[r][cpp] Add sparse = TRUE as an option for pseudobulk_matrix()
#268
There are no files selected for viewing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,162 @@ | ||
|
|
||
| #include "PseudobulkSparse.h" | ||
| namespace BPCells { | ||
|
|
||
| template <typename T> | ||
| PseudobulkStatsSparse pseudobulk_matrix_sparse(std::unique_ptr<MatrixLoader<T>> &&mat, | ||
| const std::vector<uint32_t>& cell_groups, | ||
| PseudobulkStatsMethod method, | ||
| bool transpose, | ||
| std::atomic<bool> *user_interrupt) { | ||
| MatrixIterator<T> it(std::move(mat)); | ||
| if (transpose && (it.rows() != cell_groups.size())) { | ||
| throw std::invalid_argument("pseudobulk_matrix_sparse(): Cell groups must match the number of columns in the matrix"); | ||
| } else if (!transpose && (it.cols() != cell_groups.size())) { | ||
| throw std::invalid_argument("pseudobulk_matrix_sparse(): Cell groups must match the number of columns in the matrix"); | ||
| } | ||
| // Count each group | ||
| std::vector<double> group_count_vec; | ||
| uint32_t last_cell = 0; | ||
| for (auto& cell_group : cell_groups) { | ||
| if (cell_group < last_cell) { | ||
| throw std::invalid_argument("pseudobulk_matrix_sparse(): Cell groups must be ordered"); | ||
| } | ||
| last_cell = cell_group; | ||
| if (cell_group >= group_count_vec.size()) group_count_vec.resize(cell_group + 1); | ||
| group_count_vec[cell_group]++; | ||
| } | ||
| Eigen::Map<Eigen::Array<double, 1, Eigen::Dynamic>> group_count(group_count_vec.data(), group_count_vec.size()); | ||
| // If transposed, also keep matrix transposed with groups as rows for cache-friendliness | ||
| uint32_t num_rows = transpose ? it.cols() : it.rows(); | ||
| uint32_t num_cols = group_count_vec.size(); | ||
| uint32_t tmp_size; | ||
| if (transpose) { | ||
| tmp_size = num_cols; | ||
| } else { | ||
| tmp_size = num_rows; | ||
| } | ||
| struct PseudobulkStatsSparse res; | ||
| struct PseudobulkStatsTriplet trip; | ||
| struct PseudobulkStatsTemp tmp; | ||
|
Collaborator
There was a problem hiding this comment. More comments on what these are used for |
||
| if ((method & PseudobulkStatsMethod::NonZeros) == PseudobulkStatsMethod::NonZeros) { | ||
| res.non_zeros = Eigen::SparseMatrix<double>(num_rows, num_cols); | ||
| trip.non_zeros.reserve(num_rows * num_cols); | ||
| tmp.non_zeros = std::vector<double>(tmp_size, 0.0); | ||
| } | ||
| if ((method & PseudobulkStatsMethod::Sum) == PseudobulkStatsMethod::Sum) { | ||
| res.sum = Eigen::SparseMatrix<double>(num_rows, num_cols); | ||
| trip.sum.reserve(num_rows * num_cols); | ||
| tmp.sum = std::vector<double>(tmp_size, 0.0); | ||
| } | ||
| if ((method & PseudobulkStatsMethod::Mean) == PseudobulkStatsMethod::Mean) { | ||
| res.mean = Eigen::SparseMatrix<double>(num_rows, num_cols); | ||
| trip.mean.reserve(num_rows * num_cols); | ||
| tmp.non_zeros = std::vector<double>(tmp_size, 0.0); | ||
| tmp.mean = std::vector<double>(tmp_size, 0.0); | ||
| } | ||
| if ((method & PseudobulkStatsMethod::Variance)== PseudobulkStatsMethod::Variance) { | ||
| res.var = Eigen::SparseMatrix<double>(num_rows, num_cols); | ||
| trip.var.reserve(num_rows * num_cols); | ||
| tmp.non_zeros = std::vector<double>(tmp_size, 0.0); | ||
| tmp.mean = std::vector<double>(tmp_size, 0.0); | ||
| tmp.var = std::vector<double>(tmp_size, 0.0); | ||
| } | ||
| uint32_t ncells = 0; | ||
| uint32_t g_idx = 0; | ||
| // Use a one pass strategy that can do non-zero, mean, and variance at the same time | ||
| while (it.nextCol()) { | ||
| const uint32_t col = it.currentCol(); | ||
| if (user_interrupt != NULL && *user_interrupt) return res; | ||
| while (it.load()) { | ||
| const uint32_t *row_data = it.rowData(); | ||
