vector<size_t> KMeans(const vector<double>& data, size_t K) {
    size_t dimensions = dim;
    size_t d_size = data.size();
    size_t data_size = d_size / dimensions;
    vector<size_t> clusters(data_size);
    size_t new_size = K * dimensions;
    // Initialize centroids randomly at data points
    vector<double> centroids(new_size);
    for (size_t i = 0; i < K; i++) {
        if (dim == 2) {
            size_t index = UniformRandom(data_size - 1);
            centroids[2 * i] = data[2 * index];
            centroids[2 * i + 1] = data[2 * index + 1];
        } else {
            continue;
        }
    }

    bool converged = false;
    while (!converged) {
        converged = true;
        for (size_t i = 0; i < data_size; ++i) {
            // делать точку из data[i], if dim == 2
            Point tmp;
            tmp.push_back(data[i * dimensions]);
            tmp.push_back(data[i * dimensions + 1]);
            size_t nearest_cluster = FindNearestCentroid(centroids, tmp/*FIXED*/);
            if (clusters[i] != nearest_cluster) {
                clusters[i] = nearest_cluster;
                converged = false;
            }
        }
        if (converged) {
            break;
        }
        vector<size_t> clusters_sizes(K);
        vector<double> centroids(K * dimensions);
        for (size_t i = 0; i < data_size; ++i) {     // добавить хардкод для dim == 2
            for (size_t d = 0; d < dimensions; ++d) {
                centroids[clusters[i] * dimensions + d] += data[i * dimensions + d];
            }
            ++clusters_sizes[clusters[i]];
        }
        for (size_t i = 0; i < K; ++i) {
            if (clusters_sizes[i] != 0) {
                for (size_t d = 0; d < dimensions; ++d) {
                    centroids[i * dimensions + d] /= clusters_sizes[i];
                }
            }
        }
        for (size_t i = 0; i < K; ++i) {
            if (clusters_sizes[i] == 0) {
                auto el = GetRandomPosition(centroids);
                if (dimensions == 2) {
                    centroids[i * 2] = el[0];
                    centroids[i * 2 + 1] = el[1];
                } else {
                    for (size_t j = 0; j < el.size(); ++j) {
                        centroids[i * dimensions + j] = el[j];
                    }
                }
            }
        }
    }
    return clusters;
}