ortidy 0.5.0

Operations research on tidy dataframes — a backend-agnostic dataframe façade over Google OR-Tools.

ortidy

OR (operations research) on tidy dataframes — a backend-agnostic dataframe façade over Google OR-Tools.

ortidy is the revival of pandas-or, rebuilt around Narwhals so it works natively with pandas, Polars, and more — pandas in, pandas out; Polars in, Polars out. pandas-or users: see Migrating.

The thesis: bridge operations research and the data ecosystem so solver outputs come back as tidy dataframes ready for analysis, plotting, and dashboards.

Installation

pip install ortidy

ortidy brings its own OR-Tools. Bring your own dataframe backend (pandas, polars, …) — install whichever you already use.

Quickstart

Every solver accepts a native frame, returns the same backend, and hands back a SolveResult carrying the result frame, a status, the objective, and solve metadata.

import pandas as pd
import ortidy

items = pd.DataFrame({"value": [60, 100, 120], "weight": [10, 20, 30]})

result = ortidy.knapsack(items, capacity=50)
print(result.status)        # SolveStatus.OPTIMAL
print(result.objective)     # 220
print(result.frame)         # items + an `isIncluded` boolean column

Every solver is a plain function — pass a native frame (pandas, Polars, …) and get a SolveResult back in the same backend.

Result contract

Three result shapes cover every solver:

Shape	Solvers	Result
selection	knapsack, multi_knapsack, bin_packing	rows mapped to bins/resources
edge-flow	solve_routing (more in progress)	values on an edge list
interval-schedule	scheduling (roadmap)	intervals on a timeline

Every solver returns a SolveResult:

result.frame       # native dataframe, same backend as the input
result.status      # OPTIMAL / FEASIBLE / INFEASIBLE / UNBOUNDED / MODEL_INVALID
result.objective   # objective value (None when no solution)
result.metadata    # solver name, wall time, bounds, …
bool(result)       # True when OPTIMAL or FEASIBLE — a FEASIBLE solution is a success

Solvers

Shape	Function	Problem
selection	knapsack	0/1 & multidimensional knapsack
selection	multi_knapsack	multiple knapsack
selection	bin_packing	bin packing (minimize bins)
selection	assignment	linear assignment (+ teams / allowed groups)
selection	generalized_assignment	GAP (capacity-limited agents)
selection	facility_location	uncapacitated facility location
selection	set_cover	set cover / partition
selection	blend	diet / blending LP (continuous quantities)
edge-flow	max_flow / min_cost_flow / shortest_path	network flow
edge-flow	transportation	transportation problem
edge-flow	solve_routing (+ distance_matrix)	routing (TSP/VRP/CVRP/VRPTW/pickups/optional visits)
interval-schedule	shift_scheduling	employee rostering
interval-schedule	job_shop	job-shop scheduling

See the docs for a worked example of each.

Sample datasets ship with the package:

from ortidy import data
items = data.items_knapsack()            # pandas by default
items = data.items_knapsack("polars")    # or Polars

Examples

Runnable notebooks live in examples/ — knapsack/bin-packing, a haversine world-tour TSP, and capacitated vehicle routing, each plotted with Plotly. Try them in your browser with no install:

Migrating from pandas-or

ortidy is pandas-or renamed and rebuilt. Install ortidy and import ortidy (the old pandas-or 0.1.3 on PyPI is unmaintained). The key changes:

• Solvers return a SolveResult object, not a bare frame / None.
• Rows are identified by explicit id columns, never a positional index.
• The backend is whatever you pass in (pandas, Polars, …), not just pandas.

License

Apache-2.0.