compress_sparse_triplets.hpp

Go to the documentation of this file.

#ifndef TATAMI_COMPRESS_SPARSE_TRIPLETS_H
#define TATAMI_COMPRESS_SPARSE_TRIPLETS_H
 
#include "../utils/Index_to_container.hpp"
 
#include <vector>
#include <algorithm>
#include <numeric>
#include <utility>
#include <cstddef>
 
#include "sanisizer/sanisizer.hpp"
 
namespace tatami {
 
namespace compress_triplets {
 
template<class Primary_, class Secondary_>
int is_ordered(const Primary_& primary, const Secondary_& secondary) {
    if (!std::is_sorted(primary.begin(), primary.end())) {
        return 2;
    }
 
    auto nprimary = primary.size();
    decltype(nprimary) start = 0;
    while (start < nprimary) {
        decltype(nprimary) end = start + 1;
        while (end < nprimary && primary[end] == primary[start]) {
            if (secondary[end] < secondary[end - 1]) {
                // Quit on first failure; we've seen enough.
                return 1;
            }
            ++end;
        }
        start = end;
    }
 
    return 0;
}
 
template<typename Size_, class Primary_, class Secondary_>
void order(int status, std::vector<Size_>& indices, const Primary_& primary, const Secondary_& secondary) {
    if (status == 1) {
        auto nprimary = primary.size();
        decltype(nprimary) start = 0;
        while (start < nprimary) {
            decltype(nprimary) end = start + 1;
            while (end < nprimary && primary[end] == primary[start]) {
                ++end;
            }
 
            // Checking if this particular run can be skipped.
            if (!std::is_sorted(secondary.begin() + start, secondary.begin() + end)) {
                std::sort(indices.begin() + start, indices.begin() + end, [&](Size_ left, Size_ right) -> bool {
                    return secondary[left] < secondary[right];
                });
            }
            start = end;
        }
 
    } else if (status == 2) {
        std::sort(indices.begin(), indices.end(), [&](Size_ left, Size_ right) -> bool {
            if (primary[left] == primary[right]) {
                return (secondary[left] < secondary[right]);
            }
            return (primary[left] < primary[right]);
        });
    }
}
 
}
template<class Values_, typename Pointer_ = decltype(std::declval<Values_>().size()), class PrimaryIndices_, class SecondaryIndices_>
std::vector<Pointer_> compress_sparse_triplets(std::size_t num_primary, Values_& values, const PrimaryIndices_& primary_indices, SecondaryIndices_& secondary_indices) {
    auto N = values.size();
    if (!safe_non_negative_equal(N, primary_indices.size()) || !safe_non_negative_equal(N, secondary_indices.size())) {
        throw std::runtime_error("'primary_indices', 'secondary_indices' and 'values' should have the same length");
    }
 
    int order_status = compress_triplets::is_ordered(primary_indices, secondary_indices);
    if (order_status != 0) {
        auto indices = sanisizer::create<std::vector<decltype(N)> >(N);
        std::iota(indices.begin(), indices.end(), static_cast<decltype(N)>(0));
        compress_triplets::order(order_status, indices, primary_indices, secondary_indices);
 
        // Reordering values in place. This saves memory and allows us to work with Values_, RowIndices_, etc. that may not have well-defined copy constructors.
        auto used = sanisizer::create<std::vector<unsigned char> >(N);
        for (decltype(N) i = 0; i < N; ++i) {
            if (used[i]) {
                continue;
            }
            auto current = i, replacement = indices[i];
            used[i] = 1;
 
            while (replacement != i) {
                std::swap(secondary_indices[current], secondary_indices[replacement]);
                std::swap(values[current], values[replacement]);
                current = replacement;
                used[current] = 1;
                replacement = indices[replacement];
            }
        }
    }
 
    typedef std::vector<Pointer_> Output;
    typedef typename Output::size_type OutputSize;
    Output output(sanisizer::sum<OutputSize>(num_primary, 1));
    for (auto t : primary_indices) {
        ++(output[static_cast<OutputSize>(t) + 1]);
    }
    std::partial_sum(output.begin(), output.end(), output.begin());
 
    return output;
}
 
// Back-compatibility.
template<class Values_, class RowIndices_, class ColumnIndices_>
std::vector<decltype(std::declval<Values_>().size())> compress_sparse_triplets(std::size_t nrow, std::size_t ncol, Values_& values, RowIndices_& row_indices, ColumnIndices_& column_indices, bool csr) {
    if (csr) {
        auto output = compress_sparse_triplets(nrow, values, row_indices, column_indices);
        for (decltype(nrow) r = 0; r < nrow; ++r) {
            std::fill_n(row_indices.begin() + output[r], output[r + 1] - output[r], r);
        }
        return output;
    } else {
        auto output = compress_sparse_triplets(ncol, values, column_indices, row_indices);
        for (decltype(ncol) c = 0; c < ncol; ++c) {
            std::fill_n(column_indices.begin() + output[c], output[c + 1] - output[c], c);
        }
        return output;
    }
}
 
template<bool row_, class Values_, class RowIndices_, class ColumnIndices_>
auto compress_sparse_triplets(std::size_t nrow, std::size_t ncol, Values_& values, RowIndices_& row_indices, ColumnIndices_& column_indices) {
    return compress_sparse_triplets(nrow, ncol, values, row_indices, column_indices, row_);
}
}
 
#endif