NuMojo V0.6 for Mojo 25.1

Community Showcase
1

NuMojo v0.6: Expanding Capabilities with Powerful New Features

The latest release of NuMojo, version 0.6, introduces a host of exciting new features, optimizations, and improvements designed to enhance the library’s functionality and efficiency. From new array operations to significant performance boosts, this update brings NuMojo closer to full feature parity with NumPy while taking advantage of Mojo’s powerful capabilities.

:rocket: New Features

Enhanced Array Manipulation

  • broadcast_to() Method: Enables broadcasting an array to any compatible shape, making it easier to work with differently shaped arrays in mathematical operations.
  • apply_along_axis() Function: This function allows you to apply a 1D function along a specified axis of an n-dimensional array, streamlining data transformation workflows.
  • diagonal() Function & Method: Retrieve the diagonal elements of a matrix with ease.
  • compress() and clip() Functions: These methods allow users to selectively extract and constrain array values efficiently.

Iterators and Performance Enhancements

  • New _NDAxisIter Type: Enables iteration over 1D slices along a specific axis, supporting both C-order and F-order traversal.
  • New ith() Method: Enhances _NDArrayIter and _NDIter, enabling retrieval of the i-th item directly.
  • Memory Layout Optimization: The Flags type has been introduced to replace dictionary-based tracking of memory layout, improving efficiency and readability.

:wrench: Changes and Improvements

Syntax and API Updates

NuMojo has been updated to align with Mojo 25.1’s latest syntax and best practices:

  • Constructor calls such as str() have been replaced with String().
  • index() is now replaced with Int() for better type clarity.
  • The isdigit() function has been converted into a method.
  • Arrays are now constructed directly via NDArray() instead of NDArray.__init__().

Optimized Statistical and Sorting Functions

  • Random Module Updates: Added randint() with Shape as the first argument for flexibility.
  • Statistics Module Improvements: Variance (variance()) and standard deviation (std()) now support axis-wise computation.
  • Sorting Performance Boosts: The sort() function has been significantly optimized, and argsort() now supports sorting along any axis.
  • Extrema Functions Expanded: max() and min() now operate across any axis, enhancing flexibility.

0-D Array Behavior Adjustments

  • NuMojo scalars (0-D arrays) now support basic operations and can be unpacked using either [] or .item().

:x: Deprecated & Removed Features

  • Removed Redundant Statistical Functions: cumvariance, cumstd, cumpvariance, and cumpstd have been deprecated.
  • Removed maxT() and minT() Functions: These have been replaced with more general and powerful alternatives.

:lady_beetle: Bug Fixes & Performance Enhancements

  • ravel() Fix for F-Order Arrays: Ensures correct flattening behavior.
  • NDArray.sort() Now Defaults to Last Axis (-1)
  • Boolean Evaluation Fix: NDArray.__bool__() now returns correct results.
  • Printing Performance Boost: Displaying large arrays is now significantly faster.
  • Boundary Checks for Safety: NDArrayShape and NDArrayStrides now include stricter boundary checks to prevent errors.

:open_book: Documentation & Roadmap Updates

  • All NDArray methods now adhere to the Mojo Docstring Style Guide, improving clarity and usability.
  • The project roadmap has been revised to reflect current progress and upcoming milestones.

Looking Ahead

NuMojo continues to evolve rapidly, with upcoming plans to introduce GPU support and further optimizations for large-scale numerical computations. If you’re looking to contribute or stay up-to-date with the latest developments, check out the project’s GitHub repository. Our api is documented and searchable on our readthedocs.

:rocket: Stay tuned for more powerful updates in future releases!

5 Likes