Tensor reshaping

Q: How deeply does Inside Deep Learning: Math, Algorithms, Models cover tensor reshaping?

Explains compatible dimensions, -1 inference, view tradeoffs, and flattening, but stops short of advanced memory-layout details.

As explained in Inside Deep Learning: Math, Algorithms, Models

An operation that changes a tensor's shape without changing its total number of values, used here to turn a network output into an image-shaped tensor.

Tensor reshaping changes how values are arranged without changing their total count. A tensor containing six values can become a vector, a 2×3 matrix, or another compatible shape. section 2.1.4

Use reshaping carefully in PyTorch:

Specify known dimensions and use -1 for one dimension whose size PyTorch should infer. section 2.1.4
Use reshape to change a convolutional output from (C, W, H) into (C×W×H) before a linear layer; this operation is called flattening. section 3.4.3
Prefer reshape for straightforward, safe code; view can use less memory and run faster but may fail inappropriately. section 2.1.4

Six values, several shapes

A tensor with N=6 values can become shape (3, 2), because 3×2=6. In dataset preparation, X.reshape(-1, 1) turns each scalar feature into a one-column input. For convolutional networks, nn.Flatten performs the analogous collapse of spatial dimensions. section 2.1.4 section 3.4.3

How Inside Deep Learning: Math, Algorithms, Models covers it

Cover of Inside Deep Learning: Math, Algorithms, Models

liveBook manning.com

Read Inside Deep Learning: Math, Algorithms, Models on liveBook Get the book at manning.com

Also explained in

liveBook manning.com

Grokking Machine LearningAn array operation that changes the dimensions of stored values without changing their underlying elements.
liveBook manning.com

Grokking StatisticsA NumPy array operation that changes an array’s dimensions without changing its element values.

Compare how 3 books explain tensor reshaping →

Related concepts

transpositionincludes · Transposition is an operation that exchanges a matrix’s rows and columns or, more generally, reorders the axes of a tensor without changing its values.
tensoruses · A tensor is a multidimensional, array-like structure of values indexed along zero or more dimensions and used for numerical computation.
viewequivalent to · A view is an alternate representation or access interface for data derived from or backed by underlying storage without necessarily creating an independent copy.
vectorproduces · A vector is an ordered, indexed sequence of elements, often numerical, represented as a mathematical quantity or a computational collection.
grayscale imageproduces · A grayscale image represents each pixel with a single achromatic intensity value rather than separate color components.

This explainer was generated by Manning's knowledge graph from the text of Inside Deep Learning: Math, Algorithms, Models. The section links point to the parts of the book it draws on.