FragmentedSparseMatrix.hpp

Go to the documentation of this file.

#ifndef TATAMI_FRAGMENTED_SPARSE_MATRIX_H
#define TATAMI_FRAGMENTED_SPARSE_MATRIX_H
 
#include "../base/Matrix.hpp"
#include "../utils/copy.hpp"
#include "../utils/ElementType.hpp"
#include "../utils/PseudoOracularExtractor.hpp"
#include "../utils/Index_to_container.hpp"
 
#include "primary_extraction.hpp"
#include "secondary_extraction.hpp"
 
#include <vector>
#include <algorithm>
#include <memory>
#include <utility>
#include <stdexcept>
#include <cstddef>
 
#include "sanisizer/sanisizer.hpp"
 
namespace tatami {
 
namespace FragmentedSparseMatrix_internal {
 
/********************
 *** Primary full ***
 ********************/
 
template<typename Value_, typename Index_, class ValueVectorStorage_, class IndexVectorStorage_>
class PrimaryMyopicFullDense final : public MyopicDenseExtractor<Value_, Index_> {
public:
    PrimaryMyopicFullDense(const ValueVectorStorage_& values, const IndexVectorStorage_& indices, const Index_ secondary) :
        my_values(values), my_indices(indices), my_secondary(secondary) {} 
 
    const Value_* fetch(const Index_ i, Value_* const buffer) {
        const auto& curv = my_values[i];
        const auto& curi = my_indices[i];
 
        std::fill_n(buffer, my_secondary, static_cast<Value_>(0));
        const auto curnnz = curv.size();
        for (I<decltype(curnnz)> x = 0; x < curnnz; ++x) {
            buffer[curi[x]] = curv[x];
        }
        return buffer;
    }
 
private:
    const ValueVectorStorage_& my_values;
    const IndexVectorStorage_& my_indices;
    Index_ my_secondary;
};
 
template<typename Value_, typename Index_, class ValueVectorStorage_, class IndexVectorStorage_>
class PrimaryMyopicFullSparse final : public MyopicSparseExtractor<Value_, Index_> {
public:
    PrimaryMyopicFullSparse(
        const ValueVectorStorage_& values,
        const IndexVectorStorage_& indices,
        [[maybe_unused]] const Index_ secondary /* for consistency only */,
        const Options& opt
    ) :
        my_values(values),
        my_indices(indices),
        my_needs_value(opt.sparse_extract_value),
        my_needs_index(opt.sparse_extract_index)
    {} 
 
    SparseRange<Value_, Index_> fetch(const Index_ i, Value_* const value_buffer, Index_* const index_buffer) {
        const auto& curv = my_values[i];
        const auto& curi = my_indices[i];
 
        SparseRange<Value_, Index_> output(curv.size(), NULL, NULL);
        if (my_needs_value) {
            const auto curnnz = curv.size();
            output.value = sparse_utils::extract_primary_vector(curv, static_cast<I<decltype(curnnz)> >(0), curnnz, value_buffer);
        }
        if (my_needs_index) {
            const auto curnnz = curi.size();
            output.index = sparse_utils::extract_primary_vector(curi, static_cast<I<decltype(curnnz)> >(0), curnnz, index_buffer);
        }
        return output;
    }
 
private:
    const ValueVectorStorage_& my_values;
    const IndexVectorStorage_& my_indices;
    bool my_needs_value, my_needs_index;
};
 
