Python tensorflow.int16() Examples

def diet_adam_optimizer_params():
  """Default hyperparameters for a DietAdamOptimizer.

  Returns:
    a hyperparameters object.
  """
  return tf.contrib.training.HParams(
      quantize=True,  # use 16-bit fixed-point
      quantization_scale=10.0 / tf.int16.max,
      optimizer="DietAdam",
      learning_rate=1.0,
      learning_rate_warmup_steps=2000,
      learning_rate_decay_scheme="noam",  # "noam" or "none"
      epsilon=1e-10,
      beta1=0.0,  # we can save memory if beta1=0
      beta2=0.98,
      factored_second_moment_accumulator=True,  # this saves memory
  ) 

def test__dtype_to_bytes():
    np_tf_dt = [
        (np.uint8, tf.uint8, b"uint8"),
        (np.uint16, tf.uint16, b"uint16"),
        (np.uint32, tf.uint32, b"uint32"),
        (np.uint64, tf.uint64, b"uint64"),
        (np.int8, tf.int8, b"int8"),
        (np.int16, tf.int16, b"int16"),
        (np.int32, tf.int32, b"int32"),
        (np.int64, tf.int64, b"int64"),
        (np.float16, tf.float16, b"float16"),
        (np.float32, tf.float32, b"float32"),
        (np.float64, tf.float64, b"float64"),
    ]

    for npd, tfd, dt in np_tf_dt:
        npd = np.dtype(npd)
        assert tfrecord._dtype_to_bytes(npd) == dt
        assert tfrecord._dtype_to_bytes(tfd) == dt

    assert tfrecord._dtype_to_bytes("float32") == b"float32"
    assert tfrecord._dtype_to_bytes("foobar") == b"foobar" 

def decode_pred(serialized_example):
	"""Parses prediction data from the given `serialized_example`."""

	features = tf.parse_single_example(
					serialized_example,
					features={
						'T1':tf.FixedLenFeature([],tf.string),
						'T2':tf.FixedLenFeature([], tf.string)
					})

	patch_shape = [conf.patch_size, conf.patch_size, conf.patch_size]

	# Convert from a scalar string tensor
	image_T1 = tf.decode_raw(features['T1'], tf.int16)
	image_T1 = tf.reshape(image_T1, patch_shape)
	image_T2 = tf.decode_raw(features['T2'], tf.int16)
	image_T2 = tf.reshape(image_T2, patch_shape)

	# Convert dtype.
	image_T1 = tf.cast(image_T1, tf.float32)
	image_T2 = tf.cast(image_T2, tf.float32)
	label = tf.zeros(image_T1.shape) # pseudo label

	return image_T1, image_T2, label 

def reduce_mean_support_empty(input, keepdims=False):
    return tf.cond(tf.size(input) > 0, lambda: tf.reduce_mean(input, keepdims=keepdims), lambda: tf.zeros_like(input))


# def bit_tensor_list(input):
#     assert input.dtype in [tf.uint8, tf.uint16, tf.uint32, tf.uint64], 'unsupported data type, must be uint*'
#     num_bits = 0
#     if input.dtype == tf.int8:
#         num_bits = 8
#     elif input.dtype == tf.int16:
#         num_bits = 16
#     elif input.dtype == tf.uint32:
#         num_bits = 32
#     elif input.dtype == tf.uint64:
#         num_bits = 64
#     bit_tensors = []
#     for i in range(num_bits):
#         current_bit = 1 << i
#         current_bit_tensor = tf.bitwise.bitwise_and(input, current_bit) == 1
#         bit_tensors.append(current_bit_tensor)
#     print(bit_tensors)
#     return bit_tensors 

def parse_a_line_b(value, base_num, signal_num):
    vec = tf.decode_raw(value, tf.int8)

    bases = tf.cast(tf.reshape(tf.strided_slice(vec, [0], [base_num]), [base_num]), dtype=tf.int32)
    means = tf.bitcast(
        tf.reshape(tf.strided_slice(vec, [base_num], [base_num + base_num * 4]), [base_num, 4]),
        type=tf.float32)
    stds = tf.bitcast(
        tf.reshape(tf.strided_slice(vec, [base_num * 5], [base_num * 5 + base_num * 4]), [base_num, 4]),
        type=tf.float32)
    sanum = tf.cast(tf.bitcast(
        tf.reshape(tf.strided_slice(vec, [base_num * 9], [base_num * 9 + base_num * 2]), [base_num, 2]),
        type=tf.int16), dtype=tf.int32)
    signals = tf.bitcast(
        tf.reshape(tf.strided_slice(vec, [base_num * 11], [base_num * 11 + 4 * signal_num]),
                   [signal_num, 4]), type=tf.float32)
    labels = tf.cast(
        tf.reshape(tf.strided_slice(vec, [base_num * 11 + signal_num * 4], [base_num * 11 + signal_num * 4 + 1]),
                   [1]),
        dtype=tf.int32)

