Stride
A stride is the step interval between successive positions or elements accessed, selected, or processed along a dimension or sequence.
Math and Architectures of Deep Learning
The step size with which a convolutional kernel moves across the input, affecting the output size in convolution and transposed convolution.
- How it is used
- Controls convolution and pooling stepping to determine which local positions produce output values and how feature maps are downsampled.
- How deep it goes
- Derives 1D output size from kernel bounds, then extends stepping to multidimensional and transposed convolutions.
- Examples
- Uses length-7 1D averaging windows, 3D video-volume stepping, PyTorch Conv1d calls, and 2 × 2 stride-2 pooling.
- What you can do after
- Apply it via PyTorch convolution or pooling parameters and calculate resulting convolution or transposed-convolution output sizes.
Read how Math and Architectures of Deep Learning explains stride →
Learn Generative AI with PyTorch
The step-size parameter that controls how far a convolutional operation moves and, in transposed convolution, how much upsampling occurs.
- How it is used
- Controls convolutional scanning and spatial downsampling in image-processing networks.
- How deep it goes
- Explains movement, output-size effects, padding interaction, numerical calculations, and VAE tensor-shape changes.
- Examples
- Uses a 3 × 3 image, fixed 2 × 2 weights, stride 2, padding 1, and a stride-based VAE encoder.
- What you can do after
- Apply stride via PyTorch Conv2d arguments and predict its effects on feature-map dimensions.
Fast Python: High performance techniques for large datasets
- How it is used
- Uses strides to describe element access in array dimensions and enable efficient NumPy views.
- How deep it goes
- Defines strides as byte steps and demonstrates them with reshaping, transposition, slicing, and image views.
- Examples
- Demonstrates reshaped and transposed arrays, reversed dimensions, and flipped, rotated, or sliced images.
- What you can do after
- Understand how shape and strides describe NumPy views over existing array data.
Read how Fast Python: High performance techniques for large datasets explains stride →
Deep Learning with PyTorch, Second Edition
Tensor metadata describing its shape and how elements are laid out and indexed in storage, including step size between positions.
- How it is used
- Explains how PyTorch tensors index storage and why views, slicing, transposition, and reshaping can avoid data copies.
- How deep it goes
- Derives the indexing formula and follows stride through views, transposes, reshaping, and contiguity, stopping short of deeper memory-system details.
- Examples
- Uses point tensors, indexed subtensors, transposes, contiguous copies, daily bike sequences, and size-offset-stride exercises.
- What you can do after
- Diagnose tensor layouts by predicting stride, storage offset, shared storage, and contiguity after indexing or reshaping.
Read how Deep Learning with PyTorch, Second Edition explains stride →
Grokking AI Algorithms, Second Edition
A stride is the spatial step size by which a convolutional filter moves across an input.
- How it is used
- Shrinks convolutional feature maps by increasing the distance between successive filter positions.
- How deep it goes
- Explains filter movement, padding, and output-size calculation through an 8×8-to-4×4 example and implementation.
- Examples
- Uses a 4×4 filter with stride 2 to reduce eight 8×8 feature maps to 4×4 maps.
- What you can do after
- Build it from scratch by configuring stride and calculating output dimensions in a downsampling convolution.
Read how Grokking AI Algorithms, Second Edition explains stride →
Build a Large Language Model (From Scratch)
The step size by which the sliding window advances between sampled training examples.
- How it is used
- Controls how training contexts are sampled from tokenized text for GPT-style input-target pairs.
- How deep it goes
- Defines stride and implements it in the dataset loop, including overlap behavior and context-size interactions.
- Examples
- Demonstrates stride values of 1, 2, and 4 with context sizes of 2, 4, and 8.
- What you can do after
- Apply it via GPTDatasetV1 and DataLoader to control overlap among sampled training contexts.
Read how Build a Large Language Model (From Scratch) explains stride →
Natural Language Processing in Action, Second Edition
A stride is the number of input positions by which a convolution kernel moves between successive applications.
- How it is used
- Uses stride to control how convolution kernels scan token sequences and produce pattern-location maps.
- How deep it goes
- Explains stride through adjacent-token experiments and output-length formulas, but stops short of broader optimization guidance.
- Examples
- Demonstrates stride with part-of-speech bigrams and PyTorch Conv1d configurations using stride 1.
- What you can do after
- Apply stride by selecting a scan step and calculating the resulting convolution sequence length.
Read how Natural Language Processing in Action, Second Edition explains stride →
Python How-To
- How it is used
- Uses stride to select spaced elements from a sequence and to replace selected positions in a mutable list.
- How deep it goes
- Explains stride in slice notation, positive and negative directions, and list replacement at strided positions.
- Examples
- Uses list slices to select every third element, reverse a list, and replace every second item.
- What you can do after
- Apply Python slice notation to select spaced elements or replace selected positions in a list.
Related concepts
- tensor shapecontrasts with · A tensor shape is an ordered sequence of dimension sizes specifying the extent of a tensor or array along each axis.
- offsetproduces · An offset is a position or displacement measured relative to a reference point, such as within an ordered sequence, memory region, or layout.
- dilated convolutioncontrasts with · A dilated convolution applies a filter whose sampling positions are spaced apart, expanding the receptive field without proportionally increasing kernel size or parameters.
- boundary checkused by · A boundary check is a test that verifies whether a value or operation lies within an allowed range or valid data extent.
These guides are generated from the text of Manning books by Manning's knowledge graph and link to the sections they draw on. The full books are available on liveBook and at manning.com.