/*********************
 *** Primary block ***
 *********************/
 
template<typename Value_, typename Index_, class ValueVectorStorage_, class IndexVectorStorage_>
class PrimaryMyopicBlockDense final : public MyopicDenseExtractor<Value_, Index_> {
public:
    PrimaryMyopicBlockDense(
        const ValueVectorStorage_& values,
        const IndexVectorStorage_& indices,
        const Index_ secondary,
        const Index_ block_start,
        const Index_ block_length
    ) :
        my_values(values),
        my_indices(indices),
        my_secondary(secondary),
        my_block_start(block_start),
        my_block_length(block_length)
    {} 
 
    const Value_* fetch(const Index_ i, Value_* const buffer) {
        const auto& curi = my_indices[i];
        const auto& curv = my_values[i];
 
        auto iStart = curi.begin();
        auto iEnd = curi.end();
        sparse_utils::refine_primary_block_limits(iStart, iEnd, my_secondary, my_block_start, my_block_length);
        const auto start_pos = (iStart - curi.begin());
        const auto end_pos = (iEnd - curi.begin());
 
        std::fill_n(buffer, my_block_length, static_cast<Value_>(0));
        for (auto x = start_pos; x < end_pos; ++x) {
            buffer[curi[x] - my_block_start] = curv[x];
        }
        return buffer;
    }
 
private:
    const ValueVectorStorage_& my_values;
    const IndexVectorStorage_& my_indices;
    Index_ my_secondary;
    Index_ my_block_start, my_block_length;
};
 
template<typename Value_, typename Index_, class ValueVectorStorage_, class IndexVectorStorage_>
class PrimaryMyopicBlockSparse final : public MyopicSparseExtractor<Value_, Index_> {
public:
    PrimaryMyopicBlockSparse(
        const ValueVectorStorage_& values,
        const IndexVectorStorage_& indices,
        const Index_ secondary,
        const Index_ block_start,
        const Index_ block_length,
        const Options& opt
    ) :
        my_values(values),
        my_indices(indices), 
        my_secondary(secondary),
        my_block_start(block_start),
        my_block_length(block_length),
        my_needs_value(opt.sparse_extract_value), 
        my_needs_index(opt.sparse_extract_index)
    {} 
 
    SparseRange<Value_, Index_> fetch(const Index_ i, Value_* const value_buffer, Index_* const index_buffer) {
        const auto& curi = my_indices[i];
        auto iStart = curi.begin();
        auto iEnd = curi.end();
        sparse_utils::refine_primary_block_limits(iStart, iEnd, my_secondary, my_block_start, my_block_length);
        const auto offset = iStart - curi.begin();
        const auto delta = iEnd - iStart;
 
        SparseRange<Value_, Index_> output(delta, NULL, NULL);
        if (my_needs_value) {
            output.value = sparse_utils::extract_primary_vector(my_values[i], offset, delta, value_buffer);
        }
        if (my_needs_index) {
            output.index = sparse_utils::extract_primary_vector(curi, offset, delta, index_buffer);
        }
        return output;
    }
 
private:
    const ValueVectorStorage_& my_values;
    const IndexVectorStorage_& my_indices;
    Index_ my_secondary;
    Index_ my_block_start, my_block_length;
    bool my_needs_value, my_needs_index;
};
 
/***********************
 *** Primary indexed ***
 ***********************/
 
template<typename Value_, typename Index_, class ValueVectorStorage_, class IndexVectorStorage_>
class PrimaryMyopicIndexDense final : public MyopicDenseExtractor<Value_, Index_> {
public:
    PrimaryMyopicIndexDense(
        const ValueVectorStorage_& values,
        const IndexVectorStorage_& indices,
        const Index_ secondary,
        VectorPtr<Index_> indices_ptr
    ) :
        my_values(values),
        my_indices(indices),
        my_retriever(*indices_ptr, secondary),
        my_num_indices(indices_ptr->size())
    {} 
 
    const Value_* fetch(const Index_ i, Value_* const buffer) {
        const auto& curi = my_indices[i];
        const auto& curv = my_values[i];
        std::fill_n(buffer, my_num_indices, static_cast<Value_>(0));
        my_retriever.populate(
            curi.begin(),
            curi.end(),
            [&](const auto s, const auto offset) -> void {
                buffer[s] = curv[offset];
            }
        );
        return buffer;
    }
 
