PyTorch Min: Get Minimum Value Of A PyTorch Tensor

This video will show you how to use PyTorch’s min operation to calculate the minimum of a PyTorch tensor.

First, we import PyTorch.

import torch

Then we check what version of PyTorch we are using.

print(torch.__version__)

We are using PyTorch version 0.4.1.

Let’s now create the tensor we’ll use for the PyTorch min operation example.

tensor_min_example = torch.tensor(
[
  [
      [1,-10, 1],
      [2, 2, 2],
      [3, 3, 3]
  ],
  [
      [4, 4, 4],
      [5,50, 5],
      [6, 6, 6]
  ]
]
)

We use torch.tensor to create a floating point tensor.

We pass in our data structure which is going to be 2x3x3, and we assign it to the Python variable tensor_min_example.

Let’s print the tensor_min_example Python variable to see what we have.

print(tensor_min_example)

We see that it’s a tensor.

Visually inspecting the PyTorch tensor, we see that the minimum is going to be -10.

Next, let’s programmatically calculate the minimum of the PyTorch tensor using PyTorch’s min operation.

tensor_min_value = torch.min(tensor_min_example)

So torch.min, we pass in the tensor, and we assign it to the Python variable tensor_min_value.

Let’s print the result

print(tensor_min_value)

And we see that the minimum of the tensor is -10.

One thing to note is that PyTorch returns the answer as a 0-dimensional tensor with a value of -10 inside of it.

We can double check that it is actually a PyTorch tensor by using Python’s type operation, and we pass in our tensor that has -10 inside of it.

type(tensor_min_value)

And we see that it is in fact a class of torch.Tensor.

To get the number 10 from the tensor, we’re going to use PyTorch’s item operation.

tensor_min_example.item()

So we pass in our tensor, and then we have the item.

When we do that, we get back the number 10.

Finally, just to double check that it is actually a number, let’s use the Python type operation here as well.

type(tensor_min_example.item())

So type and we’re going to get the item out of the tensor.

So that should be -10.

And we see that it is in fact of class integer.

Perfect! We were able to use PyTorch’s min operation to calculate the minimum of a PyTorch tensor.