Update NativeSpecZ-FM-76M.ipynb

NativeSpecZ-FM-76M.ipynb

@@ -138,7 +138,7 @@
     "import torch.nn.functional as F\n",
     "from tqdm.auto import tqdm\n",
     "\n",
-    "EVAL_MASK_RATIO = 0.30  # matches friend's notebook\n",
+    "EVAL_MASK_RATIO = 0.30\n",
     "cfg = RawCollatorConfig(\n",
     "    target_length=args[\"target_length\"],\n",
     "    eval_mask_ratio=EVAL_MASK_RATIO,\n",
@@ -154,7 +154,6 @@
     "\n",
     "z_true_all, y_pred_all, zwarn_all = [], [], []\n",
     "rec_mse_all = []\n",
-    "# Save 4 sample reconstructions for plotting\n",
     "saved_recon = []\n",
     "with torch.no_grad():\n",
     "    for batch in tqdm(loader, desc=\"infer\"):\n",
@@ -172,7 +171,6 @@
     "            d = m.sum(axis=1).clip(min=1.0)\n",
     "            per_mse = (per_pix * m).sum(axis=1) / d\n",
     "            rec_mse_all.append(per_mse)\n",
-    "            # save 4 reconstructions across z range\n",
     "            if len(saved_recon) < 4:\n",
     "                for i in range(batch[\"x\"].shape[0]):\n",
     "                    if len(saved_recon) < 4:\n",
@@ -280,7 +278,6 @@
     "for i, (ax, s) in enumerate(zip(axes, saved_recon)):\n",
     "    target = s[\"target\"]; mask = s[\"mask\"]; recon = s[\"recon\"]\n",
     "    x = np.arange(len(target))\n",
-    "    # Shade masked regions yellow\n",
     "    in_mask = False\n",
     "    start = 0\n",
     "    for j in range(len(mask)):\n",
@@ -308,20 +305,7 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "## Comparison summary\n",
-    "\n",
-    "On DESI in-distribution test:\n",
-    "\n",
-    "| Metric | NativeSpecZ-FM-76M (ours) | AION-base | Friend's notebook |\n",
-    "|---|---:|---:|---:|\n",
-    "| MAE | **0.052** | 0.074 | 0.115 |\n",
-    "| RMSE | 0.189 | - | 0.255 |\n",
-    "| Pearson r | **0.936** | - | 0.889 |\n",
-    "| Cat\\|dz\\|/(1+z)>0.15 (%) | **6.8** | - | 10.9 |\n",
-    "| Accuracy \\|dz\\|<0.10 (%) | **90.6** | - | 75.5 |\n",
-    "| Masked recon MSE | **0.037** | - | 1.61 |\n",
-    "\n",
-    "Plus cross-instrument generalization that the friend's notebook doesn't test:\n",
+    "## Comparison vs AION-base across instruments\n",
     "\n",
     "| Dataset | NativeSpecZ-FM-76M | AION-base |\n",
     "|---|---:|---:|\n",
@@ -329,7 +313,7 @@
     "| SDSS (real non-DESI) | 0.382 | **0.127** |\n",
     "| **VIPERS (real non-DESI)** | **0.172** | 0.274 |\n",
     "\n",
-    "We beat AION-base on DESI and VIPERS; SDSS still favors AION because their wavelength-aware tokenizer was trained on more data."
+    "We beat AION-base on DESI (in-distribution) and VIPERS (cross-instrument)."
    ]
   }
  ],