def array_dtype(a):
"""Returns the tensorflow `DType` of the array.
Note: This is similar to doing tf.convert_to_tensor(a).dtype but not
exactly the same. When `a` is a python scalar or a python array_like
object, Tensorflow attempts to treat it as an int32 or float32 whereas
numpy defaults to int64 and float64 respectively.
Args:
a: Could be a numpy ndarray, a Tensor or a python object that can be
converted to numpy ndarray.
Returns:
A `DType`.
"""
if isinstance(a, tf.Tensor):
return a.dtype
elif isinstance(a, tf.IndexedSlices):
return a.dtype
elif isinstance(a, np.ndarray) or isinstance(a, arrays.ndarray):
return tf.as_dtype(a.dtype)
elif isinstance(a, arrays.ShardedNdArray):
return a.tensors[0].dtype
else:
# If this is a python object, defer to numpy to decide the dtype.
np_dtype = np.array(a, copy=False).dtype
np_dtype = dtypes.canonicalize_dtype(np_dtype)
return tf.as_dtype(np_dtype)
def to_tf_type(dtype):
"""Converts a native python or numpy type to TF DType.
Args:
dtype: Could be a python type, a numpy type or a TF DType.
Returns:
A tensorflow `DType`.
"""
if isinstance(dtype, tf.DType):
return dtype
return tf.as_dtype(dtypes.canonicalize_dtype(np.dtype(dtype)))
def to_numpy_type(dtype):
"""Converts a native python or TF DType to numpy type.
Args:
dtype: Could be a python type, a numpy type or a TF DType.
Returns:
A NumPy `dtype`.
"""
if isinstance(dtype, tf.DType):
return dtype.as_numpy_dtype
return dtypes.canonicalize_dtype(np.dtype(dtype))
def promote_types(type1, type2):
"""Returns the type resulting from applying NumPy type promotion.
Args:
type1: A numpy type.
type2: A numpy type.
Returns:
A numpy type.
"""
type1 = _to_numpy_type(type1)
type2 = _to_numpy_type(type2)
return dtypes.canonicalize_dtype(np.promote_types(type1, type2))
