numnom 0.4.0

A fast MPS file parser for mathematical programming

numnom

A fast MPS file parser for Python, powered by Rust.

• Fast: 745 MB/s throughput on raw MPS text (4.2× faster than SCIP on MIPLIB 2017)
• Zero-copy numpy arrays: model data lands directly in numpy with no extra copies
• Compressed files: reads .mps.gz with SIMD-accelerated decompression
• Well-tested: 0 failures across all 1065 MIPLIB 2017 instances

Install

pip install numnom

# Optional: scipy.sparse integration
pip install "numnom[scipy]"

Usage

import numnom

model = numnom.parse_file("problem.mps.gz")

print(model.name, model.num_row, "rows,", model.num_col, "cols")
print("Nonzeros:", model.a_matrix.value.size)

# Numpy arrays — zero-copy from Rust
model.col_cost      # np.ndarray[float64], shape (num_col,)
model.col_lower     # np.ndarray[float64]
model.col_upper     # np.ndarray[float64]
model.row_lower     # np.ndarray[float64]
model.row_upper     # np.ndarray[float64]
model.col_integrality  # np.ndarray[uint8] — see numnom.{CONTINUOUS,INTEGER,...}

# CSC sparse matrix
A = model.a_matrix
A.start    # np.ndarray[uint32], shape (num_col + 1,)
A.index    # np.ndarray[uint32], shape (nnz,)
A.value    # np.ndarray[float64], shape (nnz,)

# Names
model.col_names    # list[str]
model.row_names    # list[str]

scipy.sparse integration

import numnom

model = numnom.parse_file("problem.mps")
A = numnom.to_scipy(model)   # scipy.sparse.csc_matrix, shape (num_row, num_col)

Parse from string

mps_text = open("problem.mps").read()
model = numnom.parse_str(mps_text)

Write MPS

numnom.write_file(model, "out.mps")
mps_text = numnom.write_str(model)

Variable type codes

model.col_integrality is a uint8 array; values map to:

Code	Constant	Meaning
0	numnom.CONTINUOUS	Continuous
1	numnom.INTEGER	Integer
2	numnom.SEMI_CONTINUOUS	Semi-continuous
3	numnom.SEMI_INTEGER	Semi-integer

License

Apache-2.0