UW MLGEO class final project

The course is project-based, with a progression throughout the quarter. The final project integrates the three technical pillars of the course — an AI-ready data set, classic machine learning, and deep learning — together with the fourth layer that runs through the 2026 edition, working with agentic AI, applied as a group to a scientific problem. The deliverables are:

• A GitHub repository that demonstrates data manipulation, algorithm design and testing, and evaluation.
• A 5-page report that places the project in context and discusses the results, in one of the two genres below.
• A 10-15 minute presentation.

Groups submit each deliverable together. This page is the rubric used both for self-evaluation and for the instructors’ evaluation.

Students are expected to use AI assistants during the project, with disclosure (see the course AI-use policy). Before submission, each group runs an agentic AI review of their repository, then critically assesses the AI’s assessment and documents at least one thing the AI got wrong or missed. Both the AI review and the group’s critique are submitted with the project.

Weights: Report 35%, GitHub repository 30%, Presentation 35%. Team contribution, assessed from the CRediT statement and peer evaluations, applies a modifier of up to ±5 points to an individual’s total.

Choose a genre for the report¶

Each group picks one of two genres for the written report. The choice changes only the Content & Research Quality criterion; every other criterion, and the repository and presentation rubrics, are common to both tracks.

• Research-paper track: the report reads as a short scientific paper — research question, data, methods, results, generalization, outlook. Graded with criterion 1a.
• Stakeholder-deliverable track: the report is a technical deliverable written for a named decision-maker (a water utility, an emergency manager, a park service, a mission planner). Graded with criterion 1b.

1. Report (5 pages) — 35%¶

• 1a. Content & Research Quality (25%) — research-paper track Quality, depth, and relevance of the research

  • Present a clear statement of an outstanding research question and place it in the context of an up-to-date literature review
  • Demonstrate the originality of the research
  • Present an AI-ready data set (see the operational definition in Chapter 2.13), preliminary analysis with correlations or descriptions of basic data features, and discuss potential data imbalance within the context of the stated problem
  • Demonstrate a strong understanding of classic machine learning and/or deep learning with an example
  • Discuss performance evaluation in the context of training the model and of generalization beyond the data and domain presented in the report
  • Discuss the computational time for training and for inference
  • Discuss the appropriateness of the computational resources used

• 1b. Content & Research Quality (25%) — stakeholder-deliverable track Fitness of the deliverable for a real decision

  • Name the decision-maker and the decision the deliverable informs, and state the operational question in their terms, grounded in a short review of current practice
  • Present an AI-ready data set (Chapter 2.13) and preliminary analysis, as in the research-paper track
  • Demonstrate a strong understanding of classic machine learning and/or deep learning with an example, evaluated against a domain baseline
  • Fitness for the named decision-maker: results reported in the units, thresholds, and lead times the decision requires
  • Honest operating limits: the conditions, regions, and seasons where the model can be trusted, and those where it cannot
  • Calibrated-uncertainty statement: predictions carry uncertainty the decision-maker can act on, with a calibration check to back it
  • Discuss the computational time for training and inference and the appropriateness of the resources used

• Structure and Organization (15%) Coherence, clarity, and flow of the report

  • Logical structure:
    • Introduction with scientific background to motivate the project and the state of the data (e.g., we are missing measurement X and propose to predict it from Y)
    • Data: what is a data sample, what are the features and dimensions, exploratory data analysis; for time series forecasting, how the data is organized for training; train-validation-test split, with a discussion of whether the test data is in-domain or out-of-domain
    • Models: design and training strategies (for deep learning: optimizer, learning rate, batch size, learning curves)
    • Generalization statement: discuss test performance and what it implies for use beyond the presented data
    • Results: what did you learn, how long does training take, how long does inference take, how would this scale to a larger data mining project
    • Conclusions and outlook, references
  • Well-organized paragraphs with transitions between ideas
  • Figures have captions and clear labels and are referenced in the text

• Clarity and Writing Style (15%) Quality and effectiveness of the writing

  • Clear, concise language; minimal grammar or spelling errors
  • Professional and academic tone appropriate for the field

• Critical Thinking & Analysis (15%) Depth of analysis and reflection

  • The analysis goes beyond description: what have you learned, what are the advantages and limitations of your approaches, what can ML methods contribute to the overall research question
  • Acknowledges alternative perspectives or limitations
  • Demonstrates original thinking and critical engagement with the research

• Communication deliverables (20%) Taught in Chapter 7

  • Audience translation: the report includes two short summaries (a paragraph each) of the same central result, written for two different audiences — for example a program manager and a field scientist, or a city planner and a journal reviewer. The two summaries must be faithful to the same numbers.
  • Downstream-impact statement: half a page on the societal, environmental, or economic relevance of the project. Who could use this result, what decision could it inform, and what could go wrong if the model is trusted beyond its domain?

