Beginner

Shape Ops & Indexing

reshape, squeeze/unsqueeze, permute, and indexing: same spirit as NumPy.

Call	Effect	Example
`x.squeeze()`	remove all size-1 dimensions	shape `(1, 3, 1)` → `(3,)`
`x.unsqueeze(dim)`	insert a new size-1 dimension at `dim`	shape `(3,)`, `unsqueeze(0)` → `(1, 3)`
`x.permute(2, 0, 1)`	reorder dimensions arbitrarily	shape `(H, W, C)` → `(C, H, W)`
`torch.cat([a, b], dim=0)`	join along an EXISTING dimension	two `(2, 3)` tensors → `(4, 3)`
`torch.stack([a, b], dim=0)`	join along a NEW dimension	two `(2, 3)` tensors → `(2, 2, 3)`

⚠️ Watch out: x.view() requires contiguous memory and fails after operations like .transpose()/.permute(): use .reshape() (or call .contiguous() first) to be safe.

Now practice it

Exercises that use what this lesson just covered.

3×3 Matrix 0–8Easy Transpose a MatrixEasy Flatten a TensorEasy Squeeze Singleton DimensionsEasy Unsqueeze at a Given DimensionEasy Permute Image Dimensions (HWC to CHW)Medium

← Tensors Basics Autograd →