call__(self, op):124                self.counter += 1125                return 'cpu:%d' % self.counter126        with self.test_session():127            with scopes.arg_scope([variables.variable], device=DevFn()):128                a = variables.variable('a', [])129                b = variables.variable('b', [])130                c = variables.variable('c', [], device='cpu:12')131                d = variables.variable('d', [])132                with tf.device('cpu:99'):133                    e_init = tf.constant(12)134                e = variables.variable('e', initializer=e_init)135            self.assertDeviceEqual(a.device, 'cpu:0')136            self.assertDeviceEqual(a.initial_value.device, 'cpu:0')137            self.assertDeviceEqual(b.device, 'cpu:1')138            self.assertDeviceEqual(b.initial_value.device, 'cpu:1')139            self.assertDeviceEqual(c.device, 'cpu:12')140            self.assertDeviceEqual(c.initial_value.device, 'cpu:12')141            self.assertDeviceEqual(d.device, 'cpu:2')142            self.assertDeviceEqual(d.initial_value.device, 'cpu:2')143            self.assertDeviceEqual(e.device, 'cpu:3')144            self.assertDeviceEqual(e.initial_value.device, 'cpu:99')145    def testVariableWithReplicaDeviceSetter(self):146        with self.test_session():147            with tf.device(tf.train.replica_device_setter(ps_tasks=2)):148                a = variables.variable('a', [])149                b = variables.variable('b', [])150                c = variables.variable('c', [], device='cpu:12')151                d = variables.variable('d', [])152                with tf.device('cpu:99'):153                    e_init = tf.constant(12)154                e = variables.variable('e', initializer=e_init)155            # The values below highlight how the replica_device_setter puts initial156            # values on the worker job, and how it merges explicit devices.157            self.assertDeviceEqual(a.device, '/job:ps/task:0/cpu:0')158            self.assertDeviceEqual(a.initial_value.device, '/job:worker/cpu:0')159            self.assertDeviceEqual(b.device, '/job:ps/task:1/cpu:0')160            self.assertDeviceEqual(b.initial_value.device, '/job:worker/cpu:0')161            self.assertDeviceEqual(c.device, '/job:ps/task:0/cpu:12')162            self.assertDeviceEqual(163                c.initial_value.device,164                '/job:worker/cpu:12')165            self.assertDeviceEqual(d.device, '/job:ps/task:1/cpu:0')166            self.assertDeviceEqual(d.initial_value.device, '/job:worker/cpu:0')167            self.assertDeviceEqual(e.device, '/job:ps/task:0/cpu:0')168            self.assertDeviceEqual(169                e.initial_value.device,170                '/job:worker/cpu:99')171    def testVariableWithVariableDeviceChooser(self):172        with tf.Graph().as_default():173            device_fn = variables.VariableDeviceChooser(174                num_parameter_servers=2)175            with scopes.arg_scope([variables.variable], device=device_fn):176                a = variables.variable('a', [])177                b = variables.variable('b', [])178                c = variables.variable('c', [], device='cpu:12')179                d = variables.variable('d', [])180                with tf.device('cpu:99'):181                    e_init = tf.constant(12)182                e = variables.variable('e', initializer=e_init)183            # The values below highlight how the VariableDeviceChooser puts initial184            # values on the same device as the variable job.185            self.assertDeviceEqual(a.device, '/job:ps/task:0/cpu:0')186            self.assertDeviceEqual(a.initial_value.device, a.device)187            self.assertDeviceEqual(b.device, '/job:ps/task:1/cpu:0')188            self.assertDeviceEqual(b.initial_value.device, b.device)189            self.assertDeviceEqual(c.device, '/cpu:12')190            self.assertDeviceEqual(c.initial_value.device, c.device)191            self.assertDeviceEqual(d.device, '/job:ps/task:0/cpu:0')192            self.assertDeviceEqual(d.initial_value.device, d.device)193            self.assertDeviceEqual(e.device, '/job:ps/task:1/cpu:0')194            self.assertDeviceEqual(e.initial_value.device, '/cpu:99')195    def