linacon 0.1.0

linear algebra in your console

linacon

Linear algebra in your Console — yayo!

linacon runs in your terminal and provides a set of linear algebra commands. All operations are performed using NumPy.

Status: sections [0]–[4] of the Feature roadmap are implemented and reachable from the CLI — basic matrix operations, property checks, rotations, factorizations and least squares, all backed by NumPy in src/linacon/core. I/O beyond headerless CSV (section [5]) is still a roadmap.

Usage

Requires Python >= 3.14, NumPy and PyArrow.

Every feature is reached as <command> <feature>, loading its matrices from CSV files:

# multiply two matrices
python -m linacon.main bmo matmul -f A.csv -g B.csv

# LU decomposition, written to CSV
python -m linacon.main factorization lu -f A.csv -out lu.csv

# ridge least squares (lambda via -k)
python -m linacon.main least-squares regularized -f A.csv -g b.csv -k 1.0

Use python -m linacon.main <command> <feature> --help to list a feature's flags.

---

Technology Stack Problem

Pure Python is too slow, approximatelly 3sec for cicle 10000 numbers, inaceptable counting the opening of the file in memory

I/O

You can set stdin and stdout to use files, pipes, or both in powerful combinations.

Loading Pipeline

The loader delegates parsing to PyArrow's pyarrow.csv.read_csv, configured for headerless numeric CSV: autogenerated column names, internal multi-threaded parsing, and no quoting/escaping for speed. The resulting Arrow table columns are stacked with np.column_stack into the final 2D NumPy array. See run.

Argument Parser

The parser is split by command (bmo, checks, rotation, factorization, least-squares), and each command exposes one sub-command per feature — every feature is reached through a command, never by direct access. Shared flags come from parent parsers: -f/--file (primary matrix), -g/--file2 (second operand), -k/--scalar, -b (start-to-end benchmark) and -out/--out (CSV output path). See get_argparse.

Output

When -out is given, the operation's result is written back as headerless CSV via PyArrow (loader-compatible). Multi-part results (e.g. P/L/U from a factorization, or the transform and transformed points from a rotation) are written one file per component as <stem>.<name><suffix>. See write_csv.

---

Feature

[0] Basic Matrix Operations

• [0.1] matrix mul (arbitrary dimensions)
• [0.2] matrix sum
• [0.3] matrix scalar mul
• [0.4] scalar dot product
• [0.5] scalar sum
• [0.6] inverse compute
• [0.7] transpose compute
• [0.8] rank compute

[1] Checks

• [1.1] invertability check
• [1.2] vectors independence check
• [1.3] vectors orthogonality check
• [1.4] symmetric check
• [1.5] triangular check
• [1.6] positive definite check

[2] Rotations

• [2.1] rotation matrix computation (input: matrix, angle, [axis], [center], [scale], [shear], [perspective])

[3] Factorization

• [3.1] gauss-jordan elimination
• [3.2] LU decomposition
• [3.3] LDU decomposition
• [3.4] QR decomposition
• [3.5] Cholesky decomposition
• [3.6] orthogonal decomposition
• [3.7] SVD
• [3.8] eigenvalue decomposition

[4] Least Squares

• [4.1] least squares
• [4.2] weighted least squares
• [4.3] least squares with regularization
• [4.4] regularization
• [4.5] linear regression