    return bases, means, stds, sanum, signals, labels 

def _convert_string_dtype(dtype):
    if dtype == 'float16':
        return tf.float16
    if dtype == 'float32':
        return tf.float32
    elif dtype == 'float64':
        return tf.float64
    elif dtype == 'int16':
        return tf.int16
    elif dtype == 'int32':
        return tf.int32
    elif dtype == 'int64':
        return tf.int64
    elif dtype == 'uint8':
        return tf.int8
    elif dtype == 'uint16':
        return tf.uint16
    else:
        raise ValueError('Unsupported dtype:', dtype) 

def set_dtype(cls, data_type: str) -> None:
        """
        Class method to set the data types
        Args:
            data_type (str): '16' or '32'
        """
        if data_type.endswith('32'):
            float_key = 'float32'
            int_key = 'int32'
        elif data_type.endswith('16'):
            float_key = 'float16'
            int_key = 'int16'
        else:
            raise ValueError("Data type not known, choose '16' or '32'")

        cls.np_float = DTYPES[float_key]['numpy']
        cls.tf_float = DTYPES[float_key]['tf']
        cls.np_int = DTYPES[int_key]['numpy']
        cls.tf_int = DTYPES[int_key]['tf'] 

def read_and_decode(filename_queue):
  reader = tf.TFRecordReader()
  _, serialized_example = reader.read(filename_queue)
  features = tf.parse_single_example(
      serialized_example,
      features={
          'image_raw': tf.FixedLenFeature([], tf.string),
          'label_raw': tf.FixedLenFeature([], tf.string),
      })
  image = tf.decode_raw(features['image_raw'], tf.int16)
  image.set_shape([IMAGE_HEIGHT * IMAGE_WIDTH])
  image = tf.cast(image, tf.float32) * (1. / 255) - 0.5
  reshape_image = tf.reshape(image, [IMAGE_HEIGHT, IMAGE_WIDTH, 1])
  label = tf.decode_raw(features['label_raw'], tf.uint8)
  label.set_shape([CHARS_NUM * CLASSES_NUM])
  reshape_label = tf.reshape(label, [CHARS_NUM, CLASSES_NUM])
  return tf.cast(reshape_image, tf.float32), tf.cast(reshape_label, tf.float32) 

def _quantize(x, params, randomize=True):
  """Quantize x according to params, optionally randomizing the rounding."""
  if not params.quantize:
    return x

  if not randomize:
    return tf.bitcast(
        tf.cast(x / params.quantization_scale, tf.int16), tf.float16)

  abs_x = tf.abs(x)
  sign_x = tf.sign(x)
  y = abs_x / params.quantization_scale
  y = tf.floor(y + tf.random_uniform(common_layers.shape_list(x)))
  y = tf.minimum(y, tf.int16.max) * sign_x
  q = tf.bitcast(tf.cast(y, tf.int16), tf.float16)
  return q 

def testIntTypes(self):
    for dtype, nptype in [
        (tf.int32, np.int32),
        (tf.uint8, np.uint8),
        (tf.uint16, np.uint16),
        (tf.int16, np.int16),
        (tf.int8, np.int8)]:
      # Test with array.
      t = tensor_util.make_tensor_proto([10, 20, 30], dtype=dtype)
      self.assertEquals(dtype, t.dtype)
      self.assertProtoEquals("dim { size: 3 }", t.tensor_shape)
      a = tensor_util.MakeNdarray(t)
      self.assertEquals(nptype, a.dtype)
      self.assertAllClose(np.array([10, 20, 30], dtype=nptype), a)
      # Test with ndarray.
      t = tensor_util.make_tensor_proto(np.array([10, 20, 30], dtype=nptype))
      self.assertEquals(dtype, t.dtype)
      self.assertProtoEquals("dim { size: 3 }", t.tensor_shape)
      a = tensor_util.MakeNdarray(t)
      self.assertEquals(nptype, a.dtype)
      self.assertAllClose(np.array([10, 20, 30], dtype=nptype), a) 

def _input(self, dtype='float32', shape=None, name=None):
        
        """Define an input for the recommender.