testVariableGPUPlacement(self):196        with tf.Graph().as_default():197            device_fn = variables.VariableDeviceChooser(placement='gpu:0')198            with scopes.arg_scope([variables.variable], device=device_fn):199                a = variables.variable('a', [])200                b = variables.variable('b', [])201                c = variables.variable('c', [], device='cpu:12')202                d = variables.variable('d', [])203                with tf.device('cpu:99'):204                    e_init = tf.constant(12)205                e = variables.variable('e', initializer=e_init)206            # The values below highlight how the VariableDeviceChooser puts initial207            # values on the same device as the variable job.208            self.assertDeviceEqual(a.device, '/gpu:0')209            self.assertDeviceEqual(a.initial_value.device, a.device)210            self.assertDeviceEqual(b.device, '/gpu:0')211            self.assertDeviceEqual(b.initial_value.device, b.device)212            self.assertDeviceEqual(c.device, '/cpu:12')213            self.assertDeviceEqual(c.initial_value.device, c.device)214            self.assertDeviceEqual(d.device, '/gpu:0')215            self.assertDeviceEqual(d.initial_value.device, d.device)216            self.assertDeviceEqual(e.device, '/gpu:0')217            self.assertDeviceEqual(e.initial_value.device, '/cpu:99')218    def testVariableCollection(self):219        with self.test_session():220            a = variables.variable('a', [], collections='A')221            b = variables.variable('b', [], collections='B')222            self.assertEquals(a, tf.get_collection('A')[0])223            self.assertEquals(b, tf.get_collection('B')[0])224    def testVariableCollections(self):225        with self.test_session():226            a = variables.variable('a', [], collections=['A', 'C'])227            b = variables.variable('b', [], collections=['B', 'C'])228            self.assertEquals(a, tf.get_collection('A')[0])229            self.assertEquals(b, tf.get_collection('B')[0])230    def testVariableCollectionsWithArgScope(self):231        with self.test_session():232            with scopes.arg_scope([variables.variable], collections='A'):233                a = variables.variable('a', [])234                b = variables.variable('b', [])235            self.assertListEqual([a, b], tf.get_collection('A'))236    def testVariableCollectionsWithArgScopeNested(self):237        with self.test_session():238            with scopes.arg_scope([variables.variable], collections='A'):239                a = variables.variable('a', [])240                with scopes.arg_scope([variables.variable], collections='B'):241                    b = variables.variable('b', [])242            self.assertEquals(a, tf.get_collection('A')[0])243            self.assertEquals(b, tf.get_collection('B')[0])244    def testVariableCollectionsWithArgScopeNonNested(self):245        with self.test_session():246            with scopes.arg_scope([variables.variable], collections='A'):247                a = variables.variable('a', [])248            with scopes.arg_scope([variables.variable], collections='B'):249                b = variables.variable('b', [])250            variables.variable('c', [])251            self.assertListEqual([a], tf.get_collection('A'))252            self.assertListEqual([b], tf.get_collection('B'))253    def testVariableRestoreWithArgScopeNested(self):254        with self.test_session():255            with scopes.arg_scope([variables.variable], restore=True):256                a = variables.variable('a', [])257                with scopes.arg_scope([variables.variable],258                                      trainable=False,259                                      collections=['A', 'B']):260                    b = variables.variable('b', [])261                c = variables.variable('c', [])262            self.assertListEqual(263                [a, b, c], variables.get_variables_to_restore())264            self.assertListEqual([a, c], tf.trainable_variables())265            self.assertListEqual([b], tf.get_collection('A'))266            self.assertListEqual([b], tf.get_collection('B'))267class GetVariablesByNameTest(tf.test.TestCase):268    def