[LAB 12 · CLIP延伸：缓存与检索（近似索引选读）](/course/notebooks/921eeaf784284ce19568c6ca7619b55a.html)

# 2 · 精确搜索：模型给分，评价定义相关 · 实现代码
[独立阅读](/course/notebook-blocks/e05d4164e12144fea000c0bf4d6b7e2f.html)
并入相关学习路径

同一CLIP缓存、exact/IVF搜索、延迟/扫描量、区域交互基线
展开完整实现代码
```
X,Q=a['gallery'],a['queries']
assert np.allclose(np.linalg.norm(X,axis=1),1,atol=2e-6)
labels=np.array([r['label'] for r in manifest['gallery']])
truth=[r['label'] for r in manifest['queries']]
S=Q@X.T
exact=np.stack([cr.stable_rank(row,k=3) for row in S])
for i,row in enumerate(exact):
    print(manifest['queries'][i]['text'],'->',[(manifest['gallery'][j]['id'],round(float(S[i,j]),4)) for j in row])
hit3=np.mean([any(labels[j]==truth[i] for j in row) for i,row in enumerate(exact)])
print('exact semantic Hit@3:',hit3)
# 反方向：每张图对 12 条文字排序。每类有两条同义候选。
i2t=np.stack([cr.stable_rank(row,k=1) for row in S.T])
print('image-to-text Hit@1:',np.mean([truth[row[0]]==labels[i] for i,row in enumerate(i2t)]))

```