Unexpected behavior from `tf.math.greaterEqual`

System information

• Have I written custom code (as opposed to using a stock example script provided in TensorFlow):
  Custom code
• OS Platform and Distribution (e.g., Linux Ubuntu 16.04):
  macOS BigSur 11.1
• TensorFlow installed from (source or binary):
  TensorFlow Java latest GitHub, based on karl/tensor-ttype

I have the following code snippet.

       Operand<TFloat32> r = tf.random.randomUniform(tf.constant(Shape.of(3,2)), TFloat32.class,   
                                                         RandomUniform.seed(1001L));
      Operand<TFloat32> rate1 = tf.constant(0.5f);
      Operand<TBool> mask = tf.math.greaterEqual(r, rate1);

The generated random sequence (shape 3,2) is:

0,0). 0.028465
0,1). 0.936395
1.0). 0.035310
1,1). 0.230565
2,0). 0.269415
2,1). 0.145439

Describe the current behavior
When I run this code, I get mask values of:

0,0). false
0,1). true
1,0). false
1,1). false
2,0). false
2,1). true

Describe the expected behavior
I expected mask values to be:

0,0). false
0,1). true
1,0). false
1,1). false
2,0). false
2,1). false

The last random number, 0.145439, is less than 0.5.

Code to reproduce the issue
Provide a reproducible test case that is the bare minimum necessary to generate the problem.

try (TestSession session = TestSession.createTestSession(TestSession.Mode.GRAPH)) {
      Ops tf = session.getTF();
       Operand<TFloat32> random = tf.random.randomUniform(tf.constant(Shape.of(3,2), TFloat32.class,   
                                                         RandomUniform.seed(1001L));
      Operand<TFloat32> rate = tf.constant(0.5f);
      Operand<TBool> mask = tf.math.greaterEqual(r, rate);

      System.out.println("******** MASK ***************");
      System.out.println("random:");
      session.print(random);
      System.out.println("rate1:");
      session.print(rate1);
      System.out.println("mask:");
      session.print(mask);
      System.out.println("******** MASK ***************");
}

If I do this code in Python it works as expected:

import tensorflow as tf
a = [
    [0.028465 ,  0.936395],
    [0.035310 , 0.230565],
    [0.269415 , 0.145439]
]
print(a);
rate = 0.5
b = tf.math.greater_equal(a, rate);
print(b)

Output:

[[0.028465, 0.936395], [0.03531, 0.230565], [0.269415, 0.145439]]
tf.Tensor(
[[False  True]
 [False False]
 [False False]], shape=(3, 2), dtype=bool)

Commencez par l’extrait Java et la TestSession en mode GRAPH, puis comparez le tenseur généré et le résultat de tf.math.greaterEqual avec l’exemple Python. Reproduisez l’incohérence signalée sur le dernier élément et déterminez si elle est produite par la liaison Java ou par la configuration du test ; le travail est terminé lorsque le comportement est expliqué et qu’un test de régression ou une correction a été vérifié.