private:
    const ValueVectorStorage_& my_values;
    const IndexVectorStorage_& my_indices;
    sparse_utils::RetrievePrimarySubsetDense<Index_> my_retriever;
    std::size_t my_num_indices;
};
 
template<typename Value_, typename Index_, class ValueVectorStorage_, class IndexVectorStorage_>
class PrimaryMyopicIndexSparse final : public MyopicSparseExtractor<Value_, Index_> {
public:
    PrimaryMyopicIndexSparse(
        const ValueVectorStorage_& values,
        const IndexVectorStorage_& indices,
        const Index_ secondary,
        VectorPtr<Index_> indices_ptr,
        const Options& opt
    ) :
        my_values(values),
        my_indices(indices),
        my_retriever(*indices_ptr, secondary),
        my_needs_value(opt.sparse_extract_value),
        my_needs_index(opt.sparse_extract_index)
    {} 
 
    SparseRange<Value_, Index_> fetch(const Index_ i, Value_* const value_buffer, Index_* const index_buffer) {
        const auto& curi = my_indices[i];
        const auto& curv = my_values[i];
        Index_ count = 0;
        auto vcopy = value_buffer;
        auto icopy = index_buffer;
 
        my_retriever.populate(
            curi.begin(),
            curi.end(),
            [&](const auto offset, const auto ix) -> void {
                ++count;
                if (my_needs_value) {
                    *vcopy = curv[offset];
                    ++vcopy;
                }
                if (my_needs_index) {
                    *icopy = ix;
                    ++icopy;
                }
            }
        );
 
        return SparseRange<Value_, Index_>(count, my_needs_value ? value_buffer : NULL, my_needs_index ? index_buffer : NULL);
    }
 
private:
    const ValueVectorStorage_& my_values;
    const IndexVectorStorage_& my_indices;
    sparse_utils::RetrievePrimarySubsetSparse<Index_> my_retriever;
    bool my_needs_value, my_needs_index;
};
 
/**********************
 *** Secondary full ***
 **********************/
 
template<typename Index_, class IndexVectorStorage_>
class ServeIndices {
public:
    ServeIndices(const IndexVectorStorage_& indices) : my_indices(indices) {}
 
private:
    const IndexVectorStorage_& my_indices;
 
public:
    typedef I<decltype(std::declval<IndexVectorStorage_>()[0].size())> Pointer;
 
    Pointer start_offset(const Index_) const {
        return 0;
    }
 
    Pointer end_offset(const Index_ primary) const {
        return my_indices[primary].size();
    }
 
    auto raw(const Index_ primary) const {
        return my_indices[primary].begin();
    }
};
 
template<typename Index_, class IndexVectorStorage_>
auto make_ServeIndices(const IndexVectorStorage_& i) {
    return ServeIndices<Index_, IndexVectorStorage_>(i);
}
 
template<typename Value_, typename Index_, class ValueVectorStorage_, class IndexVectorStorage_> 
class SecondaryMyopicFullDense final : public MyopicDenseExtractor<Value_, Index_> {
public:
    SecondaryMyopicFullDense(
        const ValueVectorStorage_& values,
        const IndexVectorStorage_& indices,
        Index_ secondary
    ) :
        my_values(values),
        my_cache(make_ServeIndices<Index_>(indices), secondary, indices.size())
    {} 
 
    const Value_* fetch(const Index_ i, Value_* const buffer) {
        std::fill_n(buffer, my_cache.size(), static_cast<Value_>(0));
        my_cache.search(
            i, 
            [&](Index_ primary, Index_ index_primary, auto ptr) -> void {
                buffer[index_primary] = my_values[primary][ptr];
            }
        );
        return buffer;
    }
 
private:
    const ValueVectorStorage_& my_values;
    sparse_utils::FullSecondaryExtractionCache<Index_, ServeIndices<Index_, IndexVectorStorage_> > my_cache;
};
 
template<typename Value_, typename Index_, class ValueVectorStorage_, class IndexVectorStorage_>
class SecondaryMyopicFullSparse final : public MyopicSparseExtractor<Value_, Index_> {
public:
    SecondaryMyopicFullSparse(
        const ValueVectorStorage_& values,
        const IndexVectorStorage_& indices,
        const Index_ secondary,
        const Options& opt
    ) :
        my_values(values),
        my_cache(make_ServeIndices<Index_>(indices), secondary, indices.size()),
        my_needs_value(opt.sparse_extract_value),
        my_needs_index(opt.sparse_extract_index)
    {} 
 