| const T *val_data = it.valData(); | ||
| const uint32_t count = it.capacity(); | ||
| // Do transpose logic to calculate row_idx, col_idx out of loop, to avoid repeating conditional checks | ||
| if (transpose) { | ||
| // Doing variance calculation also does mean and non-zero count, so specific order is required to make sure we don't double count | ||
| if ((method & PseudobulkStatsMethod::Variance) == PseudobulkStatsMethod::Variance) { | ||
| for (uint32_t i = 0; i < count; i++) update_mean_and_variance(val_data[i], tmp.non_zeros[cell_groups[row_data[i]]], tmp.mean[cell_groups[row_data[i]]], tmp.var[cell_groups[row_data[i]]]); | ||
| } else if ((method & PseudobulkStatsMethod::Mean) == PseudobulkStatsMethod::Mean) { | ||
| for (uint32_t i = 0; i < count; i++) update_mean(val_data[i], tmp.non_zeros[cell_groups[row_data[i]]], tmp.mean[cell_groups[row_data[i]]]); | ||
| } else if ((method & PseudobulkStatsMethod::NonZeros) == PseudobulkStatsMethod::NonZeros) { | ||
| for (uint32_t i = 0; i < count; i++) tmp.non_zeros[cell_groups[row_data[i]]] += 1; | ||
| } | ||
| if ((method & PseudobulkStatsMethod::Sum) == PseudobulkStatsMethod::Sum) { | ||
| for (uint32_t i = 0; i < count; i++) tmp.sum[cell_groups[row_data[i]]] += val_data[i]; | ||
| } | ||
| } else { | ||
| if ((method & PseudobulkStatsMethod::Variance) == PseudobulkStatsMethod::Variance) { | ||
| for (uint32_t i = 0; i < count; i++) update_mean_and_variance(val_data[i], tmp.non_zeros[row_data[i]], tmp.mean[row_data[i]], tmp.var[row_data[i]]); | ||
| } else if ((method & PseudobulkStatsMethod::Mean) == PseudobulkStatsMethod::Mean) { | ||
| for (uint32_t i = 0; i < count; i++) update_mean(val_data[i], tmp.non_zeros[row_data[i]], tmp.mean[row_data[i]]); | ||
| } else if ((method & PseudobulkStatsMethod::NonZeros) == PseudobulkStatsMethod::NonZeros) { | ||
| for (uint32_t i = 0; i < count; i++) tmp.non_zeros[row_data[i]] += 1; | ||
| } | ||
| if ((method & PseudobulkStatsMethod::Sum) == PseudobulkStatsMethod::Sum) { | ||
| for (uint32_t i = 0; i < count; i++) tmp.sum[row_data[i]] += val_data[i]; | ||
| } | ||
| } | ||
| } | ||
| if (transpose) { | ||
| for (uint32_t i = 0; i < tmp.var.size(); i++) { | ||
| if (tmp.mean[i] > 0) trip.var.emplace_back(col, i, (tmp.var[i] + (tmp.mean[i] * tmp.mean[i] * tmp.non_zeros[i] * ((group_count_vec[i] - tmp.non_zeros[i]) / group_count_vec[i]))) / (group_count_vec[i] - 1)); | ||
| } | ||
| for (uint32_t i = 0; i < tmp.mean.size(); i++) { | ||
| if (tmp.mean[i] > 0) trip.mean.emplace_back(col, i, tmp.mean[i] * tmp.non_zeros[i] / group_count_vec[i]); | ||
| } | ||
| for (uint32_t i = 0; i < tmp.non_zeros.size(); i++) { | ||
| if (tmp.non_zeros[i] > 0) trip.non_zeros.emplace_back(col, i, tmp.non_zeros[i]); | ||
| } | ||
| for (uint32_t i = 0; i < tmp.sum.size(); i++) { | ||
| if (tmp.sum[i] > 0) trip.sum.emplace_back(col, i, tmp.sum[i]); | ||
| } | ||
| std::fill(tmp.var.begin(), tmp.var.end(), 0.0); | ||
| std::fill(tmp.mean.begin(), tmp.mean.end(), 0.0); | ||
| std::fill(tmp.sum.begin(), tmp.sum.end(), 0.0); | ||
| std::fill(tmp.non_zeros.begin(), tmp.non_zeros.end(), 0.0); | ||
| continue; | ||
| } | ||
| // transpose == false | ||
| ncells += 1; | ||
| if (ncells < group_count_vec[g_idx]) continue; | ||
| for (uint32_t i = 0; i < tmp.var.size(); i++) { | ||
|
Collaborator
There was a problem hiding this comment. Unclear why we would do this at a first glance (even though I wrote the original code for the corrections for sparsity). Means we need some comments! |
||
| if (tmp.mean[i] > 0) trip.var.emplace_back(i, g_idx, (tmp.var[i] + (tmp.mean[i] * tmp.mean[i] * tmp.non_zeros[i] * (group_count_vec[g_idx] - tmp.non_zeros[i]) / group_count_vec[g_idx])) / (group_count_vec[g_idx] - 1)); | ||
