for i in range(1, batch_size):
      col = 1
      if np.random.rand() < 0.25:
        # w.p. 25%, drop out the second entry
        expected_str[i][col] = b"default"