testGetVariableGivenNameScoped(self):269        with self.test_session():270            with tf.variable_scope('A'):271                a = variables.variable('a', [5])272                b = variables.variable('b', [5])273                self.assertEquals([a], variables.get_variables_by_name('a'))274                self.assertEquals([b], variables.get_variables_by_name('b'))275    def testGetVariablesByNameReturnsByValueWithScope(self):276        with self.test_session():277            with tf.variable_scope('A'):278                a = variables.variable('a', [5])279                matched_variables = variables.get_variables_by_name('a')280                # If variables.get_variables_by_name returns the list by reference, the281                # following append should persist, and be returned, in subsequent calls282                # to variables.get_variables_by_name('a').283                matched_variables.append(4)284                matched_variables = variables.get_variables_by_name('a')285                self.assertEquals([a], matched_variables)286    def testGetVariablesByNameReturnsByValueWithoutScope(self):287        with self.test_session():288            a = variables.variable('a', [5])289            matched_variables = variables.get_variables_by_name('a')290            # If variables.get_variables_by_name returns the list by reference, the291            # following append should persist, and be returned, in subsequent calls292            # to variables.get_variables_by_name('a').293            matched_variables.append(4)294            matched_variables = variables.get_variables_by_name('a')295            self.assertEquals([a], matched_variables)296class GlobalStepTest(tf.test.TestCase):297    def testStable(self):298        with tf.Graph().as_default():299            gs = variables.global_step()300            gs2 = variables.global_step()301            self.assertTrue(gs is gs2)...

losses_test.py

Source:losses_test.py

...19import tensorflow as tf20from inception.slim import losses21class LossesTest(tf.test.TestCase):22    def testL1Loss(self):23        with self.test_session():24            shape = [5, 5, 5]25            num_elem = 5 * 5 * 526            weights = tf.constant(1.0, shape=shape)27            wd = 0.0128            loss = losses.l1_loss(weights, wd)29            self.assertEquals(loss.op.name, 'L1Loss/value')30            self.assertAlmostEqual(loss.eval(), num_elem * wd, 5)31    def testL2Loss(self):32        with self.test_session():33            shape = [5, 5, 5]34            num_elem = 5 * 5 * 535            weights = tf.constant(1.0, shape=shape)36            wd = 0.0137            loss = losses.l2_loss(weights, wd)38            self.assertEquals(loss.op.name, 'L2Loss/value')39            self.assertAlmostEqual(loss.eval(), num_elem * wd / 2, 5)40class RegularizersTest(tf.test.TestCase):41    def testL1Regularizer(self):42        with self.test_session():43            shape = [5, 5, 5]44            num_elem = 5 * 5 * 545            tensor = tf.constant(1.0, shape=shape)46            loss = losses.l1_regularizer()(tensor)47            self.assertEquals(loss.op.name, 'L1Regularizer/value')48            self.assertAlmostEqual(loss.eval(), num_elem, 5)49    def testL1RegularizerWithScope(self):50        with self.test_session():51            shape = [5, 5, 5]52            num_elem = 5 * 5 * 553            tensor = tf.constant(1.0, shape=shape)54            loss = losses.l1_regularizer(scope='L1')(tensor)55            self.assertEquals(loss.op.name, 'L1/value')56            self.assertAlmostEqual(loss.eval(), num_elem, 5)57    def testL1RegularizerWithWeight(self):58        with self.test_session():59            shape = [5, 5, 5]60            num_elem = 5 * 5 * 561            tensor = tf.constant(1.0, shape=shape)62            weight = 0.0163            loss = losses.l1_regularizer(weight)(tensor)64            self.assertEquals(loss.op.name, 'L1Regularizer/value')65            self.assertAlmostEqual(loss.eval(), num_elem * weight, 5)66    def testL2Regularizer(self):67        with self.test_session():68            shape = [5, 5, 5]69            num_elem = 5 * 5 * 570            tensor = tf.constant(1.0, shape=shape)71            loss = losses.l2_regularizer()(tensor)72            