changes to p6

Author: AymenEcharghaoui

mathprobs/p6/assignment.ipynb

@@ -10,9 +10,9 @@
     "\n",
     "The homework has two parts:\n",
     "\n",
-    "**Part 1 — Debugging.** Find and fix every bug in `model.py` so this notebook trains the model end-to-end to ≥ 0.95 copy-task accuracy.\n",
+    "**Part 1 — Debugging (Main Part).** Find and fix every bug in `model.py` so this notebook trains the model end-to-end to ≥ 0.95 copy-task accuracy.\n",
     "\n",
-    "**Part 2 — Layer recovery.** A teammate trained the model to convergence, then mishandled the checkpoint and the 8 blocks got shuffled. You're given the shuffled weights. Recover the correct block ordering so the model works again.\n",
+    "**Part 2 — Layer recovery (Bonus).** A teammate trained the model to convergence, then mishandled the checkpoint and the 8 blocks got shuffled. You're given the shuffled weights. Recover the correct block ordering so the model works again.\n",
     "\n",
     "The dataset (already on disk in `data/`) contains length-17 sequences of the form\n",
     "\n",
@@ -28,13 +28,13 @@
     "\n",
     "## Deliverables\n",
     "\n",
-    "**Part 1**\n",
+    "**Part 1 (2pts)**\n",
     "- **(A)** A table listing every bug you found with: file/line, cause (one sentence), minimal fix. (There are 12 bugs.)\n",
     "- **(B)** A fixed `model.py` such that this notebook reports `PASS` for Part 1.\n",
     "\n",
-    "**Part 2**\n",
+    "**Part 2 (Bonus)**\n",
     "- **(C)** The recovered permutation `block_order` (a list of 8 integers). The notebook will PASS only when both copy accuracy ≥ 0.95 **and** the MSE checksum against the trained model's reference output is below `1e-5`.\n",
-    "- **(D)** A very concise paragraph describing the procedure you used to recover the order. Brute-force search is **NOT** acceptable. Think analytical / algorithmic.\n",
+    "- **(D)** A very concise paragraph describing the procedure you used to recover the order. Brute-force search is **NOT** acceptable. Think analytical / algorithmic. Hint : What quantity might be predictive of the depth in the NN ?\n",
     "\n",
     "## Submission\n",
     "\n",