Packages
Calculates 4x4 homogeneous transformation matrices for 3D coordinate operations (translation, rotation, scaling, shearing) and converts between rotation matrices, Euler angles, and quaternions.
However, for new projects, consider the alternatives mentioned in the documentation (Pytransform3d, scipy.spatial.transform, transforms3d) which may offer more active…
Equinox provides neural network and model building on top of JAX with PyTorch-like syntax, plus PyTree manipulation, filtered transformations, and runtime error handling—all while remaining fully compatible with core JAX operations.
TensorFlow Probability provides probabilistic modeling, statistical inference, and Bayesian machine learning tools integrated with TensorFlow, including distributions, variational inference, MCMC sampling, and neural network layers with uncertainty quantification.
Provides compiled C++ kernels and extensions that accelerate operations on nested, variable-sized data structures with performance comparable to NumPy.
Uproot reads and writes ROOT files (the data format used in high-energy physics) directly in Python using NumPy, without requiring the C++ ROOT library.
Install it if you work with ROOT files or need to integrate them into Python data pipelines.
LinearOperator abstracts structured matrix operations in PyTorch, enabling efficient computation on large matrices without materializing them in memory by exploiting their algebraic structure.
However, the pre-alpha status and incomplete feature set (symmetric, generic square, rectangular, and sparse operators not yet implemented) mean you should verify…
Provides unit-aware measurement objects for Python that support conversion between different units and arithmetic operations across multiple measurement types including distance, weight, temperature, energy, speed, volume, time, and area.
ECOS is a Python wrapper for a numerical solver that handles convex second-order cone programs (SOCPs), accepting linear and cone constraints and returning optimal solutions.
TensorFlow Addons provides experimental operators, layers, metrics, losses, and optimizers that extend core TensorFlow with functionality not yet ready for the main library.
However, the dormant status (no releases in 990 days) means you should verify compatibility with your current TensorFlow and Python versions before committing.
Provides fast implementations of core algorithms for multi-objective optimization, including nondominated set generation, dominance filtering, hypervolume and quality metrics, and empirical attainment function computation.
boost-histogram provides Python bindings to Boost::Histogram, a C++14 library for fast histogram creation and manipulation with support for multiple axis types, storage modes, and advanced indexing.
Provides integration modules connecting Optuna hyperparameter optimization with third-party ML frameworks like PyTorch, scikit-learn, TensorFlow, XGBoost, LightGBM, and others.
Python wrapper for the TA-Lib C library providing technical analysis indicators and candlestick pattern recognition for financial market data.
However, you must pre-install the underlying C library on your system—this is non-trivial on some platforms.
Claripy is an abstraction layer for constraint solvers that wraps Z3 and other backends, letting you build and solve symbolic constraints over bit-vectors and other domains without being tied to a specific solver implementation.
pymoo implements single- and multi-objective optimization algorithms with visualization and decision-making tools for solving complex optimization problems in Python.
Provides a TensorBoard plugin and standalone server for profiling and visualizing ML model performance across multiple devices, showing execution timelines, memory usage, and computational graphs.
The main gotcha is the internet requirement for full UI rendering; if you profile in offline environments, verify that limitation first.
NeMo Gym provides infrastructure for building, running, and scaling evaluation and training environments where agents interact with tasks, datasets, verifiers, and execution state to solve problems.
However, the strict Python 3.13.14+ requirement and heavy dependency footprint may conflict with existing projects.
emcee implements affine-invariant ensemble sampling for Markov chain Monte Carlo (MCMC), enabling Bayesian inference and parameter estimation through parallel sampling of posterior distributions.
Install it if you need to perform Bayesian inference or sample from complex posterior distributions.
Provides a unified Python interface to solve convex quadratic programs using a choice of backend solvers, returning either the primal solution or primal and dual results.
Install it if you need to solve convex quadratic programs or want to experiment with multiple QP solvers.
Reads, writes, and analyzes SVG Path objects and Bézier curves, providing geometric tools to transform, intersect, and measure path elements.
Rounds numbers by significant figures, decimal places, or uncertainty, and formats them in multiple scientific and publication styles with results that match expected mathematical behavior.
Enables PyTorch to run computations on Huawei Ascend NPU hardware, bridging PyTorch's tensor operations to Ascend AI Processors.
Provides precompiled qhull binaries (convex hull, Delaunay triangulation, Voronoi diagrams, halfspace intersection) as a Python package via the cmeel distribution system.
However, install only after clarifying the upstream license terms—the package metadata does not declare them, creating legal ambiguity for proprietary or…
Bayesian optimization using Gaussian processes to find the maximum of an expensive unknown function with minimal iterations, balancing exploration and exploitation automatically.
Amply parses AMPL data syntax (sets and parameters) into Python data structures, letting you load optimization problem data without a full AMPL installation.
Implements the Leiden community detection algorithm for graphs, exposing a C++ implementation to Python via igraph for partitioning networks into communities using multiple optimization methods.
ai-edge-litert runs machine learning models on mobile and embedded devices with low latency and small binary footprint, supporting Android, iOS, and other operating systems.
TensorBoard is a web-based visualization suite for inspecting TensorFlow training runs, displaying scalar metrics, histograms, images, and computational graphs from event logs.
tf-nightly provides nightly builds of TensorFlow, an open-source framework for numerical computation and machine learning that runs on CPUs, GPUs, TPUs, and edge devices.
Hist provides a user-friendly interface for creating, filling, and analyzing histograms, built on top of boost-histogram with support for named axes, advanced indexing, and integrated plotting.
SALib implements global sensitivity analysis methods (Sobol, Morris, FAST, DGSM, PAWN, HDMR, and others) to quantify how model inputs affect outputs in systems modeling and uncertainty analysis.
Install it if you need to understand parameter importance or uncertainty propagation in a computational model.
OSMnx downloads, models, and analyzes street networks and geospatial features from OpenStreetMap, letting you work with walking, driving, or biking networks, amenities, building footprints, and routing data.
Install it if you need to work with street networks, urban amenities, or geospatial features from OpenStreetMap.
Provides statistical computations and diagnostic functions for Bayesian model analysis, including posterior analysis, model checking, and comparison metrics.
Solves linear assignment problems using Jonker-Volgenant and related algorithms, supporting single and batch operations on square and rectangular cost matrices.
Install it if you need to solve assignment problems and want more flexibility or batch processing capabilities.
Tests whether a graph is planar, computes planar embeddings, draws planar graphs as ASCII art, and isolates forbidden subgraphs using algorithms from the Edge Addition Planarity Suite.
MXNet is a deep learning framework that enables you to build and train neural networks with support for multiple programming interfaces and hardware accelerators.
No, not for new projects.
Vector provides 2D, 3D, and 4D space-time vector operations optimized for working with arrays of vectors rather than individual vectors in loops, with support for multiple coordinate systems and data backends.
Install it if you need to work with spatial vectors at scale or in high-energy physics contexts.
ScaNN performs efficient vector similarity search at scale using search space pruning and quantization, supporting multiple distance functions including inner product and Euclidean distance.
coffea provides columnar data analysis tools for high-energy physics (HEP) experiments, enabling efficient manipulation of ROOT file data and event structures using NumPy-like syntax with horizontal scaling support.
Moyopy is a Python binding for Moyo, a crystal symmetry finder that identifies and analyzes symmetry operations in crystalline structures.