    SparseRange<Value_, Index_> fetch(const Index_ i, Value_* const value_buffer, Index_* const index_buffer) {
        Index_ count = 0;
        my_cache.search(
            i,
            [&](const Index_ primary, const Index_, const auto ptr) -> void {
                if (my_needs_value) {
                    value_buffer[count] = my_values[primary][ptr];
                }
                if (my_needs_index) {
                    index_buffer[count] = primary;
                }
                ++count;
            }
        );
        return SparseRange<Value_, Index_>(count, my_needs_value ? value_buffer : NULL, my_needs_index ? index_buffer : NULL);
    }
 
private:
    const ValueVectorStorage_& my_values;
    sparse_utils::FullSecondaryExtractionCache<Index_, ServeIndices<Index_, IndexVectorStorage_> > my_cache;
    bool my_needs_value, my_needs_index;
};
 
/***********************
 *** Secondary block ***
 ***********************/
 
template<typename Value_, typename Index_, class ValueVectorStorage_, class IndexVectorStorage_>
class SecondaryMyopicBlockDense final : public MyopicDenseExtractor<Value_, Index_> {
public:
    SecondaryMyopicBlockDense(
        const ValueVectorStorage_& values,
        const IndexVectorStorage_& indices,
        const Index_ secondary,
        const Index_ block_start,
        const Index_ block_length
    ) :
        my_values(values),
        my_cache(make_ServeIndices<Index_>(indices), secondary, block_start, block_length)
    {}
 
    const Value_* fetch(const Index_ i, Value_* const buffer) {
        std::fill_n(buffer, my_cache.size(), static_cast<Value_>(0));
        my_cache.search(
            i,
            [&](const Index_ primary, const Index_ index_primary, const auto ptr) -> void {
                buffer[index_primary] = my_values[primary][ptr];
            }
        );
        return buffer;
    }
 
private:
    const ValueVectorStorage_& my_values;
    sparse_utils::BlockSecondaryExtractionCache<Index_, ServeIndices<Index_, IndexVectorStorage_> > my_cache;
};
 
template<typename Value_, typename Index_, class ValueVectorStorage_, class IndexVectorStorage_>
class SecondaryMyopicBlockSparse final : public MyopicSparseExtractor<Value_, Index_> {
public:
    SecondaryMyopicBlockSparse(
        const ValueVectorStorage_& values,
        const IndexVectorStorage_& indices,
        const Index_ secondary,
        const Index_ block_start,
        const Index_ block_length,
        const Options& opt
    ) :
        my_values(values),
        my_cache(make_ServeIndices<Index_>(indices), secondary, block_start, block_length),
        my_needs_value(opt.sparse_extract_value),
        my_needs_index(opt.sparse_extract_index)
    {} 
 
    SparseRange<Value_, Index_> fetch(const Index_ i, Value_* const value_buffer, Index_* const index_buffer) {
        Index_ count = 0;
        my_cache.search(
            i, 
            [&](const Index_ primary, const Index_, const auto ptr) -> void {
                if (my_needs_value) {
                    value_buffer[count] = my_values[primary][ptr];
                }
                if (my_needs_index) {
                    index_buffer[count] = primary;
                }
                ++count;
            }
        );
        return SparseRange<Value_, Index_>(count, my_needs_value ? value_buffer : NULL, my_needs_index ? index_buffer : NULL);
    }
 
private:
    const ValueVectorStorage_& my_values;
    sparse_utils::BlockSecondaryExtractionCache<Index_, ServeIndices<Index_, IndexVectorStorage_> > my_cache;
    bool my_needs_value, my_needs_index;
};
 