        Parameters
        ----------
        dtype: str
            Data type: "float16", "float32", "float64", "int8", "int16", "int32", "int64", "bool", or "string".
        shape: list or tuple
            Input shape.
        name: str
            Name of the input.

        Returns
        -------
        Tensorflow placeholder
            Defined tensorflow placeholder.
        """
        if dtype not in self._str_to_dtype:
            raise ValueError
        else:
            return tf.placeholder(self._str_to_dtype[dtype], shape=shape, name=name) 

def diet_adam_optimizer_params():
  """Default hyperparameters for a DietAdamOptimizer.

  Returns:
    a hyperparameters object.
  """
  return hparam.HParams(
      quantize=True,  # use 16-bit fixed-point
      quantization_scale=10.0 / tf.int16.max,
      optimizer="DietAdam",
      learning_rate=1.0,
      learning_rate_warmup_steps=2000,
      learning_rate_decay_scheme="noam",  # "noam" or "none"
      epsilon=1e-10,
      beta1=0.0,  # we can save memory if beta1=0
      beta2=0.98,
      factored_second_moment_accumulator=True,  # this saves memory
  ) 

def parse_withlabel_function(example_proto):
        time1 = time.time()
        features = {
            'image': tf.FixedLenFeature([], tf.string),
            'image_shape': tf.FixedLenFeature([], tf.string),
            'image_label': tf.FixedLenFeature([], tf.string)
        }

        content = tf.parse_single_example(example_proto, features=features)

        content['image_shape'] = tf.decode_raw(content['image_shape'], tf.int32)
        content['image_label'] = tf.decode_raw(content['image_label'], tf.int16)
        content['image'] = tf.decode_raw(content['image'], tf.int16)
        content['image'] = tf.reshape(content['image'], content['image_shape'])
        print('parse using time: ', time.time() - time1)
        return content['image'], content['image_label'] 

def testOnesLike(self):
    for dtype in [tf.float32, tf.float64, tf.int32,
                  tf.uint8, tf.int16, tf.int8,
                  tf.complex64, tf.complex128, tf.int64]:
      numpy_dtype = dtype.as_numpy_dtype
      with self.test_session():
        # Creates a tensor of non-zero values with shape 2 x 3.
        d = tf.constant(np.ones((2, 3), dtype=numpy_dtype), dtype=dtype)
        # Constructs a tensor of zeros of the same dimensions and type as "d".
        z_var = tf.ones_like(d)
        # Test that the type is correct
        self.assertEqual(z_var.dtype, dtype)
        z_value = z_var.eval()

      # Test that the value is correct
      self.assertTrue(np.array_equal(z_value, np.array([[1] * 3] * 2)))
      self.assertEqual([2, 3], z_var.get_shape()) 

def testDtype(self):
    with self.test_session():
      d = tf.fill([2, 3], 12., name="fill")
      self.assertEqual(d.get_shape(), [2, 3])
      # Test default type for both constant size and dynamic size
      z = tf.ones([2, 3])
      self.assertEqual(z.dtype, tf.float32)
      self.assertEqual([2, 3], z.get_shape())
      self.assertAllEqual(z.eval(), np.ones([2, 3]))
      z = tf.ones(tf.shape(d))
      self.assertEqual(z.dtype, tf.float32)
      self.assertEqual([2, 3], z.get_shape())
      self.assertAllEqual(z.eval(), np.ones([2, 3]))
      # Test explicit type control
      for dtype in (tf.float32, tf.float64, tf.int32,
                    tf.uint8, tf.int16, tf.int8,
                    tf.complex64, tf.complex128, tf.int64, tf.bool):
        z = tf.ones([2, 3], dtype=dtype)
        self.assertEqual(z.dtype, dtype)
        self.assertEqual([2, 3], z.get_shape())
        self.assertAllEqual(z.eval(), np.ones([2, 3]))
        z = tf.ones(tf.shape(d), dtype=dtype)
        self.assertEqual(z.dtype, dtype)
        self.assertEqual([2, 3], z.get_shape())
        self.assertAllEqual(z.eval(), np.ones([2, 3])) 

