| import torch | |
| def eval_knn(x_train, y_train, x_test, y_test, k=200): | |
| """ k-nearest neighbors classifier accuracy """ | |
| d = torch.cdist(x_test, x_train) | |
| topk = torch.topk(d, k=k, dim=1, largest=False) | |
| labels = y_train[topk.indices] | |
| pred = torch.empty_like(y_test) | |
| for i in range(len(labels)): | |
| x = labels[i].unique(return_counts=True) | |
| pred[i] = x[0][x[1].argmax()] | |
| acc = (pred == y_test).float().mean().cpu().item() | |
| del d, topk, labels, pred | |
| return acc | |