A PyTorch tensor is a multidimensional array of numbers. It holds data in a form that deep learning models can use: image pixels, for example, must become numerical inputs before a model can classify them. Tensors resemble NumPy arrays, but PyTorch can also use them in GPU-accelerated training. section 2.1 section 2.2.1
Start with a Python list of heights. `torch.tensor(heights, dtype=torch.float64)` turns it into a tensor of 64-bit floating-point values. The `dtype` argument controls the numerical type: `float64` offers more precision than `float32` at a computational cost. PyTorch also provides constructors such as `torch.zeros(2, 3)` and `torch.ones(1, 4, 5)` when the desired shape is known. section 2.1.1
import torch
heights_tensor = torch.tensor(heights, #1
dtype=torch.float64) #2
See the list-to-tensor conversion and its explicit `dtype`. section 2.1.1
Use square brackets to select values as you would with a Python list: `heights_tensor[2]` selects one height, while `heights_tensor[-5:]` selects the last five. Check `.shape` before combining tensors; incompatible shapes can cause errors. A 46-value height tensor has shape `(46,)`. Concatenating centimeter and foot measurements produces 92 values, which can be reshaped to `(2, 46)`—one row per unit. In that 2D tensor, `[1, -2]` selects the second row’s second-to-last value. section 2.1.2 section 2.1.3
Operations can act on selected values or along a dimension. `torch.median(heights_reshaped[0, :])` finds the median centimeter height. `torch.mean(heights_reshaped, dim=1)` returns one average for each row; `torch.max(..., dim=1)` returns both maximum values and their indices. Here, `dim` specifies which dimension the operation reduces. section 2.1.4
From image pixels to model inputs
For grayscale clothing images, `T.ToTensor()` converts image values from integers in the range 0–255 to floating-point tensors in the range 0–1. `T.Normalize([0.5], [0.5])` then moves them to the range −1 to 1. The resulting tensors are ready to enter an image-classification workflow. section 2.2.2 section 2.2.1