self.assertEquals(loss.op.name, 'L2Regularizer/value')73            self.assertAlmostEqual(loss.eval(), num_elem / 2, 5)74    def testL2RegularizerWithScope(self):75        with self.test_session():76            shape = [5, 5, 5]77            num_elem = 5 * 5 * 578            tensor = tf.constant(1.0, shape=shape)79            loss = losses.l2_regularizer(scope='L2')(tensor)80            self.assertEquals(loss.op.name, 'L2/value')81            self.assertAlmostEqual(loss.eval(), num_elem / 2, 5)82    def testL2RegularizerWithWeight(self):83        with self.test_session():84            shape = [5, 5, 5]85            num_elem = 5 * 5 * 586            tensor = tf.constant(1.0, shape=shape)87            weight = 0.0188            loss = losses.l2_regularizer(weight)(tensor)89            self.assertEquals(loss.op.name, 'L2Regularizer/value')90            self.assertAlmostEqual(loss.eval(), num_elem * weight / 2, 5)91    def testL1L2Regularizer(self):92        with self.test_session():93            shape = [5, 5, 5]94            num_elem = 5 * 5 * 595            tensor = tf.constant(1.0, shape=shape)96            loss = losses.l1_l2_regularizer()(tensor)97            self.assertEquals(loss.op.name, 'L1L2Regularizer/value')98            self.assertAlmostEqual(loss.eval(), num_elem + num_elem / 2, 5)99    def testL1L2RegularizerWithScope(self):100        with self.test_session():101            shape = [5, 5, 5]102            num_elem = 5 * 5 * 5103            tensor = tf.constant(1.0, shape=shape)104            loss = losses.l1_l2_regularizer(scope='L1L2')(tensor)105            self.assertEquals(loss.op.name, 'L1L2/value')106            self.assertAlmostEqual(loss.eval(), num_elem + num_elem / 2, 5)107    def testL1L2RegularizerWithWeights(self):108        with self.test_session():109            shape = [5, 5, 5]110            num_elem = 5 * 5 * 5111            tensor = tf.constant(1.0, shape=shape)112            weight_l1 = 0.01113            weight_l2 = 0.05114            loss = losses.l1_l2_regularizer(weight_l1, weight_l2)(tensor)115            self.assertEquals(loss.op.name, 'L1L2Regularizer/value')116            self.assertAlmostEqual(117                loss.eval(),118                num_elem *119                weight_l1 +120                num_elem *121                weight_l2 /122                2,123                5)124class CrossEntropyLossTest(tf.test.TestCase):125    def testCrossEntropyLossAllCorrect(self):126        with self.test_session():127            logits = tf.constant([[10.0, 0.0, 0.0],128                                  [0.0, 10.0, 0.0],129                                  [0.0, 0.0, 10.0]])130            labels = tf.constant([[1, 0, 0],131                                  [0, 1, 0],132                                  [0, 0, 1]])133            loss = losses.cross_entropy_loss(logits, labels)134            self.assertEquals(loss.op.name, 'CrossEntropyLoss/value')135            self.assertAlmostEqual(loss.eval(), 0.0, 3)136    def testCrossEntropyLossAllWrong(self):137        with self.test_session():138            logits = tf.constant([[10.0, 0.0, 0.0],139                                  [0.0, 10.0, 0.0],140                                  [0.0, 0.0, 10.0]])141            labels = tf.constant([[0, 0, 1],142                                  [1, 0, 0],143                                  [0, 1, 0]])144            loss = losses.cross_entropy_loss(logits, labels)145            self.assertEquals(loss.op.name, 'CrossEntropyLoss/value')146            self.assertAlmostEqual(loss.eval(), 10.0, 3)147    def testCrossEntropyLossAllWrongWithWeight(self):148        with self.test_session():149            logits = tf.constant([[10.0, 0.0, 0.0],150                                  [0.0, 10.0, 0.0],151                                  [0.0, 0.0, 10.0]])152            labels = tf.constant([[0, 0, 1],153                                  [1, 0, 0],154                                  [0, 1, 0]])155            loss = losses.cross_entropy_loss(logits, labels, weight=0.5)156            self.assertEquals(loss.op.name, 'CrossEntropyLoss/value')157            self.assertAlmostEqual(loss.eval(), 5.0, 3)158if __name__ == '__main__':...

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.