def testDtype(self):
    with self.test_session():
      d = tf.fill([2, 3], 12., name="fill")
      self.assertEqual(d.get_shape(), [2, 3])
      # Test default type for both constant size and dynamic size
      z = tf.zeros([2, 3])
      self.assertEqual(z.dtype, tf.float32)
      self.assertEqual([2, 3], z.get_shape())
      self.assertAllEqual(z.eval(), np.zeros([2, 3]))
      z = tf.zeros(tf.shape(d))
      self.assertEqual(z.dtype, tf.float32)
      self.assertEqual([2, 3], z.get_shape())
      self.assertAllEqual(z.eval(), np.zeros([2, 3]))
      # Test explicit type control
      for dtype in [tf.float32, tf.float64, tf.int32,
                    tf.uint8, tf.int16, tf.int8,
                    tf.complex64, tf.complex128, tf.int64, tf.bool]:
        z = tf.zeros([2, 3], dtype=dtype)
        self.assertEqual(z.dtype, dtype)
        self.assertEqual([2, 3], z.get_shape())
        self.assertAllEqual(z.eval(), np.zeros([2, 3]))
        z = tf.zeros(tf.shape(d), dtype=dtype)
        self.assertEqual(z.dtype, dtype)
        self.assertEqual([2, 3], z.get_shape())
        self.assertAllEqual(z.eval(), np.zeros([2, 3])) 

def testNumpyConversion(self):
    self.assertIs(tf.float32, tf.as_dtype(np.float32))
    self.assertIs(tf.float64, tf.as_dtype(np.float64))
    self.assertIs(tf.int32, tf.as_dtype(np.int32))
    self.assertIs(tf.int64, tf.as_dtype(np.int64))
    self.assertIs(tf.uint8, tf.as_dtype(np.uint8))
    self.assertIs(tf.uint16, tf.as_dtype(np.uint16))
    self.assertIs(tf.int16, tf.as_dtype(np.int16))
    self.assertIs(tf.int8, tf.as_dtype(np.int8))
    self.assertIs(tf.complex64, tf.as_dtype(np.complex64))
    self.assertIs(tf.complex128, tf.as_dtype(np.complex128))
    self.assertIs(tf.string, tf.as_dtype(np.object))
    self.assertIs(tf.string, tf.as_dtype(np.array(["foo", "bar"]).dtype))
    self.assertIs(tf.bool, tf.as_dtype(np.bool))
    with self.assertRaises(TypeError):
      tf.as_dtype(np.dtype([("f1", np.uint), ("f2", np.int32)])) 

def testZeros(self):
    with self.test_session(use_gpu=True):
      for dtype in tf.uint8, tf.int16, tf.int32, tf.int64:
        zero = tf.constant(0, dtype=dtype)
        one = tf.constant(1, dtype=dtype)
        bads = [one // zero]
        if dtype in (tf.int32, tf.int64):
          bads.append(one % zero)
        for bad in bads:
          try:
            result = bad.eval()
          except tf.OpError as e:
            # Ideally, we'd get a nice exception.  In theory, this should only
            # happen on CPU, but 32 bit integer GPU division is actually on
            # CPU due to a placer bug.
            # TODO(irving): Make stricter once the placer bug is fixed.
            self.assertIn('Integer division by zero', str(e))
          else:
            # On the GPU, integer division by zero produces all bits set.
            # But apparently on some GPUs "all bits set" for 64 bit division
            # means 32 bits set, so we allow 0xffffffff as well.  This isn't
            # very portable, so we may need to expand this list if other GPUs
            # do different things.
            self.assertTrue(tf.test.is_gpu_available())
            self.assertIn(result, (-1, 0xff, 0xffffffff)) 

def _quantize(x, params, randomize=True):
  """Quantize x according to params, optionally randomizing the rounding."""
  if not params.quantize:
    return x

  if not randomize:
    return tf.bitcast(
        tf.cast(x / params.quantization_scale, tf.int16), tf.float16)

  abs_x = tf.abs(x)
  sign_x = tf.sign(x)
  y = abs_x / params.quantization_scale
  y = tf.floor(y + tf.random_uniform(common_layers.shape_list(x)))
  y = tf.minimum(y, tf.int16.max) * sign_x
  q = tf.bitcast(tf.cast(y, tf.int16), tf.float16)
  return q 

def diet_adam_optimizer_params():
  """Default hyperparameters for a DietAdamOptimizer.

