k_to_accuracies = {}
classifier = KNearestNeighbor()
for k in k_choices:
    print k
    acc=np.zeros(num_folds)
    for i, fold in enumerate(range(num_folds)):
        X_tests=X_train_folds[fold]
        y_tests=y_train_folds[fold]

        if fold == 0:
            X_trains=X_train_folds[1]
            y_trains=y_train_folds[1]
        else:
            X_trains=X_train_folds[0]
            y_trains=y_train_folds[0]


        for train in range(num_folds-1):
            if train+1 != fold:
                X_trains=np.concatenate((X_trains, X_train_folds[train+1]))
                y_trains=np.concatenate((y_trains, y_train_folds[train+1]))

        classifier.train(X_trains, y_trains)
        dists = classifier.compute_distances_no_loops(X_tests)
        y_test_pred = classifier.predict_labels(dists, k=5)
        num_correct = np.sum(y_test_pred == y_tests)
        acc[i] = float(num_correct) / 1000
    k_to_accuracies[k]=acc