Tensor shape
A tensor shape is an ordered sequence of dimension sizes specifying the extent of a tensor or array along each axis.
Fast Python: High performance techniques for large datasets
- How it is used
- Uses shape to interpret array axes and choose efficient views for large-array processing.
- How deep it goes
- Works through shapes alongside byte strides and shared memory, while deferring fuller allocation details.
- Examples
- Uses a ten-element array, grayscale image transformations, and scalar addition to expose shape-dependent behavior.
- What you can do after
- Diagnose/debug with it by checking `.shape`, `.strides`, and shared memory before assuming an array transformation avoids a copy.
Read how Fast Python: High performance techniques for large datasets explains tensor shape →
Deep Learning with PyTorch, Second Edition
The sizes of an array or tensor along each axis, here specifically the channels-height-width layout for image tensors.
- How it is used
- Organizes images into channel-first tensors and batches suitable for neural-network input.
- How deep it goes
- Works through shape inspection, axis permutation, batching, and time-series reshaping, with brief links to strides and broadcasting.
- Examples
- Uses RGB image tensors, a batch of PNG files, coordinate points, and hourly records grouped into days.
- What you can do after
- Operate/configure PyTorch tensors by checking shapes and arranging image or time-series axes for the intended use.
Read how Deep Learning with PyTorch, Second Edition explains tensor shape →
Deep Learning with R, Third Edition
An output shape is the ordered set of dimension sizes describing the tensor produced by a layer.
- How it is used
- Uses tensor shape to describe data organization and check dimensional changes through neural-network layers.
- How deep it goes
- Defines ordered dimensions, works through MNIST and data-layout examples, and applies output-shape summaries to model debugging.
- Examples
- Shows MNIST images, vector datasets, sequences, images, video, and Keras model summaries with changing output shapes.
- What you can do after
- Diagnose/debug with it by inspecting tensor dimensions and checking layer output shapes before or during model construction.
Read how Deep Learning with R, Third Edition explains tensor shape →
Learn Generative AI with PyTorch
The tuple of dimension sizes describing the structure of a multidimensional object.
- How it is used
- Uses dimension sizes to prevent operation errors and guide tensor construction, concatenation, reshaping, indexing, and model data representation.
- How deep it goes
- Defines dimensions, demonstrates shape inspection and transformation, and stops short of broader tensor algebra or model-specific shape rules.
- Examples
- Demonstrates height measurements with shapes (46), (92), and (2, 46), plus a music object shaped (4, 2, 16, 84).
- What you can do after
- Apply it via PyTorch to inspect dimensions, combine tensors, reshape data, and access values with dimension-matched indexes.
Read how Learn Generative AI with PyTorch explains tensor shape →
Inside Deep Learning: Math, Algorithms, Models
The dimensions of a tensor, used to describe its layout and reshaping operations.
- How it is used
- Configures generator outputs and discriminator inputs while preserving a flexible batch dimension.
- How deep it goes
- Defines shape notation and implements reshaping through output-size multiplication and View, without developing convolutional layouts.
- Examples
- Uses (-1, 28, 28) to reshape generated flat values into batches of black-and-white MNIST images.
- What you can do after
- Apply tensor shapes via PyTorch by configuring out_shape and reshaping network outputs with View.
Read how Inside Deep Learning: Math, Algorithms, Models explains tensor shape →
Related concepts
- stridecontrasts with · A stride is the step interval between successive positions or elements accessed, selected, or processed along a dimension or sequence.
- circleincludes · Circle is a book-defined Python class whose instances represent circles with a radius data attribute and operations for changing or retrieving that radius.
- broadcastingused by · Broadcasting is the distribution or implicit replication of data from one source across multiple recipients or compatible positions for communication or computation.
- collision detectionused by · Collision detection is the computation of whether modeled objects or entities intersect, touch, overlap, or occupy conflicting spatial regions.
- tensoruses · A tensor is a multidimensional, array-like structure of values indexed along zero or more dimensions and used for numerical computation.
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.