  Returns:
    a hyperparameters object.
  """
  return tf.contrib.training.HParams(
      quantize=True,  # use 16-bit fixed-point
      quantization_scale=10.0 / tf.int16.max,
      optimizer="DietAdam",
      learning_rate=1.0,
      learning_rate_warmup_steps=2000,
      learning_rate_decay_scheme="noam",  # "noam" or "none"
      epsilon=1e-10,
      beta1=0.0,  # we can save memory if beta1=0
      beta2=0.98,
      factored_second_moment_accumulator=True,  # this saves memory
  ) 

def _quantize(x, params, randomize=True):
  """Quantize x according to params, optionally randomizing the rounding."""
  if not params.quantize:
    return x

  if not randomize:
    return tf.bitcast(
        tf.cast(x / params.quantization_scale, tf.int16), tf.float16)

  abs_x = tf.abs(x)
  sign_x = tf.sign(x)
  y = abs_x / params.quantization_scale
  y = tf.floor(y + tf.random_uniform(common_layers.shape_list(x)))
  y = tf.minimum(y, tf.int16.max) * sign_x
  q = tf.bitcast(tf.cast(y, tf.int16), tf.float16)
  return q 

def diet_adam_optimizer_params():
  """Default hyperparameters for a DietAdamOptimizer.

  Returns:
    a hyperparameters object.
  """
  return tf.contrib.training.HParams(
      quantize=True,  # use 16-bit fixed-point
      quantization_scale=10.0 / tf.int16.max,
      optimizer="DietAdam",
      learning_rate=1.0,
      learning_rate_warmup_steps=2000,
      learning_rate_decay_scheme="noam",  # "noam" or "none"
      epsilon=1e-10,
      beta1=0.0,  # we can save memory if beta1=0
      beta2=0.98,
      factored_second_moment_accumulator=True,  # this saves memory
  ) 

def _quantize(x, params, randomize=True):
  """Quantize x according to params, optionally randomizing the rounding."""
  if not params.quantize:
    return x

  if not randomize:
    return tf.bitcast(
        tf.cast(x / params.quantization_scale, tf.int16), tf.float16)

  abs_x = tf.abs(x)
  sign_x = tf.sign(x)
  y = abs_x / params.quantization_scale
  y = tf.floor(y + tf.random_uniform(common_layers.shape_list(x)))
  y = tf.minimum(y, tf.int16.max) * sign_x
  q = tf.bitcast(tf.cast(y, tf.int16), tf.float16)
  return q 

def testIsFloating(self):
    self.assertEqual(tf.as_dtype("int8").is_floating, False)
    self.assertEqual(tf.as_dtype("int16").is_floating, False)
    self.assertEqual(tf.as_dtype("int32").is_floating, False)
    self.assertEqual(tf.as_dtype("int64").is_floating, False)
    self.assertEqual(tf.as_dtype("uint8").is_floating, False)
    self.assertEqual(tf.as_dtype("uint16").is_floating, False)
    self.assertEqual(tf.as_dtype("complex64").is_floating, False)
    self.assertEqual(tf.as_dtype("complex128").is_floating, False)
    self.assertEqual(tf.as_dtype("float32").is_floating, True)
    self.assertEqual(tf.as_dtype("float64").is_floating, True)
    self.assertEqual(tf.as_dtype("string").is_floating, False)
    self.assertEqual(tf.as_dtype("bool").is_floating, False)
    self.assertEqual(tf.as_dtype("bfloat16").is_integer, False) 

def testIsComplex(self):
    self.assertEqual(tf.as_dtype("int8").is_complex, False)
    self.assertEqual(tf.as_dtype("int16").is_complex, False)
    self.assertEqual(tf.as_dtype("int32").is_complex, False)
    self.assertEqual(tf.as_dtype("int64").is_complex, False)
    self.assertEqual(tf.as_dtype("uint8").is_complex, False)
    self.assertEqual(tf.as_dtype("uint16").is_complex, False)
    self.assertEqual(tf.as_dtype("complex64").is_complex, True)
    self.assertEqual(tf.as_dtype("complex128").is_complex, True)
    self.assertEqual(tf.as_dtype("float32").is_complex, False)
    self.assertEqual(tf.as_dtype("float64").is_complex, False)
    self.assertEqual(tf.as_dtype("string").is_complex, False)
    self.assertEqual(tf.as_dtype("bool").is_complex, False)
    self.assertEqual(tf.as_dtype("bfloat16").is_integer, False) 