/***********************
 *** Secondary index ***
 ***********************/
 
template<typename Value_, typename Index_, class ValueVectorStorage_, class IndexVectorStorage_>
class SecondaryMyopicIndexDense final : public MyopicDenseExtractor<Value_, Index_> {
public:
    SecondaryMyopicIndexDense(
        const ValueVectorStorage_& values,
        const IndexVectorStorage_& indices,
        const Index_ secondary,
        VectorPtr<Index_> indices_ptr
    ) :
        my_values(values),
        my_cache(make_ServeIndices<Index_>(indices), secondary, std::move(indices_ptr))
    {}
 
    const Value_* fetch(const Index_ i, Value_* const buffer) {
        std::fill_n(buffer, my_cache.size(), static_cast<Value_>(0));
        my_cache.search(
            i,
            [&](Index_ primary, Index_ index_primary, auto ptr) -> void {
                buffer[index_primary] = my_values[primary][ptr];
            }
        );
        return buffer;
    }
 
private:
    const ValueVectorStorage_& my_values;
    sparse_utils::IndexSecondaryExtractionCache<Index_, ServeIndices<Index_, IndexVectorStorage_> > my_cache;
};
 
template<typename Value_, typename Index_, class ValueVectorStorage_, class IndexVectorStorage_>
class SecondaryMyopicIndexSparse final : public MyopicSparseExtractor<Value_, Index_> {
public:
    SecondaryMyopicIndexSparse(
        const ValueVectorStorage_& values,
        const IndexVectorStorage_& indices,
        const Index_ secondary,
        VectorPtr<Index_> indices_ptr,
        const Options& opt
    ) :
        my_values(values),
        my_cache(make_ServeIndices<Index_>(indices), secondary, std::move(indices_ptr)),
        my_needs_value(opt.sparse_extract_value),
        my_needs_index(opt.sparse_extract_index)
    {} 
 
    SparseRange<Value_, Index_> fetch(const Index_ i, Value_* const value_buffer, Index_* const index_buffer) {
        Index_ count = 0;
        my_cache.search(
            i,
            [&](const Index_ primary, const Index_, const auto ptr) -> void {
                if (my_needs_value) {
                    value_buffer[count] = my_values[primary][ptr];
                }
                if (my_needs_index) {
                    index_buffer[count] = primary;
                }
                ++count;
            }
        );
        return SparseRange<Value_, Index_>(count, my_needs_value ? value_buffer : NULL, my_needs_index ? index_buffer : NULL);
    }
 
private:
    const ValueVectorStorage_& my_values;
    sparse_utils::IndexSecondaryExtractionCache<Index_, ServeIndices<Index_, IndexVectorStorage_> > my_cache;
    bool my_needs_value, my_needs_index;
};
 
}
struct FragmentedSparseMatrixOptions {
    bool check = true;
};
 
template<typename Value_, typename Index_, class ValueVectorStorage_, class IndexVectorStorage_>
class FragmentedSparseMatrix : public Matrix<Value_, Index_> {
public:
    FragmentedSparseMatrix(
        const Index_ nrow,
        const Index_ ncol,
        ValueVectorStorage_ values,
        IndexVectorStorage_ indices,
        const bool row_sparse,
        const FragmentedSparseMatrixOptions& options
    ) : 
        my_nrow(nrow),
        my_ncol(ncol),
        my_values(std::move(values)),
        my_indices(std::move(indices)),
        my_row_sparse(row_sparse)
    {
        if (options.check) {
            if (my_values.size() != my_indices.size()) {
                throw std::runtime_error("'values' and 'indices' should be of the same length");
            }
 
