c773824f4f22a14375a2451264d1b5708addd0f9,rllib/models/tests/test_distributions.py,TestDistributions,_stability_test,#TestDistributions#Any#Any#Any#Any#Any#,27

Before Change

        ]
        inputs = np.zeros(shape=network_output_shape, dtype=np.float32)
        for batch_item in range(network_output_shape[0]):
            for num in range(len(inputs[batch_item])):
                inputs[batch_item][num] = np.random.choice(extreme_values)
        dist = distribution_cls(inputs, {})
        for _ in range(100):
            sample = dist.sample()
            if fw != "tf":

After Change

        ]
        inputs = np.zeros(shape=network_output_shape, dtype=np.float32)
        for batch_item in range(network_output_shape[0]):
            for num in range(len(inputs[batch_item]) // 2):
                inputs[batch_item][num] = np.random.choice(extreme_values)
            else:
                // For Gaussians, the second half of the vector is
                // log standard deviations, and should therefore be
                // the log of a positive number >= 1.
                inputs[batch_item][num] = np.log(
                    max(1, np.random.choice((extreme_values))))

        dist = distribution_cls(inputs, {})
        for _ in range(100):
            sample = dist.sample()
            if fw != "tf":

Instances