Rather than an exam, the cumulative evaluation in PIC16B is a course project. The purpose of the project is for you to demonstrate your ability to independently identify and solve complex problems using Python.
This project is open-ended, and should match your interests and goals. You are free to work on a project that is connected to other work you are doing, in another class, research, an internship, or a job. However, your project must be original work – i.e. no submitting code that you already wrote.
You have considerable freedom to set the topic and goal of your project, subject to my approval. There are a few requirements. If you aren’t sure about whether your project idea fulfills these requirements, please approach me as early as possible to discuss.
Use Tools Beyond PIC16A
Part of the purpose of the project is for you to demonstrate what you learned during your time in PIC16B. For this reason, you are required to demonstrate tools from the course. A project that uses basic
pandas and decision trees to analyze a data set won’t receive full credit – that’s PIC16A stuff. Some possibilities:
- Use TensorFlow and/or OpenCV for image classification or language modeling in a data set that you scraped yourself.
- Perform a detailed analysis and visualization of a large data set, using SQL to access the data.
- Perform and visualize a complex simulation by solving partial differential equations. Use boundary conditions drawn from real data.
- Analyze a networked data set using
networkxand other tools.
- Write an app to visualize different kinds of dynamical processes (like disease spread or opinion dynamics) evolving on networks.
- Use webscraping and database skills to access and consolidate a range of disparate data sources into a new, publicly available data set.
You might wish to use tools from later in the course in your project. In this case, you’ll need to learn these tools early! You can approach me as a group, and we will discuss a way to ensure that you are able to learn the needed content in time to work on your project.
You are allowed and even encouraged to additionally use tools not covered in PIC16B. However, you must still incorporate some content from PIC16B as well in order to receive full credit.
A common pattern on machine learning websites such as Kaggle is:
- Acquire cleaned data.
- Fit several machine learning models to the data.
- Assess performance and report test score.
This is not enough for a successful course project. While you are welcome to do a machine learning project, I expect more than the above for a quarter-long effort. A project whose primary punchline is “we got 93% test accuracy on a public data set” is likely to be an A- project at best.
Here’s how you can deepen your project:
- Perform a significant data acquisition task. Obtain, clean several novel data sets on which to base your analysis.
- Perform a detailed analysis of model error and bias. Does your model have the same performance on different subgroups of the data? Why is that? Can you adjust your model to reduce bias?
- Construct complex visualizations to help your audience understand why your model works on the data, and what we can learn from the data using the model.
Is Our Project Good Enough?
When planning what you want to achieve with your project, I encourage to keep in mind two simple questions:
Will I learn something by completing this project?
Will I be proud of this project once it’s done?
If the answer to both questions is yes, then you likely have a good project idea on your hands. Feel free to talk to me (early!) if you aren’t sure whether your project idea is appropriate for the course.
Here is a continuously-updated list of project assignments. Each of these has their own descriptions and specs. The due date is shown. Note: this is exactly the same list as the one here.
There are also a few requirements that apply to your final project submission, which you may wish to keep in mind early on.
Format and Structure
- Your project must be hosted on GitHub as a version-controlled repository.
- Functions and classes should not be defined in notebooks. Rather, your project must include functions and classes defined in two or more
.pyfiles, which must in turn be packaged as a module that can be imported. To use your functions and classes in a notebook, you will import these functions and classes using
- Code repetition has been minimized by efficient use of functions and classes.
- Your provisional and final project submissions must include a blog post that clearly demonstrates how to use your project tools. Your code and explanation must be sufficiently careful that a user could clone your repository and run exactly the code demonstrated on your blog. It’s fine for group members to collaborate on a single blog post and post the same one to each of their blogs.
- All your source code must include thorough comments and comprehensive docstrings on functions, classes, and methods.
- In your project’s GitHub repository (not your blog post), you should have a top-level
README.mdfile that includes a Group Contributions Statement. Your group contributions statement should make clear the primary ways in which each group member contributed to the project.
- I expect all group members to contribute to the project repository. The commit history of the project repository is public, and so I will be checking to ensure that everyone has made commits. If I see that only one group member has been committing code, I will have questions.
Generally speaking, all group members will receive the same project grade. The only exceptions are if I see a highly inequitable Group Contributions Statement or GitHub commit history.