            if (my_row_sparse) {
                if (!safe_non_negative_equal(my_indices.size(), my_nrow)) {
                    throw std::runtime_error("length of 'indices' should be equal to number of rows'");
                }
            } else {
                if (!safe_non_negative_equal(my_indices.size(), my_ncol)) {
                    throw std::runtime_error("length of 'indices' should be equal to number of columns");
                }
            }
 
            const ElementType<ElementType<IndexVectorStorage_> > max_index = (my_row_sparse ? my_ncol : my_nrow);
            const auto num_indices = my_indices.size();
            for (I<decltype(num_indices)> i = 0; i < num_indices; ++i) {
                const auto& curv = my_values[i];
                const auto& curi = my_indices[i];
                if (!safe_non_negative_equal(curv.size(), curi.size())) {
                    throw std::runtime_error("corresponding elements of 'values' and 'indices' should have the same length");
                }
 
                for (const auto x : curi) {
                    if (x < 0 || x >= max_index) {
                        throw std::runtime_error("'indices' should contain non-negative integers less than the number of " + (my_row_sparse ? std::string("columns") : std::string("rows")));
                    }
                }
 
                const auto curnnz = curi.size();
                for (I<decltype(curnnz)> j = 1; j < curnnz; ++j) {
                    if (curi[j] <= curi[j - 1]) {
                        throw std::runtime_error("my_indices should be strictly increasing within each element of 'indices'");
                    }
                }
 
                // Check that iterator subtraction is safe.
                // Various functions in 'sparse_utils' will subtract iterators when converting the lower_bound return value to an index.
                // We cast to Index_ as it gives us a chance to skip the check at compile time, given that the size is no greater than the dimension extents.
                sanisizer::can_ptrdiff<I<decltype(curi.begin())> >(attest_for_Index(static_cast<Index_>(curnnz)));
            }
        }
    }
 
    // Back-compatibility only.
    FragmentedSparseMatrix(Index_ nrow, Index_ ncol, ValueVectorStorage_ values, IndexVectorStorage_ indices, bool row_sparse, bool check = true) :
        FragmentedSparseMatrix(
            nrow,
            ncol,
            std::move(values),
            std::move(indices),
            row_sparse,
            [&]{
                FragmentedSparseMatrixOptions fopt;
                fopt.check = check;
                return fopt;
            }()
        )
    {}
private:
    Index_ my_nrow, my_ncol;
    ValueVectorStorage_ my_values;
    IndexVectorStorage_ my_indices;
    bool my_row_sparse;
 
public:
    Index_ nrow() const { return my_nrow; }
 
    Index_ ncol() const { return my_ncol; }
 
    bool is_sparse() const { return true; }
 
    double is_sparse_proportion() const { return 1; }
 
    bool prefer_rows() const { return my_row_sparse; }
 
    double prefer_rows_proportion() const { return static_cast<double>(my_row_sparse); }
 
    bool uses_oracle(const bool) const { return false; }
 
    using Matrix<Value_, Index_>::dense;
 
    using Matrix<Value_, Index_>::sparse;
 
private:
    Index_ secondary() const {
        if (my_row_sparse) {
            return my_ncol;
        } else {
            return my_nrow;
        }
    }
 
    /*****************************
     ******* Dense myopic ********
     *****************************/
private:
    std::unique_ptr<MyopicDenseExtractor<Value_, Index_> > dense(
        const bool row,
        const Options&
    ) const {
        if (my_row_sparse == row) {
            return std::make_unique<FragmentedSparseMatrix_internal::PrimaryMyopicFullDense<Value_, Index_, ValueVectorStorage_, IndexVectorStorage_> >(
                my_values, my_indices, secondary()
            );
        } else {
            return std::make_unique<FragmentedSparseMatrix_internal::SecondaryMyopicFullDense<Value_, Index_, ValueVectorStorage_, IndexVectorStorage_> >(
                my_values, my_indices, secondary()
            ); 
        }
    }
 
