{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Lesson 37: Image Classification in Practice\n",
    "\n",
    "Every classifier so far has been evaluated with a single number: overall accuracy. That number hides a lot. This lesson builds a 4-class classifier on a deliberately imbalanced dataset and shows why accuracy alone can be misleading, using the tools that reveal what's actually going wrong: the **confusion matrix**, and **per-class precision and recall**."
   ]
  },
  {
   "cell_type": "code",
   "id": "57fdea49",
   "source": "import numpy as np\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\nimport matplotlib.pyplot as plt",
   "metadata": {},
   "execution_count": null,
   "outputs": []
  },
  {
   "cell_type": "markdown",
   "id": "2733014c",
   "source": "## A 4-class, imbalanced dataset\n\nFour noisy synthetic shapes (plus, circle, square, triangle). The test set has a roughly even mix of all four, but the *training* set is deliberately starved of triangles — a stand-in for the common real-world situation where some classes are just rarer to collect than others.",
   "metadata": {}
  },
  {
   "cell_type": "code",
   "id": "94248cb5",
   "source": "SHAPES = ['plus', 'circle', 'square', 'triangle']\n\ndef make_image(shape_type, cx, cy, size=16):\n    img = np.zeros((size, size), dtype=np.float32)\n    if shape_type == 'plus':\n        img[cy-1:cy+2, cx-3:cx+4] = 1.0\n        img[cy-3:cy+4, cx-1:cx+2] = 1.0\n    elif shape_type == 'circle':\n        yy, xx = np.mgrid[0:size, 0:size]\n        img[((xx-cx)**2 + (yy-cy)**2) <= 9] = 1.0\n    elif shape_type == 'square':\n        img[cy-3:cy+4, cx-3:cx+4] = 1.0\n    elif shape_type == 'triangle':\n        yy, xx = np.mgrid[0:size, 0:size]\n        h = 7\n        mask = (yy >= cy - h//2) & (yy <= cy + h//2) & (np.abs(xx - cx) <= (yy - (cy - h//2)) * 0.8)\n        img[mask] = 1.0\n    return img\n\ndef make_dataset(rng_local, n, position_range=(4, 12), noise=0.5):\n    imgs, labels = [], []\n    for _ in range(n):\n        shape_type = rng_local.choice(SHAPES)\n        cx, cy = rng_local.integers(*position_range), rng_local.integers(*position_range)\n        img = make_image(shape_type, cx, cy)\n        img = np.clip(img + rng_local.normal(0, noise, img.shape), 0, 1).astype(np.float32)\n        imgs.append(img)\n        labels.append(SHAPES.index(shape_type))\n    return np.array(imgs, dtype=np.float32), np.array(labels, dtype=np.int64)\n\nrng = np.random.default_rng(7)\nX_train, y_train = make_dataset(rng, 400)\nX_test, y_test = make_dataset(rng, 200)\n\n# downsample triangle in the training set only, to create class imbalance\nmask = ~((y_train == 3) & (rng.random(len(y_train)) < 0.92))\nX_train, y_train = X_train[mask], y_train[mask]\n\nprint('training class counts:', dict(zip(SHAPES, np.bincount(y_train))))\nprint('test class counts:    ', dict(zip(SHAPES, np.bincount(y_test))))\n\nfig, axes = plt.subplots(1, 4, figsize=(8, 2.2))\nfor ax, name in zip(axes, SHAPES):\n    ax.imshow(X_train[y_train == SHAPES.index(name)][0], cmap='gray')\n    ax.set_title(name, fontsize=9)\n    ax.axis('off')\nplt.show()",
   "metadata": {},
   "execution_count": null,
   "outputs": []
  },
  {
   "cell_type": "code",
   "id": "b2a478d6",
