Make tf.random.get_seed() doesn't depend on op count, but the number …

…of calls to it.

It was difficult to change internal op counts because there are many tests that depend on the specific seed that's currently based on op counts.  With this change, get_seed() doesn't depend on the internal op count but the number of calls to get_seed()

PiperOrigin-RevId: 294523232
Change-Id: I4c0b7ea69adef8aceafcfa4754c30e9ea7a8d1e9

tf_slim/layers/optimizers_test.py

@@ -191,7 +191,7 @@ def testGradientNoise(self):
       session.run(train, feed_dict={x: 5})
       var_value, global_step_value = session.run([var, global_step])
       # Due to randomness the following number may change if graph is different.
-      self.assertAlmostEqual(var_value, 9.86912, 4)
+      self.assertAlmostEqual(var_value, 9.801016, 4)
       self.assertEqual(global_step_value, 1)
 
   @disable_resource_variables
@@ -209,7 +209,7 @@ def testGradientNoiseWithClipping(self):
       variables.global_variables_initializer().run()
       session.run(train, feed_dict={x: 5})
       var_value, global_step_value = session.run([var, global_step])
-      self.assertAlmostEqual(var_value, 9.86912, 4)
+      self.assertAlmostEqual(var_value, 9.801016, 4)
       self.assertEqual(global_step_value, 1)
 
   def testGradientClip(self):

tf_slim/learning_test.py

@@ -885,7 +885,7 @@ def testTrainingSubsetsOfVariablesOnlyUpdatesThoseVariables(self):
 
         # Update only biases.
         loss = sess.run(train_biases)
-        self.assertGreater(loss, .5)
+        self.assertGreater(loss, .45)
         new_weights, new_biases = sess.run([weights, biases])
 
         # Check that the biases have been updated, but weights have not.