    std::unique_ptr<MyopicDenseExtractor<Value_, Index_> > dense(
        const bool row,
        const Index_ block_start,
        const Index_ block_end,
        const Options&
    ) const {
        if (my_row_sparse == row) {
            return std::make_unique<FragmentedSparseMatrix_internal::PrimaryMyopicBlockDense<Value_, Index_, ValueVectorStorage_, IndexVectorStorage_> >(
                my_values, my_indices, secondary(), block_start, block_end
            );
        } else {
            return std::make_unique<FragmentedSparseMatrix_internal::SecondaryMyopicBlockDense<Value_, Index_, ValueVectorStorage_, IndexVectorStorage_> >(
                my_values, my_indices, secondary(), block_start, block_end
            );
        }
    }
 
    std::unique_ptr<MyopicDenseExtractor<Value_, Index_> > dense(
        const bool row,
        VectorPtr<Index_> subset_ptr,
        const Options&
    ) const {
        if (my_row_sparse == row) {
            return std::make_unique<FragmentedSparseMatrix_internal::PrimaryMyopicIndexDense<Value_, Index_, ValueVectorStorage_, IndexVectorStorage_> >(
                my_values, my_indices, secondary(), std::move(subset_ptr)
            );
        } else {
            return std::make_unique<FragmentedSparseMatrix_internal::SecondaryMyopicIndexDense<Value_, Index_, ValueVectorStorage_, IndexVectorStorage_> >(
                my_values, my_indices, secondary(), std::move(subset_ptr)
            );
        }
    }
 
    /******************************
     ******* Sparse myopic ********
     ******************************/
private:
    std::unique_ptr<MyopicSparseExtractor<Value_, Index_> > sparse(
        const bool row,
        const Options& opt
    ) const {
        if (my_row_sparse == row) {
            return std::make_unique<FragmentedSparseMatrix_internal::PrimaryMyopicFullSparse<Value_, Index_, ValueVectorStorage_, IndexVectorStorage_> >(
                my_values, my_indices, secondary(), opt
            );
        } else {
            return std::make_unique<FragmentedSparseMatrix_internal::SecondaryMyopicFullSparse<Value_, Index_, ValueVectorStorage_, IndexVectorStorage_> >(
                my_values, my_indices, secondary(), opt
            ); 
        }
    }
 
    std::unique_ptr<MyopicSparseExtractor<Value_, Index_> > sparse(
        const bool row,
        const Index_ block_start,
        const Index_ block_end,
        const Options& opt
    ) const {
        if (my_row_sparse == row) {
            return std::make_unique<FragmentedSparseMatrix_internal::PrimaryMyopicBlockSparse<Value_, Index_, ValueVectorStorage_, IndexVectorStorage_> >(
                my_values, my_indices, secondary(), block_start, block_end, opt
            );
        } else {
            return std::make_unique<FragmentedSparseMatrix_internal::SecondaryMyopicBlockSparse<Value_, Index_, ValueVectorStorage_, IndexVectorStorage_> >(
                my_values, my_indices, secondary(), block_start, block_end, opt
            );
        }
    }
 
    std::unique_ptr<MyopicSparseExtractor<Value_, Index_> > sparse(
        const bool row,
        VectorPtr<Index_> subset_ptr,
        const Options& opt
    ) const {
        if (my_row_sparse == row) {
            return std::make_unique<FragmentedSparseMatrix_internal::PrimaryMyopicIndexSparse<Value_, Index_, ValueVectorStorage_, IndexVectorStorage_> >(
                my_values, my_indices, secondary(), std::move(subset_ptr), opt
            );
        } else {
            return std::make_unique<FragmentedSparseMatrix_internal::SecondaryMyopicIndexSparse<Value_, Index_, ValueVectorStorage_, IndexVectorStorage_> >(
                my_values, my_indices, secondary(), std::move(subset_ptr), opt
            );
        }
    }
 
