The *_like Functions in PyTorch

2026-05-16 · Source: Deep Learning on Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Novice, short

Summary

PyTorch offers a suite of "_like" functions designed to simplify tensor creation by automatically matching the shape of an existing input tensor. These functions eliminate the need for manually specifying tensor dimensions, which can become cumbersome for multi-dimensional tensors like `(3, 4, 5, 6)`. Key functions include `torch.ones_like()`, which creates a tensor filled with ones; `torch.empty_like()`, which produces an uninitialized tensor; and `torch.full_like()`, allowing users to specify a custom fill value. For random tensor generation, `torch.rand_like()` populates a tensor with uniform random numbers between 0 and 1, `torch.randn_like()` uses a normal distribution (mean 0, variance 1), and `torch.randint_like()` generates random integers within a specified range (inclusive lower bound, exclusive upper bound). This approach enhances code readability and reduces errors in machine learning workflows.

Key takeaway

For Machine Learning Engineers creating new tensors, understanding PyTorch's `_like` functions is crucial for writing cleaner, less error-prone code. You should integrate these functions into your tensor initialization routines to automatically match shapes from existing tensors, thereby avoiding repetitive manual dimension specification. This practice will significantly improve code maintainability and efficiency in your PyTorch projects, especially when dealing with complex model architectures.

Key insights

PyTorch's `_like` functions streamline tensor creation by automatically inheriting shape from an existing tensor.

Principles

• Automate shape inference for new tensors
• Reduce manual shape specification errors

Method

Use `_like` functions (e.g., `torch.zeros_like()`, `torch.rand_like()`) with an existing tensor as input to create a new tensor of identical shape, populated with fixed or random values.

In practice

• Initialize weight matrices with `torch.randn_like()`
• Create mask tensors with `torch.ones_like()`
• Generate placeholder tensors with `torch.empty_like()`

Topics

• PyTorch
• Tensor Creation
• *_like Functions
• torch.ones_like
• torch.empty_like

Best for: AI Student, Machine Learning Engineer, Data Scientist

Related on AIssential

• The Most Complete Guide to PyTorch for Data Scientists — PyTorch offers high customizability and a Pythonic interface for building and training neural networks.
• ProcFunc: Function-Oriented Abstractions for Procedural 3D Generation in Python — ProcFunc streamlines Blender-based procedural 3D generation in Python, enabling diverse synthetic data creation.
• Build a PyTorch ReLU Kernel with Hugging Face Kernels (CPU + Metal) — The Hugging Face Kernels Library standardizes kernel distribution and execution across diverse hardware and software.
• Hello, PyTorch! Let the Framework Do the Calculus — PyTorch automates gradient computation and parameter management, simplifying machine learning model development.
• Understanding PyTorch Buffers — PyTorch buffers ensure non-learnable tensors within modules are correctly managed across devices, preventing runtime errors.
• Simplify Sparse Deep Learning with Universal Sparse Tensor in nvmath-python — The Universal Sparse Tensor (UST) in nvmath-python v0.9.0 accelerates sparse operations via flexible formats and zero-cost interoperability.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Deep Learning on Medium.