CSci574 / AI 520 Machine Learning: Classroom Session 08/25/2026
Overview
Derek Harter explains the setup and workflow for Graduate Machine Learning, including GitHub organization access, SSH authentication, Docker-based Python environments, VS Code Dev Containers, and Jupyter notebooks. He walks through Assignment 1, which covers scalar and NumPy implementations of the sigmoid function, vectorization, feature scaling, and error metrics, while outlining weekly deadlines, commit expectations, and a final data-analysis project.
Key takeaways
- Accepting the CSCI 574 GitHub organization invitation is a prerequisite for accessing and accepting individual assignment repositories.
- Using the provided Docker container aligns Python and scientific-library versions with the autograder, while the /opt/base virtual environment is the intended notebook kernel.
- Assignment notebooks must execute sequentially from top to bottom because the grader reruns every cell, not just the tests for a completed function.
- Assignment 1 expects students to commit work at least once per task, with roughly six or seven commits reflecting progress across its subtasks.
- The sigmoid exercises contrast scalar computation with math.exp and array-based computation with NumPy vectorization.
- The final project asks students to select a dataset and apply a reproducible workflow that can include cleaning, visualization, regression or classification, and model improvement.
Chapters
- Derek Harter checks screen sharing and audio with an online student before beginning the classroom session.
- He plans to record class meetings and post them for students who cannot attend at the scheduled time.
- Class resources include a playlist of lecture videos and a separate playlist for recorded classroom sessions.
- Assignments are due Sundays by midnight, approximately every one to two weeks, with grading planned for Mondays.
- Assignment 1 is due that Sunday, after roughly a week and a half of preparation time.
- Students must accept the invitation to the CSCI 574 machine-learning GitHub organization before accepting assignments.
- The instructor asks students to join promptly and contact him if an invitation is missing or has expired.
- Organization membership lets students access their own repositories and assignment templates, while solutions remain restricted.
- Students need a GitHub account, Git tools, SSH, and a working Python environment; an existing GitHub account is acceptable.
- Docker is recommended because its containers use expected versions of Python, NumPy, and scikit-learn for the autograder.
- VS Code is recommended for Git integration and Docker-based development, but neither VS Code nor Docker is mandatory.
- Generate an SSH key with ssh-keygen and add its public key to GitHub's authentication-key settings.
- Test authentication before cloning; the first SSH connection may prompt students to trust GitHub's host signature.
- Accepting the first-host prompt adds the site's signature to known hosts and avoids a hidden confirmation blocking a VS Code clone.
- The public CSCI 574 class-resources repository can be cloned using HTTPS because students have read-only access.
- The repository contains Python notebooks for course units 1 through 11, including Python versions of examples originally presented in Octave.
- Course pages also link to lecture playlists, notebooks, and supplementary machine-learning materials.
- The Microsoft Dev Containers extension lets VS Code edit and run code inside a local Docker container.
- On Windows or macOS, Docker Desktop must be running; the first container build may take time to download components.
- The course containers use Debian Linux and include tools and packages such as Python, NumPy, and scikit-learn.
- Students are expected to be comfortable programming in a high-level language and able to pick up Python syntax.
- Basic calculus, linear algebra, and probability are useful background, with derivatives and linear algebra especially relevant to neural networks.
- Course topics include data exploration and cleaning, supervised learning, linear and logistic regression, support vector machines, decision trees, and random forests.
- The course has seven or eight regular assignments plus an open-ended final project due at the end of the course.
- Students choose a dataset and apply course methods, potentially including data cleaning, visualization, regression or classification, and model improvement.
- A midterm checkpoint requires students to choose a dataset and create a repository; suggested project folders include data, notebooks, and source code.
- The GitHub Classroom invitation creates an individual repository under the course organization, named with the course, assignment, and student's GitHub username.
- Students can also locate their assignment repository through their GitHub organization if the invitation link is no longer available.
- Assignment instructions are in the IPython notebook, and assignment repositories should be cloned with SSH to enable pushing work.
- VS Code's lower-left status identifies the active Dev Container, and its terminal shows that commands run inside the container.
- The local project directory is mounted into the container, so file edits made on the host and inside the container are shared.
- The container starts JupyterHub on port 8888 and forwards it to the host, allowing notebooks to run in a web browser.
- Assignment notebooks are expected to run sequentially from top to bottom; the grader reruns every cell and deducts points if execution stops.
- The container creates a Python virtual environment under /opt/base and installs packages using requirements.txt.
- When running notebooks in VS Code, select the /opt/base Python kernel so the notebook uses the intended environment and libraries.
- Choose the SSH repository URL when cloning an assignment; VS Code may otherwise default to HTTPS.
- Git commits provide both a backup and a way to restore work by recloning the latest pushed state.
- Make at least one commit per task, with roughly six or seven commits expected for the subtasks in Assignment 1.
- Assignment 1 asks students to implement basic machine-learning functions, beginning with a scalar sigmoid using math.exp rather than NumPy.
- Implement functions in the assignment's source/assignment task code, then use the notebook's pytest checks to verify behavior.
- Tests check expected outputs such as sigmoid(0) = 0.5; a passing example alone is not enough if other test cases still fail.
- The notebook's auto-reload setup can load changes from the source file without manually restarting the import, and the test panel can show passing checks.
- NumPy vectorization applies operations such as exponentiation to an array of 100 values in one operation instead of using explicit Python loops.
- The scalar math.exp function accepts individual values, while NumPy operations can work across arrays.
- Feature scaling adjusts values to a common mean and standard deviation, a preprocessing step that can be important before model training.
- Assignment functions for mean absolute error and root mean squared error preview cost metrics used with regression models.
Transcript excerpt
0:29 Um yeah, give me a moment. I'm still setting stuff up here for the people that are joining me online. Uh See if I can change my display settings. Ta-da. No, screen at least.
2:15 >> Okay, getting there. Um, so uh So, uh can you see my screen? Am I sharing that? So, I'm not 100% certain. I think I am. For my Zoom caller here. >> Yes, Professor. >> Okay, great. Thank you. You can hear me, sounds like. That's good. Right. Oh,
3:16 okay. Let me make certain that Okay, okay. Sorry, so I had a little problem getting in the room. Uh I think I'm ready to go. Um, so we don't have too many people here yet. So, I'm going to repeat a few of the things I did on Thursday. I actually had a few new people join the class since then. Uh since we met kind of for the first time. So, I've gotten not a whole lot of students, so I'm kind of looking for feedback on what to do with the face-to-face stuff. I'm going to be doing the Zoom and coming into these, uh, to the classroom that we have
4:04 scheduled, for a while till I figure out, whether we want to keep doing that or not. So, and I'm am going to be recording these, uh, class sessions and putting those online if you're not, you know, if you're an online student not available during this time. The first one should be up there. So, hopefully everybody's able to get into the course and has been reading the announcements. So, you should find links to the, uh, the regular playlist of a bunch of videos that I already have for the lecture topics and things. Plus another playlist that I'll be putting these,
This is the start of the transcript. Save this video free for the whole transcript, searchable and time-coded.
In this course
Related notes
Summary, takeaways, and chapters were generated by AI from the video's transcript and may contain errors. The video belongs to its creator, Derek Harter.