    /*******************************
     ******* Dense oracular ********
     *******************************/
public:
    std::unique_ptr<OracularDenseExtractor<Value_, Index_> > dense(
        const bool row,
        std::shared_ptr<const Oracle<Index_> > oracle,
        const Options& opt
    ) const {
        return std::make_unique<PseudoOracularDenseExtractor<Value_, Index_> >(std::move(oracle), dense(row, opt));
    }
 
    std::unique_ptr<OracularDenseExtractor<Value_, Index_> > dense(
        const bool row,
        std::shared_ptr<const Oracle<Index_> > oracle,
        const Index_ block_start,
        const Index_ block_end,
        const Options& opt
    ) const {
        return std::make_unique<PseudoOracularDenseExtractor<Value_, Index_> >(std::move(oracle), dense(row, block_start, block_end, opt));
    }
 
    std::unique_ptr<OracularDenseExtractor<Value_, Index_> > dense(
        const bool row,
        std::shared_ptr<const Oracle<Index_> > oracle,
        VectorPtr<Index_> subset_ptr,
        const Options& opt
    ) const {
        return std::make_unique<PseudoOracularDenseExtractor<Value_, Index_> >(std::move(oracle), dense(row, std::move(subset_ptr), opt));
    }
 
    /********************************
     ******* Sparse oracular ********
     ********************************/
public:
    std::unique_ptr<OracularSparseExtractor<Value_, Index_> > sparse(
        const bool row,
        std::shared_ptr<const Oracle<Index_> > oracle,
        const Options& opt
    ) const {
        return std::make_unique<PseudoOracularSparseExtractor<Value_, Index_> >(std::move(oracle), sparse(row, opt));
    }
 
    std::unique_ptr<OracularSparseExtractor<Value_, Index_> > sparse(
        const bool row,
        std::shared_ptr<const Oracle<Index_> > oracle,
        const Index_ block_start,
        const Index_ block_end,
        const Options& opt
    ) const {
        return std::make_unique<PseudoOracularSparseExtractor<Value_, Index_> >(std::move(oracle), sparse(row, block_start, block_end, opt));
    }
 
    std::unique_ptr<OracularSparseExtractor<Value_, Index_> > sparse(
        const bool row,
        std::shared_ptr<const Oracle<Index_> > oracle,
        VectorPtr<Index_> subset_ptr,
        const Options& opt
    ) const {
        return std::make_unique<PseudoOracularSparseExtractor<Value_, Index_> >(std::move(oracle), sparse(row, std::move(subset_ptr), opt));
    }
};
 
template<typename Value_, typename Index_, class ValueVectorStorage_ = std::vector<std::vector<Value_> >, class IndexVectorStorage_ = std::vector<std::vector<Index_> > >
class FragmentedSparseColumnMatrix final : public FragmentedSparseMatrix<Value_, Index_, ValueVectorStorage_, IndexVectorStorage_> {
public:
    FragmentedSparseColumnMatrix(Index_ nrow, Index_ ncol, ValueVectorStorage_ values, IndexVectorStorage_ indices, bool check = true) : 
        FragmentedSparseMatrix<Value_, Index_, ValueVectorStorage_, IndexVectorStorage_>(nrow, ncol, std::move(values), std::move(indices), false, check) {}
};
 
template<typename Value_, typename Index_, class ValueVectorStorage_ = std::vector<std::vector<Value_> >, class IndexVectorStorage_ = std::vector<std::vector<Index_> > >
class FragmentedSparseRowMatrix final : public FragmentedSparseMatrix<Value_, Index_, ValueVectorStorage_, IndexVectorStorage_> {
public:
    FragmentedSparseRowMatrix(Index_ nrow, Index_ ncol, ValueVectorStorage_ values, IndexVectorStorage_ indices, bool check = true) : 
        FragmentedSparseMatrix<Value_, Index_, ValueVectorStorage_, IndexVectorStorage_>(nrow, ncol, std::move(values), std::move(indices), true, check) {}
};
 
 
}
 
#endif