ML Labs Exam — Study Plan (ML_LABS_EXAM_001 → retake)
md2 min readSummary of AIDE500
Where you stand
Attempt 1: 54/120 (45%) — T/F 7/10 · MCQ 20/30 · Code 19/36 · Open 0/30 · Fill 8/14. Strengths: transforms, NumPy, CNN shape math, ReLU/loss recall, matplotlib basics. The 66 missing points come from just a handful of gaps (below) + 30 points of unanswered open questions.
The materials
| File | What it covers |
|---|---|
Study_01_Labs_1-3_Jupyter_Plot_Numpy.md |
Jupyter modes, matplotlib, NumPy (dtype, axis, broadcasting, random) |
Study_02_Labs_4-5_Tensors_Autograd.md |
tensor creation/copy-vs-share, cat/mul, in-place _, autograd, backward/step/zero_grad, freeze |
Study_03_Labs_6-7_NN_and_nnModule.md |
resnet, Linear/Conv/ReLU/Softmax, output-size formula, class Net, training loop basics |
Study_04_Labs_8-9_Transforms_Dataloaders_Classifier.md |
ToTensor/Normalize, DataLoader, CIFAR-10 full pipeline, logits/argmax, no_grad |
Study_05_Lab10_Augmentation_and_Exam_Review.md |
4 augmentation families + all 47 questions mapped, model answers for Q32–Q37 |
The method (per guide, ~45–60 min)
- Read the guide once without stopping.
- Run the actual notebook cells it refers to (Labs 1→10 in order).
- Retype the key code block from memory (training loop, Net class, shape trace).
- Self-check: answer every numbered question out loud, then compare with the answer lines.
- Tick the exam-question IDs you now own (each guide flags them with ⚠).
Your 7 targeted weak areas → where they're fixed
| # | Gap (attempt 1) | Guide section |
|---|---|---|
| R1 | logits vs probabilities (q5, q24) | G4 §loss + §testing |
| R2 | copy vs share (q17, q31) | G2 §cheat-sheet + §bridge |
| R3 | axis semantics (q14) | G1 §axis |
| R4 | Normalize ranges (q23, q43) | G4 §transforms |
| R5 | eval protocol & loop (q28, q34, q41, q42, q47) | G4 §loop + G5 Q34 |
| R6 | Jupyter + squeeze (q38, q46) | G1 §Jupyter + G4 §viz |
| R7 | Open questions (q32–q37 = 30 pts) | G5 §model answers — highest ROI |
Suggested schedule
- Day 1: G1 + G2 (Labs 1–5) — fixes R2, R3, R6.
- Day 2: G3 + G4 (Labs 6–9) — fixes R1, R4, R5.
- Day 3: G5 + rerun Lab 10 — fixes R7, then re-attempt every self-check you previously missed.
- Day 4: Redo exam cold (see below).
Resubmitting
- Open
ML_LABS_EXAM_001.htmlin your browser (works from the Lab Notebooks folder). - Answer every question — especially Section 4 (6 × 5 pts). Partial answers earn rubric points; blank earns 0.
- Click "Export answers" → saves
ML_LABS_EXAM_001_answers.json(to Downloads; copy it into this folder to overwrite the old one). - Tell me: "Grade my exam" → I grade against
ML_LABS_EXAM_001_answer_key.jsonwith partial credit and give you a new report. - Alternative: ask me for
ML_LABS_EXAM_002— a fresh exam (same blueprint: 47 Qs / 120 pts, new items) if you want a clean retake after studying. Your current answer file is unchanged, so nothing is lost either way.
Tip: the exam never shows your score in the browser — grading only happens when you ask. Export any time; you can export, study more, and export again.