| } | ||
| for (uint32_t i = 0; i < tmp.mean.size(); i++) { | ||
| if (tmp.mean[i] > 0) trip.mean.emplace_back(i, g_idx, tmp.mean[i] * tmp.non_zeros[i] / group_count_vec[g_idx]); | ||
| } | ||
| for (uint32_t i = 0; i < tmp.non_zeros.size(); i++) { | ||
| if (tmp.non_zeros[i] > 0) trip.non_zeros.emplace_back(i, g_idx, tmp.non_zeros[i]); | ||
| } | ||
| for (uint32_t i = 0; i < tmp.sum.size(); i++) { | ||
| if (tmp.sum[i] > 0) trip.sum.emplace_back(i, g_idx, tmp.sum[i]); | ||
| } | ||
| std::fill(tmp.var.begin(), tmp.var.end(), 0.0); | ||
| std::fill(tmp.mean.begin(), tmp.mean.end(), 0.0); | ||
| std::fill(tmp.sum.begin(), tmp.sum.end(), 0.0); | ||
| std::fill(tmp.non_zeros.begin(), tmp.non_zeros.end(), 0.0); | ||
| g_idx += 1; | ||
| ncells = 0; | ||
| } | ||
| if ((method & PseudobulkStatsMethod::NonZeros) == PseudobulkStatsMethod::NonZeros) res.non_zeros.setFromTriplets(trip.non_zeros.begin(), trip.non_zeros.end()); | ||
| if ((method & PseudobulkStatsMethod::Sum) == PseudobulkStatsMethod::Sum) res.sum.setFromTriplets(trip.sum.begin(), trip.sum.end()); | ||
| if ((method & PseudobulkStatsMethod::Mean) == PseudobulkStatsMethod::Mean) res.mean.setFromTriplets(trip.mean.begin(), trip.mean.end()); | ||
| if ((method & PseudobulkStatsMethod::Variance) == PseudobulkStatsMethod::Variance) res.var.setFromTriplets(trip.var.begin(), trip.var.end()); | ||
| return res; | ||
| }; | ||
| template PseudobulkStatsSparse pseudobulk_matrix_sparse<uint32_t>(std::unique_ptr<MatrixLoader<uint32_t>>&& mat, | ||
| const std::vector<uint32_t>& cell_groups, | ||
| PseudobulkStatsMethod method, | ||
| bool transpose, | ||
| std::atomic<bool> *user_interrupt); | ||
| template PseudobulkStatsSparse pseudobulk_matrix_sparse<float>(std::unique_ptr<MatrixLoader<float>>&& mat, | ||
| const std::vector<uint32_t>& cell_groups, | ||
| PseudobulkStatsMethod method, | ||
| bool transpose, | ||
| std::atomic<bool> *user_interrupt); | ||
| template PseudobulkStatsSparse pseudobulk_matrix_sparse<double>(std::unique_ptr<MatrixLoader<double>>&& mat, | ||
| const std::vector<uint32_t>& cell_groups, | ||
| PseudobulkStatsMethod method, | ||
| bool transpose, | ||
| std::atomic<bool> *user_interrupt); | ||
| } // namespace BPCells | ||
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,35 @@ | ||
| #pragma once | ||
| #include "Pseudobulk.h" | ||
| #include <unordered_map> | ||
|
|
||
| namespace BPCells { | ||
|
|
||
| struct PseudobulkStatsSparse { | ||
|
Collaborator
There was a problem hiding this comment. Docstrings would be helpful, why do we need all three of these new structs? I would even throw |
||
| Eigen::SparseMatrix<double> non_zeros; | ||
| Eigen::SparseMatrix<double> sum; | ||
| Eigen::SparseMatrix<double> mean; | ||
| Eigen::SparseMatrix<double> var; | ||
| }; | ||
|
|
||
| struct PseudobulkStatsTriplet { | ||
| std::vector<Eigen::Triplet<double>> non_zeros; | ||
| std::vector<Eigen::Triplet<double>> sum; | ||
| std::vector<Eigen::Triplet<double>> mean; | ||
| std::vector<Eigen::Triplet<double>> var; | ||
| }; | ||
|
|
||
| struct PseudobulkStatsTemp { | ||
| std::vector<double> non_zeros; | ||
| std::vector<double> sum; | ||
| std::vector<double> mean; | ||
| std::vector<double> var; | ||
| }; | ||
|
|
||
| template <typename T> | ||
| PseudobulkStatsSparse pseudobulk_matrix_sparse(std::unique_ptr<MatrixLoader<T>> &&mat, | ||
| const std::vector<uint32_t>& cell_groups, | ||
| PseudobulkStatsMethod method, | ||
| bool transpose, | ||
| std::atomic<bool> *user_interrupt); | ||
|
|
||
| } // namespace BPCells | ||
Oops, something went wrong.
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.