Class project
← Back to course home · Schedule · Syllabus · Policies
Beginning in the second week of class, groups of about four students work together on a machine learning project throughout the semester, using one of several real-world public policy problems. Each week, every group submits a project assignment. In addition to connecting the readings and discussion topics to the policy domain, these updates give you a chance to get input and feedback and iterate.
Project Descriptions
This semester, we have set up two projects
- Reducing Jail Rebookings through Proactive Mental Health Outreach
- Supporting Advocacy for Civil Rights by Prioritizing State Bills that are likely to Pass
⚠️ Data security. Project data is sensitive and must remain in the secure computing environment provided for the course. See the data security policy — violations result in automatic failure of the class.
Deliverables at a glance
| # | Deliverable | Due | Weight |
|---|---|---|---|
| 1 | Project proposal | Tue, Sep 15 | 10% |
| 2 | Peer reviews of three proposals | Tue, Sep 22 | 5% |
| 3 | Weekly progress updates | Most Tuesdays | 25% |
| 4 | Final presentation | Thu, Dec 3 | 10% |
| 5 | Final report and code | Wed, Dec 9 | 25% |
20% of your grade is class attendance and participation and 5% is weekly feedback forms — see policies.
1. Project proposal
Submitted as a group, 4–5 pages not including figures, tables, or references. It should cover the project scope and preliminary descriptive statistics about the entities in your dataset, and include the provided scoping worksheet as an appendix.
Useful reference: the Data Science Project Scoping Guide.
2. Proposal peer reviews
Each student reviews three other groups' proposals. Reviews should be constructive and specific: is the problem well scoped, is the proposed formulation the right one, what would you imorove?
3. Weekly Project Assignments
Short update assignments, generally due Tuesdat, that guide the check-in discussions. These typically take the form of filling results or modeling details into a handful of template slides. Over the semester these build up the technical core of the project:
- Analytical formulation, baselines, and cohort/label queries (week 6)
- Modeling plan and temporal validation configuration (week 8) — how the scope in your proposal is formulated as an ML problem, and the elements of the pipeline your group will build
- V0 baseline results and planned feature list (week 9) — the specific features to be built, the underlying data, the level at which information is available, aggregation strategies over time or geography, and your plan for handling missing values
- V0 modeling results (week 10)
- Interpretability results (week 11)
- Feature importances and crosstabs (week 12)
- Bias and disparity results (week 13)
Updates are graded for completeness and correctness. We expect this work to be iterative: errors identified in one week's update that are corrected by the next week results in revision of the previous score up to 80% of the total possible.
4. Final presentation
15 minutes plus 3 minutes for questions. The final presentation should be geared toward the relevant decision makers for your project: an overview of the problem and approach, your results, policy recommendations, and limitations of the work.
6. Final report
Approximately 10 pages, accompanying the final presentation. It should include:
- An executive summary, no more than one page, succinctly describing the project, results, and recommendations.
- An overview of the problem, its significance, and the scope and goals of the work.
- A description of the methodology and results, with a link to well-documented code in your group's course GitHub repository.
- A brief (1–2 paragraph) design of a field trial to evaluate the accuracy of the resulting model in practice, as well as its ability to help the organization achieve its goals.
- Concluding lessons and recommendations for the partner organization.
- Optionally, a proposal for future work beyond the scope of this project — novel ML methods that could improve on the current work, new policy interventions to evaluate, or related research opportunities.
Project iteration targets
The point of the weekly cadence is to have a working end-to-end system early and improve it, rather than building the pieces separately and integrating at the end.
| Iteration | Weeks | Focus |
|---|---|---|
| 1 | 5–6 | End-to-end shell — the simplest possible full pipeline |
| 2 | 7–8 | Feature development |
| 3 | 9–10 | Models and evaluation |
| 4 | 11–12 | Interpreting the models |
| Final | 13–14 | Final model choice, disparities, impact |
Templates and worksheets
Scoping worksheets, update slide templates, and report templates are in the project/ directory.