MSDS Time Estimator

Bayesian time estimator for CU Boulder MSDS courses, calibrated against student reviews.

What it does

• Predicts calendar days for a target course from one or more anchor courses you have already finished.
• Uses crowdsourced CU Boulder MSDS review data, then updates course estimates with a tempered Bayesian weight.
• Adjusts for focus ratio, ADHD profile, and self-rated math/coding/writing strengths.
• Returns uncertainty-aware outputs, including per-course standard deviations and 80% intervals.

Quick start

git clone <YOUR_GITHUB_REPO_URL>
cd msds-time-estimator
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
python -m modules.form_intake
python example_run.py

What those commands do:

1. install the dependencies
2. refresh data_points.csv from the live sheet
3. run an example profile and print predictions

If you prefer the notebook flow instead, open time_estimator.ipynb after install and run the profile/prediction cells there.

If you are not technical, start with the nontechnical guide.

How it works (short)

The model anchors on hours, not just calendar days: your anchor-course days and hours/week are converted into study hours, then compared against each course's posterior mean hours to estimate your pace. Course posteriors are updated from student review data with a tempered effective weight of min(N, 5) so high-review courses tighten confidence without completely drowning the prior. For courses that still do not have student review rows, the intake step can fall back to Coursera-published hours from the same MSDS review sheet. Course predictions transfer anchor pace by topic similarity, and degree-total uncertainty uses the same course-topic matrix for CLT correlations. See MODEL.md for the full derivation and implementation details.

Data sources

Live MSDS Reviews sheet:

• https://docs.google.com/spreadsheets/d/1lplPW_5DI-wgB_q6qgxmr9WTP12-JVV_yDXsKxLFMiM/edit?gid=158871878#gid=158871878

python -m modules.form_intake pulls from that sheet's published MSDS Reviews tab and refreshes the local review observations automatically.

Customizing for your own use

The fastest way to customize the example CLI run is to edit the common = dict(...) block in example_run.py.

Key knobs:

• anchor course and anchor duration: anchor_courses, anchor_days_list
• anchor and target schedule: anchor_hours_per_week, target_hours_per_week
• focus ratio and ADHD flags: user_focus_ratio, has_adhd, medicated
• self-ratings: skill_priors=(math, coding, writing) on a 1-10 scale

If you want to call the model directly, use modules.pipeline.build_profile(...) and modules.pipeline.build_predictions(...).

Simplified model alternative

For a leaner alternative to the full Bayesian/topic-transfer pipeline, use modules.simplified_predictor. It estimates baseline course hours from the guide difficulty weighted-OLS fit when a course has difficulty, falls back to expected_hours otherwise, multiplies by your median completed-course ratio, then converts predicted hours into calendar days from your study schedule.

from modules import simplified_predictor

params = simplified_predictor.load_course_params()
prediction = simplified_predictor.predict_course_from_history(
    "DTSA5501",
    [{"course_id": "DTSA5001", "actual_hours": 40.0}],
    params,
    hours_per_week=10.0,
)

If you prefer the notebook, edit the profile input cell or switch back to pipeline.prompt_profile() for an interactive prompt.

Contributing

Add your own course review here:

• http://tinyurl.com/cu-elective-review

More reviews make the posterior estimates better.

Limitations

• Some courses still rely on Coursera-published hours rather than student-reviewed hour averages.
• Focus-ratio constants are heuristic, not individually fitted.
• No cross-validation or held-out calibration study yet.

License

MIT. See LICENSE.