rmath.array API Reference

The rmath.array module provides the foundation of RMath: the Array class. This is a parallel, N-dimensional numeric array that supports matrix operations, reductions, and seamless interop with NumPy, PyTorch, and JAX without the GIL.

Creation & Basic Ops

array_basics.py

import rmath.array as ra

# Multiple constructors
z = ra.zeros(3, 3)
o = ra.ones(2, 4)

# Create from a Python list
a = ra.Array([[1.0, 2.0], [3.0, 4.0]])

print(f"ones shape : {o.shape}")
print(f"array sum  : {a.sum()}")
print(f"array mean : {a.mean()}")
print(f"transpose  :
{a.transpose()}")

ones shape : [2, 4] array sum : 10.0 array mean : 2.5 transpose : Array([ [ 1.0000, 3.0000], [ 2.0000, 4.0000]])

API Reference (Selected Methods)

The Array class contains over 100 methods for high-performance numerical computing. Below is a subset.

Method	Description
`ra.zeros(*shape)`	Create an array of zeros.
`ra.ones(*shape)`	Create an array of ones.
`ra.randn(*shape)`	Normal random samples.
`ra.arange(start, end, step)`	Step-wise sequence of values.
`a.shape`	Returns the dimensions of the array.
`a.reshape(rows, cols)`	Reshape the array into new dimensions.
`a.transpose()`	Transpose the array dimensions.
`a.matmul(other)`	Matrix multiplication (equivalent to `a @ other`).
`a.sum()`	Sum of all elements via Kahan compensation.
`a.mean()` / `a.variance()`	Statistical reductions.
`a.lazy()`	Returns a `LazyPipeline` to fuse loop operations without allocations.
`a.to_numpy()`	Zero-copy conversion to a NumPy ndarray.
`a.from_numpy(arr)`	Create an RMath Array from a NumPy ndarray.

Performance Note: All math operations like .sin(), .exp(), and reductions like .sum() automatically dispatch to a background Rayon thread pool if the array has more than 8,192 elements.