---
title: "From Python"
description: "Run a benchmark from a Python script and analyse the results with pandas."
---

> Documentation Index
> Fetch the complete documentation index at: https://docs.cpbenchy.com/llms.txt
> Use this file to discover all available pages before exploring further.

# From Python

1. **Run**

   ```python
   import cpbenchy

   results = cpbenchy.run(
       "data/opb/",                  # files, directories, glob patterns or a CPMpy dataset
       solvers=["ortools", "exact"],
       time_limit=60,
       out="results/opb",
   )
   ```

   It takes the same options as [the command line](/getting-started/first-benchmark/), and stores the
   results in `out` as they come in. Running the script again only does the runs that are missing.
2. **Analyse**

   ```python
   df = results.to_pandas()              # one row per run
   df.groupby("solver").solved.sum()     # solved instances per solver
   ```

   Later, `cpbenchy.load("results/opb")` gives the stored results back.

## Follow the runs

Pass a function to see each result as its run finishes:

```python
cpbenchy.run("data/opb/", solvers=["ortools"], time_limit=60, on_result=print)
```

## Next steps

- **[Results](/guides/results/)**: what each result holds, and more ways to analyse them.
- **[Experiments](/guides/experiments/)**: different settings per solver, results in a loop, and
  callbacks for each solution found.
- **The [library](/library/)**: ready-made extras to pass with `plugins=`, such as checking every
  solution.
- **[Extending cpbenchy](/plugins/overview/)**: record your own measurements with an observer, or benchmark your own formats with a loader.
- The [Python API reference](/reference/python/) lists every argument.

Source: https://docs.cpbenchy.com/getting-started/python/index.mdx