NuMojo is a library for numerical computing in Mojo 🔥 similar to NumPy, SciPy in Python.
Table of Contents
NuMojo intends to capture a wide swath of numerics capability present in the Python packages NumPy, SciPy, and Scikit.
NuMojo intends to try and get the most out of the capabilities of Mojo including vectorization, parallelization, and GPU acceleration (once available). Currently, NuMojo extends (most of) the standard library math functions to work on array inputs.
We intend NuMojo to be a building block for other Mojo packages that need fast math under the hood without the added weight of a ML back and forward propagation system
NuMojo is not a machine learning library, it will never include back-propagation in the base library.
Our main goal is to implement a fast, comprehensive numerics library in Mojo.
Following are some brief long-term goals. Some of them have already been implemented or partially implemented.
Core data types:
- Native n-dimensional array (
numojo.core.ndarray.NDArray). - Native 2-dimensional array, i.e., matrix (
numojo.mat.matrix.Matrix). - Native fixed-dimension array (to be implemented when trait parameterization is available).
Routines and objects:
- Array creation routines (
numojo.routines.creation) - Array manipulation routines (
numojo.routines.manipulation) - Bit-wise operations (
numojo.routines.bitwise) - Constants (
numojo.routines.constants) - Input and output (
numojo.routines.io) - Text files (
numojo.routines.files) - Text formatting options (
numojo.routines.formatting) - Linear algebra (
numojo.routines.linalg) - Decompositions (
numojo.routines.decompositions) - Products of matrices and vectors (
numojo.routines.products) - Solving (
numojo.routines.solving) - Logic functions (
numojo.routines.logic) - Comparison (
numojo.routines.comparison) - Array contents (
numojo.routines.contents) - Truth value testing (
numojo.routines.truth) - Mathematical functions (
numojo.routines.math) - Arithmetic operations (
numojo.routines.arithmetic) - Exponents and logarithms (
numojo.routines.exponents) - Extrema finding (
numojo.routines.extrema) - Floating point routines (
numojo.routines.floating) - Hyperbolic functions (
numojo.routines.hyper) - Indexing (
numojo.routines.indexing) - Miscellaneous (
numojo.routines.misc) - Rounding (
numojo.routines.rounding) - Sums, products, differences (
numojo.routines.sums,numojo.routines.products,numojo.routines.differences) - Trigonometric functions (
numojo.routines.trig) - Random sampling (
numojo.routines.random) - Sorting, searching, and counting (
numojo.routines.sorting,numojo.routines.searching) - Statistics (
numojo.routines.statistics) - Averages and variances (
numojo.routines.averages) - Calculus, Integration & Derivatives etc
- Optimizers
- Function approximators
Please find all the available functions and objects here.
For a detailed roadmap, please refer to the docs/roadmap.md file.
An example of n-dimensional array (NDArray type) goes as follows.
import numojo as nm
from numojo.prelude import *
fn main() raises:
# Generate two 1000x1000 matrices with random float64 values
var A = nm.random.randn(shape=Shape(1000, 1000))
var B = nm.random.randn(shape=Shape(1000, 1000))
# Generate a 3x2 matrix from string representation
var X = nm.fromstring[f32]("[[1.1, -0.32, 1], [0.1, -3, 2.124]]")
# Print array
print(A)
# Array multiplication
var C = A @ B
# Array inversion
var I = nm.inv(A)
# Array slicing
var A_slice = A[1:3, 4:19]
# Get scalar from array
var A_item = A[Idx(291, 141)]An example of matrix (Matrix type) goes as follows.
from numojo import mat
from numojo.prelude import *
fn main() raises:
# Generate two 1000x1000 matrices with random float64 values
var A = mat.rand(shape=(1000, 1000))
var B = mat.rand(shape=(1000, 1000))
# Generate 1000x1 matrix (column vector) with random float64 values
var C = mat.rand(shape=(1000, 1))
# Generate a 4x3 matrix from string representation
var F = mat.fromstring[i8](
"[[12,11,10],[9,8,7],[6,5,4],[3,2,1]]", shape=(4, 3)
)
# Matrix slicing
var A_slice = A[1:3, 4:19]
var B_slice = B[255, 103:241:2]
# Get scalar from matrix
var A_item = A[291, 141]
# Flip the column vector
print(C[::-1, :])
# Sort and argsort along axis
print(mat.sort(A, axis=1))
print(mat.argsort(A, axis=0))
# Sum the matrix
print(mat.sum(B))
print(mat.sum(B, axis=1))
# Matrix multiplication
print(A @ B)
# Matrix inversion
print(A.inv())
# Solve linear algebra
print(mat.solve(A, B))
# Least square
print(mat.lstsq(A, C))There are two approach to install and use the Numojo package.
This approach invovles building a standalone package file mojopkg.
- Clone the repository.
- Build the package using
mojo package numojo - Move the numojo.mojopkg into the directory containing the your code.
This approach does not require buiding a package file. Instead, when you compile your code, you can include the path of NuMojo reporsitory with the following command:
mojo run -I "../NuMojo" example.mojoThis is more flexible as you are able to edit the NuMojo source files when testing your code.
In order to allow VSCode LSP to resolve the imported numojo package, you can:
- Go to preference page of VSCode.
- Go to
Mojo › Lsp: Include Dirs - Click
add itemand write the path where the Numojo repository is located, e.g./Users/Name/Programs/NuMojo. - Restart the Mojo LSP server.
Now VSCode can show function hints for the Numojo package!
Any contributions you make are greatly appreciated. For more details and guidelines on contributions, please check here
This library is still very much a work in progress and may change at any time.
Distributed under the Apache 2.0 License with LLVM Exceptions. See LICENSE and the LLVM License for more information.
Built in native Mojo which was created by Modular.