• Formatting, Citations & Disclosure (10%)

  • Follows formatting requirements (margins, font, length); correct use of a citation style (e.g., APA, MLA, Chicago)
  • CRediT authorship statement: every student is associated with specific contributions
  • AI-use disclosure: which AI tools were used, for what tasks, and what the group verified themselves

2. GitHub Repository — 30%¶

• Code Quality (15%) Quality and functionality of the scripts and code

  • The code is clean, well-documented, and follows good programming practices
  • All scripts run without errors when the environment is set up
  • Code is modular, with reusable functions where appropriate
  • Scripts achieve the intended outcomes (generating plots, performing analysis)

• Reproducibility (25%) Ease of reproducing the analysis and results

  • The repository includes clear instructions (in a README.md) for setting up the environment and running the code
  • Jupyter notebooks, scripts, and any other files necessary to recreate the analysis are provided
  • Data, or instructions to access the data, are included or referenced appropriately
  • Outputs (plots, tables) are reproducible from the code

• Organization & Structure (15%) Organization and clarity of the repository

  • Clear folder structure (separate folders for code, data, results)
  • File and folder names are descriptive and intuitive
  • The README provides a clear overview of the repository and how to find things

• Documentation (15%) Clarity and completeness of documentation

  • README explains the project, dependencies, and setup instructions
  • Code and notebooks are documented, with comments explaining key sections
  • Functions have docstrings

• Environment Setup (10%) Dependency management

  • Includes a complete, working environment specification (pixi.toml with pixi.lock, or environment.yml/requirements.txt)
  • The environment lists all necessary dependencies with correct versions
  • Setup instructions are clear and easy to follow

• Version Control Practices (10%) Effective use of Git and GitHub

  • Commits are frequent, descriptive, and reflect the progression of the project; commit messages disclose substantial AI assistance
  • Clear use of branches where applicable
  • Issues, pull requests, or other GitHub collaboration tools are used effectively

• AI review and critique (submitted alongside the repository, graded within Reproducibility and Documentation)

  • The output of an agentic AI review of the repository
  • The group’s written critique of that review, documenting at least one error or omission in the AI’s assessment

3. Presentation (10-15 minutes) — 35%¶

• Content (25%) Quality and depth of research

  • A clear statement of an outstanding research question, supported by references to the scientific literature
  • Present the AI-ready data set and preliminary analysis
  • Demonstrate a classic machine learning example and a deep learning example
  • Report the computational time for training and inference, and the appropriateness of the resources used

• Structure and Organization (15%) Coherence and flow

  • Logical structure (intro, body, conclusion)
  • Key points are clearly distinguished and emphasized

• Clarity and Delivery (20%) Effectiveness in communicating ideas

  • Clear articulation, adequate volume and pace
  • Minimal reliance on notes, eye contact with the audience
  • Confident and professional demeanor

• Visual Aids (10%) Supporting materials (slides, charts, plots)

  • Visual aids support rather than distract from the presentation
  • Information is presented clearly and uses appropriate design principles
  • Slides do not overwhelm with text or complex visuals

• Engagement and Interaction (10%)

  • Encourages audience interaction through questions or participation
  • Responds effectively to audience questions and comments

• Critical Thinking and Analysis (15%)

  • Demonstrates original thinking and critical engagement with the research
  • Identifies limitations and future directions

• Professionalism (5%)

  • Respect for time limits, prepared with material, respectful manner

4. Team contribution modifier (±5 points)¶

Individual grades can be adjusted by up to ±5 points based on the CRediT statement, peer evaluations, and the commit history. Every enrolled student must appear in the CRediT statement with specific, verifiable contributions. A student who cannot explain the parts of the project attributed to them should expect the instructor to ask them to walk through the code unaided.