Graduate Visit Scheduler

MILP-based scheduling utilities for graduate visit days. Created by Alex Dowling and Jeff Kantor at the University of Notre Dame.

Motivation

Most chemical engineering graduate programs in the U.S. host admitted or prospective students for a visit or open house. A core part of those events is one-on-one or small-group meetings with individual faculty members. Building a fair, feasible schedule is challenging: faculty availability, room locations, and student preferences must all be respected simultaneously.

This package is the meeting scheduler used by Notre Dame Chemical and Biomolecular Engineering, released as a general-purpose, open-source tool. Under the hood, it formulates a mixed-integer linear program (MILP) in Pyomo, solves it with a standard solver, visualizes the results, and can export customized student schedules to DOCX.

Data Gathering

Each visitor is asked to rank up to four faculty members for meetings and to choose up to two topic areas within the department. Likewise, faculty are asked to provide any afternoon meeting conflicts. This information is used to generate three key input files:

• A faculty catalog YAML file (names, buildings, rooms, areas, status, and optional aliases).
• A run configuration YAML file (time slots by building, breaks, area weights, and any faculty availability constraints).
• A visitor preferences CSV file.

Research areas are department-specific. In the included example data, they are represented as simple labels such as Area1 and Area2, but you can define any set of areas in your faculty catalog. Time slots should be written with explicit AM/PM labels or 24-hour times such as 1:00 PM-1:25 PM or 13:00-13:25. Bare labels like 1:00-1:25 still work via a visit-day heuristic, but they emit a warning because they are ambiguous.

What It Produces

• An optimal assignment of visitors to faculty across time slots.
• Visual summaries (plots) of the resulting schedule.
• Optional DOCX exports with individualized visitor schedules.

Install

pip install grad-visitor-scheduler

Solver setup:

• HiGHS is installed by default via the highspy dependency.
• To use CBC, install the solver binary with conda:

conda install -c conda-forge coincbc

CI and Coverage

GitHub Actions runs the test suite on pushes to main and on pull requests targeting main. Coverage is uploaded to Codecov from the Python 3.11 job. For this public repository, coverage upload uses GitHub OIDC (no CODECOV_TOKEN secret required).

Automated PyPI Releases

Releases are tag-driven via GitHub Actions:

• Push a semantic version tag like v0.2.0.
• Workflow .github/workflows/release.yml runs tests, builds sdist/wheel, and checks package metadata.
• Package is published to TestPyPI first.
• CI performs a smoke install of that exact tagged version from TestPyPI.
• If smoke install passes, CI publishes the same artifact to PyPI.

Publishing uses PyPI Trusted Publishing (OIDC), so no API token secret is required. Configure both PyPI and TestPyPI trusted publishers to point to this repository and the release.yml workflow file. Detailed release instructions are in docs/releasing.md.

Manual release workflow runs are also supported (workflow_dispatch):

• target=testpypi: build, publish to TestPyPI, smoke install, stop.
• target=pypi: run the full pipeline (TestPyPI + smoke + PyPI publish).
• Optional version input lets smoke install pin a specific version.

Quickstart

from pathlib import Path
from grad_visit_scheduler import scheduler_from_configs, Solver

root = Path("examples")

s = scheduler_from_configs(
    root / "faculty_formulation.yaml",
    root / "config_two_buildings_close.yaml",
    root / "data_formulation_visitors.csv",
    solver=Solver.HIGHS,
)

sol = s.schedule_visitors(
    group_penalty=0.2,
    min_visitors=2,
    max_visitors=8,
    min_faculty=1,
    max_group=2,
    faculty_breaks=1,
    student_breaks=1,
    tee=False,
    run_name="demo",
)

if sol is not None:
    sol.plot_visitor_schedule(save_files=True, show_solution_rank=False)
    sol.plot_faculty_schedule(save_files=True, show_solution_rank=False)
    sol.export_visitor_docx("visitor_schedule.docx")
else:
    print(s.infeasibility_report())

Note: the examples/ folder referenced above is included in the repository, but it is not packaged on PyPI. If you installed from PyPI, clone the repo to access the example files.

All shipped example configs now use explicit AM/PM slot labels to avoid ambiguity warnings.

Buildings and Movement

Notre Dame CBE is split across two buildings (Nieuwland Science Hall and McCourtney Hall) separated by about a 7-minute walk. A key aspect of the scheduler is to ensure that any visitor who needs to move buildings does so during their break slot. A typical schedule uses six meeting slots, and each visitor and faculty member gets at least one middle-slot break.

The run config now supports one, two, or many buildings. building_order controls plotting/export ordering. Movement behavior is configured with the movement section:

• movement.policy: none: close-proximity buildings; no explicit travel-time constraints.
• movement.policy: travel_time: explicit inter-building travel-time lag constraints.
• movement.policy: nonoverlap_time: auto-derive inter-building lag constraints from absolute slot timestamps.
• movement.phase_slot: earliest slot allowed by building (for staggered starts like Building A first vs Building B first).
• movement.travel_slots: auto: auto-derive lag matrix under travel_time for shifted/nonuniform clocks.

Important shifted-clock note: if buildings have offset time grids, policy: none can allow real-time overlaps across adjacent slot indices. Prefer policy: nonoverlap_time (or policy: travel_time with travel_slots: auto) for those schedules.

Legacy Mode.* options are still available with FutureWarning.

See docs/movement.md and scripts/run_building_configuration_examples.py for full examples. For common configuration mistakes and fixes, see the Failure Modes and Diagnostics section in docs/movement.md.

Refine the Schedule

The solver exposes several tunable parameters on schedule_visitors to refine the schedule:

• group_penalty: penalize group meetings to bias toward one-on-one meetings; higher values discourage multi-visitor meetings.
• min_visitors: minimum number of visitors each available faculty member must meet.
• max_visitors: maximum number of visitors each faculty member may meet.
• min_faculty: minimum number of faculty each visitor must meet.
• max_group: maximum size of a meeting group at any time slot.
• faculty_breaks: minimum automatic faculty breaks. Faculty-unavailable slots outside the break window count toward this total.
• student_breaks: minimum automatic visitor breaks within the configured break-option slots.
• enforce_breaks: legacy compatibility alias for setting both counts to 0 or 1 together.
• debug_infeasible: if True, build the model before raising pre-solve contradiction errors (useful for IIS/manual inspection of s.model).
• tee: print solver output for debugging.
• run_name: label used when saving plots/exports.

For advanced hard constraints, per-entity bounds, pre-solve checks, and debug infeasibility workflows, see docs/advanced_customization.md.

To generate multiple ranked schedules, use no-good cuts:

top = s.schedule_visitors_top_n(n_solutions=3, faculty_breaks=2, student_breaks=1)
report = top.summarize(ranks_to_plot=(1, 2), show_solution_rank=True)
print(report["summary"])
print(report["compact"])

Export DOCX

Preferred modern path:

sol = s.schedule_visitors(...)
if sol is not None:
    sol.export_visitor_docx("visitor_schedule.docx")

Legacy helper grad_visit_scheduler.export_visitor_docx(...) remains available for compatibility but emits FutureWarning.

License

This software is released under the BSD-3-Clause license. Please adapt it to your needs and share.