   "source": "class CNN(nn.Module):\n    def __init__(self):\n        super().__init__()\n        self.conv = nn.Sequential(\n            nn.Conv2d(1, 16, 5, padding=2), nn.ReLU(),\n            nn.MaxPool2d(2),\n            nn.Conv2d(16, 32, 5, padding=2), nn.ReLU(),\n            nn.AdaptiveMaxPool2d(1),\n        )\n        self.fc = nn.Linear(32, 4)\n\n    def forward(self, x):\n        return self.fc(self.conv(x).flatten(1))\n\ntorch.manual_seed(0)\nmodel = CNN()\nopt = torch.optim.Adam(model.parameters(), lr=0.01)\nXt = torch.tensor(X_train).unsqueeze(1); yt = torch.tensor(y_train)\nfor _ in range(300):\n    opt.zero_grad()\n    loss = F.cross_entropy(model(Xt), yt)\n    loss.backward()\n    opt.step()\n\nXte = torch.tensor(X_test).unsqueeze(1)\nwith torch.no_grad():\n    preds = model(Xte).argmax(1).numpy()\n\nacc = (preds == y_test).mean()\nprint(f'overall test accuracy: {acc:.1%}')",
   "metadata": {},
   "execution_count": null,
   "outputs": []
  },
  {
   "cell_type": "markdown",
   "id": "26cf352b",
   "source": "82% overall accuracy sounds fine on its own. It hides something important: the model is not equally good at all four classes.\n\n## Confusion matrix\n\nRow `i`, column `j` counts test examples of true class `i` predicted as class `j`. A perfect classifier is diagonal; everything off the diagonal is a specific, nameable mistake.",
   "metadata": {}
  },
  {
   "cell_type": "code",
   "id": "38bf8aa6",
   "source": "cm = np.zeros((4, 4), dtype=int)\nfor t, p in zip(y_test, preds):\n    cm[t, p] += 1\n\nprint('confusion matrix (rows=true, cols=predicted):')\nprint(f'{\"\":>10}' + ''.join(f'{s:>10}' for s in SHAPES))\nfor i, s in enumerate(SHAPES):\n    print(f'{s:>10}' + ''.join(f'{cm[i, j]:>10}' for j in range(4)))\n\nfig, ax = plt.subplots(figsize=(4.5, 4))\nim = ax.imshow(cm, cmap='Blues')\nax.set_xticks(range(4)); ax.set_xticklabels(SHAPES, rotation=45)\nax.set_yticks(range(4)); ax.set_yticklabels(SHAPES)\nax.set_xlabel('predicted'); ax.set_ylabel('true')\nfor i in range(4):\n    for j in range(4):\n        ax.text(j, i, cm[i, j], ha='center', va='center',\n                 color='white' if cm[i, j] > cm.max() / 2 else 'black')\nplt.title('Confusion matrix')\nplt.tight_layout()\nplt.show()",
   "metadata": {},
   "execution_count": null,
   "outputs": []
  },
  {
   "cell_type": "markdown",
   "id": "f44a0b1d",
   "source": "## True positives, false positives, true negatives, false negatives\n\nEvery cell of the confusion matrix above is a specific kind of correctness or mistake, but the standard vocabulary for talking about them is binary: pick one class and ask only \"is it this, or not?\" Collapsing the 4-class matrix down to \"triangle vs. everything else\" gives exactly four outcomes:\n\n- **True positive (TP)**: actually a triangle, predicted triangle. A hit.\n- **False negative (FN)**: actually a triangle, predicted something else. A miss — the model let it slip past.\n- **False positive (FP)**: actually *not* a triangle, predicted triangle anyway. A false alarm.\n- **True negative (TN)**: actually not a triangle, correctly predicted not-triangle.\n\nThis is the same 2x2 table underlying every binary classifier's evaluation (a medical test's \"positive/negative\" result, a spam filter's \"spam/not spam\" decision) — a multi-class confusion matrix is just this table computed once per class, with everything off that class's row/column collapsed into \"not this class.\"",
   "metadata": {}
  },
  {
   "cell_type": "code",
   "id": "06300321",
   "source": "cls = SHAPES.index('triangle')\ntp = cm[cls, cls]\nfn = cm[cls, :].sum() - tp          # true triangle, predicted something else\nfp = cm[:, cls].sum() - tp          # predicted triangle, actually something else\ntn = cm.sum() - tp - fn - fp        # everything else, correctly not called triangle\n\nprint(f'{\"\":>18}{\"predicted triangle\":>20}{\"predicted NOT triangle\":>24}')\nprint(f'{\"actually triangle\":>18}{tp:>20}{fn:>24}')\nprint(f'{\"actually NOT triangle\":>18}{fp:>20}{tn:>24}')\nprint()\nprint(f'TP={tp}, FP={fp}, FN={fn}, TN={tn}, total={tp+fp+fn+tn} (test set size={len(y_test)})')",
