How to Get a Type of a Tensor in PyTorch

The primary and most efficient way to get the data type of a tensor is to use the .dtype attribute. We can use the print() function to print the type in the console.

import torch

# Creating a Tensor
tensor = torch.tensor([11.0, 21.0, 31.0])

print(tensor.dtype)
# Output: torch.float32

You can see that it created a tensor of type torch.float32 values. It is the type of elements in the tensor.

We can set the data type while creating a tensor. By default, it creates a tensor with torch.float32 type, but using the dtype attribute, we can also set it.

import torch

# Float64 tensor
tensor_float64 = torch.tensor([11, 21], dtype=torch.float64)
print(tensor_float64.dtype)
# Output: torch.float64


# Int32 tensor
tensor_int32 = torch.tensor([11, 21], dtype=torch.int32)
print(tensor_int32.dtype)
# Output: torch.int32


# Boolean tensor
tensor_bool = torch.tensor([True, False], dtype=torch.bool)
print(tensor_bool.dtype)
# Output: torch.bool

Getting the object type

To get the type of the object (not the individual elements’ type), use the built-in Python type() method.

import torch

tensor = torch.tensor([90, 100])

print(type(tensor))
# Output: <class 'torch.Tensor'>

The above output shows that the object is of type Tensor. It can be anything from a Parameter or a custom subclass.

Getting the full type string

If you want detailed type info, you can use the .type() method on a tensor.

import torch

tensor = torch.tensor([1.0])

print(tensor.type())

# Output: torch.FloatTensor

.is_* Methods

You can check if the tensor matches a specific dtype (e.g., .is_floating_point(), .is_complex()) using is.* family methods.

import torch

tensor = torch.tensor([11.0, 21.0])

print(tensor.is_floating_point())
# Output: True

print(tensor.is_complex())
# Output: False

The input tensor has a floating value. So, is_floating_point() method returns True. However, the values are not of type complex. So, the is_complex() method returns False.