miplog

Parse MIP/LP solver log files into a unified, machine-readable format.

Every solver writes its log differently — different column names, different ways to say "optimal", different units. miplog reads any supported solver's log and gives you back the same shape, so you can analyze runs across solvers without writing one parser per format.

Supported solvers: SCIP 10–11, Gurobi 11–13, Xpress 9, HiGHS 1.12–14, CPLEX 12.7–12.8, CBC 2.9, COPT 8.0, OptVerse 2.0, Mosek 11.0.

Python

pip install miplog

import miplog

log = miplog.parse_file("run.log.gz")   # plain or gzipped, solver auto-detected
print(log["solver"], log["termination"]["status"])
print(f"obj = {log['bounds']['primal']}, gap = {log['bounds']['gap']}")

# B&B progress is stored columnar — drop straight into pandas/numpy
import pandas as pd
df = pd.DataFrame(log["progress"])

Available functions: parse_file(path, solver=None), parse_text(text), split_concatenated(text) for Mittelmann-style bundled runs.

Command line

Requires the Rust toolchain (install via rustup.rs):

cargo install miplog

miplog run.log                    # human-readable summary
miplog run.log --format json      # machine-readable JSON
miplog run.log -o run.json.gz     # compressed archive (extension-inferred)
cat run.log | miplog -            # stdin works too

Sample output (a SCIP run hitting a time limit on glass4):

solver: scip 11.0.0
problem: glass4
status: time-limit in 2.00s
primal: 4350038500
dual: 800004879.356464
gap: 443.75%
sols: 2
presolve: 396→393 rows, 322→317 cols
convergence: ████████████████████

       time     nodes           dual         primal     gap  event
       0.00         1     8.000024e8              -       -
       0.00         1     8.000024e8     4.450042e9  456.2%  H
    … same for 4 more rows …
       0.00         1     8.000031e8     4.450042e9  456.2%
       0.10         1     8.000033e8     4.450042e9  456.2%
       0.10         1     8.000044e8     4.450042e9  456.2%
       0.10         1     8.000046e8     4.450042e9  456.2%
    … same for 5 more rows …
       0.60         1     8.000049e8     4.450042e9  456.2%
       0.90         1     8.000049e8     4.350038e9  443.8%  H
    … same for 3 more rows …
       1.70         1     8.000049e8     4.350038e9  443.8%

Identical-looking rows are elided. Incumbent updates (H for heuristic, * for branch-found solution) are always kept. Pass --no-progress to suppress the table entirely.

Rust

use miplog::{autodetect, input};

let text = input::read_file("run.log.gz")?;  // plain or gzipped
let log = autodetect(&text)?;                // solver auto-detected
println!("{log}");
# Ok::<(), Box<dyn std::error::Error>>(())

Full API reference: docs.rs/miplog.

License

MIT OR Apache-2.0.