def _remove_subj_tp(self, subj, tp, img, label):
        img = tf.decode_raw(img, tf.float32)
        label = tf.cast(tf.decode_raw(label, tf.int16), tf.float32)
        # hard coded shape because the images are a fixed size
        img = tf.reshape(img, tf.stack([4, 60, 256, 256]))
        label = tf.reshape(label, tf.stack([1, 60, 256, 256]))
        img = self._process_data(img)
        label = self._process_data(label)
        return img, label 

def testIntTypesWithImplicitRepeat(self):
    for dtype, nptype in [
        (tf.int64, np.int64),
        (tf.int32, np.int32),
        (tf.uint8, np.uint8),
        (tf.uint16, np.uint16),
        (tf.int16, np.int16),
        (tf.int8, np.int8)]:
      t = tensor_util.make_tensor_proto([10], shape=[3, 4], dtype=dtype)
      a = tensor_util.MakeNdarray(t)
      self.assertAllEqual(np.array([[10, 10, 10, 10],
                                    [10, 10, 10, 10],
                                    [10, 10, 10, 10]], dtype=nptype), a) 

def test_forward_right_shift():
    _test_forward_right_shift((7,), 'int32')
    _test_forward_right_shift((3, 11), 'int16') 

def test_forward_unpack():
    _test_forward_unpack((3,), 0, 'int32')
    _test_forward_unpack((3,), -1, 'int16')
    _test_forward_unpack((21, 23, 3), 2, 'float32')

#######################################################################
# Range
# ----- 

def testStringConversion(self):
    self.assertIs(tf.float32, tf.as_dtype("float32"))
    self.assertIs(tf.float64, tf.as_dtype("float64"))
    self.assertIs(tf.int32, tf.as_dtype("int32"))
    self.assertIs(tf.uint8, tf.as_dtype("uint8"))
    self.assertIs(tf.uint16, tf.as_dtype("uint16"))
    self.assertIs(tf.int16, tf.as_dtype("int16"))
    self.assertIs(tf.int8, tf.as_dtype("int8"))
    self.assertIs(tf.string, tf.as_dtype("string"))
    self.assertIs(tf.complex64, tf.as_dtype("complex64"))
    self.assertIs(tf.complex128, tf.as_dtype("complex128"))
    self.assertIs(tf.int64, tf.as_dtype("int64"))
    self.assertIs(tf.bool, tf.as_dtype("bool"))
    self.assertIs(tf.qint8, tf.as_dtype("qint8"))
    self.assertIs(tf.quint8, tf.as_dtype("quint8"))
    self.assertIs(tf.qint32, tf.as_dtype("qint32"))
    self.assertIs(tf.bfloat16, tf.as_dtype("bfloat16"))
    self.assertIs(tf.float32_ref, tf.as_dtype("float32_ref"))
    self.assertIs(tf.float64_ref, tf.as_dtype("float64_ref"))
    self.assertIs(tf.int32_ref, tf.as_dtype("int32_ref"))
    self.assertIs(tf.uint8_ref, tf.as_dtype("uint8_ref"))
    self.assertIs(tf.int16_ref, tf.as_dtype("int16_ref"))
    self.assertIs(tf.int8_ref, tf.as_dtype("int8_ref"))
    self.assertIs(tf.string_ref, tf.as_dtype("string_ref"))
    self.assertIs(tf.complex64_ref, tf.as_dtype("complex64_ref"))
    self.assertIs(tf.complex128_ref, tf.as_dtype("complex128_ref"))
    self.assertIs(tf.int64_ref, tf.as_dtype("int64_ref"))
    self.assertIs(tf.bool_ref, tf.as_dtype("bool_ref"))
    self.assertIs(tf.qint8_ref, tf.as_dtype("qint8_ref"))
    self.assertIs(tf.quint8_ref, tf.as_dtype("quint8_ref"))
    self.assertIs(tf.qint32_ref, tf.as_dtype("qint32_ref"))
    self.assertIs(tf.bfloat16_ref, tf.as_dtype("bfloat16_ref"))
    with self.assertRaises(TypeError):
      tf.as_dtype("not_a_type")