Milestone map
Milestone map
2 milestones
Do the course work in public
2–3 weeks (10 hrs/week)
Take Advanced Learning Algorithms for free and do lessons 1, 3, 5 and 6: a neural network built from scratch and a random forest, each on real data. Keep everything in one public repo or notebook as you go. OSSU lists a DeepLearning.AI course on Coursera, where graded work is paid. We use fast.ai's free course, lessons 1, 3, 5 and 6 (neural networks from scratch, then random forests).
Proof required
Link your public repo or notebook with lessons 1, 3, 5 and 6: a neural network built from scratch and a random forest, each on real data, and a screenshot or log showing the self-check passed.
What gets checked
- Each notebook runs top to bottom.
- The from-scratch network matches a library model's results.
- Validation scores are reported, not training scores.
Common mistakes
- Copying posted answers: verifiers read the history.
- Skipping the self-check, then asking for review too early.
Powstik Guide
A public repo or notebook with lessons 1, 3, 5 and 6: a neural network built from scratch and a random forest, each on real data.
Steps
- Make a public repo or notebook first.
- Do one unit at a time and commit after each.
- Run the self-check: Run each lesson notebook, then answer the chapter questionnaire in the free book before you ask for review.
- Write a short README: what is here and how to run it.
- Ask three people who know the topic to check it.
Template
README Course: Advanced Learning Algorithms (fast.ai Practical Deep Learning for Coders) What is here: How to run it: Self-check result: What was hardest:
What gets sent back
- No history, just one big upload.
- Self-check not shown.
- Code that does not run from the README.
Resources
Foundationstart here
Depthgo deeper
What a verifier looks for
- Each notebook runs top to bottom.
- The from-scratch network matches a library model's results.
- Validation scores are reported, not training scores.
- Ask one question about a choice in the code; the answer should be clear.
You'll sign in first, then come straight back here.
Beat a baseline with a neural net or a forest
1–3 weeks (5–8 hrs/week)
Use what you learned. Make a model on a real dataset you chose that beats a simple baseline, with what you tried and what did not work. Keep it small enough to finish in about two weeks.
Proof required
Link a model on a real dataset you chose that beats a simple baseline, with what you tried and what did not work, with a README that says how to run or read it.
What gets checked
- The baseline and the model use the same held-out data.
- The gain is real, not noise (repeat runs shown).
- Failed tries are listed honestly.
Common mistakes
- A project too big to finish.
- A demo that only works on your machine.
Powstik Guide
A model on a real dataset you chose that beats a simple baseline, with what you tried and what did not work.
Steps
- Write one line: what it does and for whom.
- Build the smallest version that works.
- Test it on someone else's machine or with someone else reading it.
- Write the README.
- Ask three people to check it.
Template
README What it does: Who it is for: How to run it: What I would do next:
What gets sent back
- Does not run from the README.
- Not clearly the person's own work.
- Claims more than it does.
Resources
Foundationstart here
Depthgo deeper
What a verifier looks for
- The baseline and the model use the same held-out data.
- The gain is real, not noise (repeat runs shown).
- Failed tries are listed honestly.
- Check that the work is the person's own.
You'll sign in first, then come straight back here.