   "metadata": {},
   "execution_count": null,
   "outputs": []
  },
  {
   "cell_type": "markdown",
   "id": "f50b41fd",
   "source": "## Precision and recall\n\nTwo numbers per class, computed directly from TP/FP/FN:\n- **Recall** = TP / (TP + FN) — of everything that really *was* class `i`, what fraction did the model catch? Low recall means the model misses that class often.\n- **Precision** = TP / (TP + FP) — of everything the model *called* class `i`, what fraction actually was? Low precision means the model cries wolf on that class often.\n\n(A less commonly needed but related pair, built from the other two quadrants: **specificity** = TN / (TN + FP), how well the model avoids false alarms on the negative class, and its complement the **false positive rate** = FP / (FP + TN) = 1 − specificity.)",
   "metadata": {}
  },
  {
   "cell_type": "code",
   "id": "fccace41",
   "source": "print(f'{\"class\":>10} {\"precision\":>10} {\"recall\":>8} {\"support\":>8}')\nfor i, name in enumerate(SHAPES):\n    tp = cm[i, i]\n    fn = cm[i, :].sum() - tp\n    fp = cm[:, i].sum() - tp\n    precision = tp / (tp + fp) if (tp + fp) > 0 else float('nan')\n    recall = tp / (tp + fn) if (tp + fn) > 0 else float('nan')\n    support = cm[i, :].sum()\n    print(f'{name:>10} {precision:>10.2f} {recall:>8.2f} {support:>8}')",
   "metadata": {},
   "execution_count": null,
   "outputs": []
  },
  {
   "cell_type": "markdown",
   "id": "171afc8c",
   "source": "Triangle — the class starved to just 10 training examples — has high precision but noticeably lower recall: when the model does say \"triangle,\" it's usually right, but it fails to recognize a large fraction of the actual triangles, defaulting instead to whichever classes it saw plenty of during training. That is the standard signature of class imbalance, and it is completely invisible in the single overall-accuracy number from before. Plus and circle, by contrast, are both well-represented in training but get confused *with each other* — a different failure mode entirely, caused by genuine visual ambiguity between the two shapes under heavy noise rather than by a lack of data.\n\nThe practical lesson: always inspect the confusion matrix and per-class metrics before trusting a single accuracy figure, especially on any dataset where classes aren't naturally balanced.\n\n### Exercise\n\n1. Reduce the noise level (`noise=0.5` in `make_dataset`) to `0.2` and rerun. Does the plus/circle confusion mostly disappear? Does the triangle recall problem also improve, or does it persist — and why would data scarcity not be fixed by less noise?\n2. Try weighting the loss by inverse class frequency (`F.cross_entropy(logits, yt, weight=class_weights)`, where `class_weights[i] = 1 / count(class i)`) instead of downsampling triangles further. Does it recover triangle recall, and at what cost to the other classes' precision?\n3. Compute the **F1 score** (the harmonic mean of precision and recall, `2 * p * r / (p + r)`) for each class. Why might F1 be a better single number to track per-class than accuracy, when accuracy is only meaningful in aggregate?",
   "metadata": {}
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "name": "python",
   "version": "3.10.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}