Spatial coordinate indexing dense to sparse

spatial-coordinate-indexing-dense-to-sparse

Summary

Convert dense spatial coordinate matrices into sparse CSR (Compressed Sparse Row) format adjacency and distance matrices for memory-efficient k-nearest-neighbor graph storage and retrieval. This skill is essential for scalable spatial analysis when working with large numbers of observations where most pairwise distances are irrelevant.

When to use

You have computed k-nearest neighbors for spatial coordinates (e.g., via pynndescent or another NN backend) and need to store the resulting adjacency and distance information in a memory-efficient format compatible with downstream graph algorithms. Specifically, when k << n_obs (number of observations), sparse representations dramatically reduce memory footprint while preserving sparsity patterns needed for spatial statistics.

When NOT to use

• Input coordinates are already indexed as a sparse matrix — this skill assumes dense indices and distances from a nearest-neighbor query.
• You need dense pairwise distance matrix for all observation pairs, not just k-nearest neighbors — sparse indexing discards non-neighbor distances intentionally.
• Downstream analysis requires random access to arbitrary (i,j) entries; CSR format is optimized for row-wise iteration, not column-wise or random lookups.

Inputs

• k-nearest neighbor indices (array of shape [n_obs, k])
• k-nearest neighbor distances (array of shape [n_obs, k])
• n_obs: total number of observations (integer)
• k / n_neighs: number of neighbors per observation (integer)
• set_diag: boolean flag indicating whether to set diagonal entries

Outputs

• adj: CSR sparse matrix (float32, shape [n_obs, n_obs]) — adjacency indicator matrix
• dst: CSR sparse matrix (float64, shape [n_obs, n_obs]) — distance matrix
• (adj, dst) tuple ready for graph-based spatial analysis

How to apply

After querying a nearest-neighbor index (e.g., NNDescent with Euclidean metric) for k-nearest neighbors on spatial coordinates, extract the flattened row indices, column indices, and distance values. Construct two CSR sparse matrices: (1) an adjacency matrix (float32) with indicator values of 1.0 at (i, j) positions for each k-neighbor pair, and (2) a distance matrix (float64) with the actual distance values. Set diagonal entries to 1.0 and 0.0 respectively if set_diag=True. Verify that both matrices have shape (n_obs, n_obs), share identical sparsity patterns (same row/column indices), and are stored in CSR format for efficient row-wise access during subsequent spatial operations.

Related tools

• pynndescent (Nearest-neighbor descent backend that queries spatial coordinates and returns k-neighbor indices and distances for input to sparse matrix construction) — https://github.com/lmcinnes/pynndescent
• scipy.sparse (CSR matrix construction and manipulation; provides csr_matrix class for efficient sparse storage)
• numpy (Array operations for flattening indices, creating sparse matrix data arrays, and setting diagonal entries)
• Squidpy (Spatial omics framework that integrates k-NN graph construction via GraphBuilderCSR and PynndescentKNNBuilder subclasses) — https://github.com/scverse/squidpy

Examples

from scipy.sparse import csr_matrix; import numpy as np; indices_flat = np.repeat(np.arange(n_obs), k); cols = knn_indices.flatten(); data_adj = np.ones(len(indices_flat), dtype=np.float32); adj = csr_matrix((data_adj, (indices_flat, cols)), shape=(n_obs, n_obs), dtype=np.float32); dst = csr_matrix((knn_distances.flatten(), (indices_flat, cols)), shape=(n_obs, n_obs), dtype=np.float64)

Evaluation signals

• Adjacency and distance matrices have identical shape (n_obs, n_obs) and identical sparsity patterns (same row/column indices).
• Both matrices are in CSR format (scipy.sparse.csr_matrix); verify via .format property and .has_sorted_indices.
• Adjacency matrix contains only 1.0 values (and 0.0 on diagonal if set_diag=False); distance matrix contains non-negative floats.
• Number of non-zero entries equals n_obs * k (plus n_obs if set_diag=True), reflecting k neighbors per observation.
• Diagonal entries of adj are 1.0 and diagonal entries of dst are 0.0 when set_diag=True; row-wise iteration produces k+1 neighbors per observation.

Limitations

• Sparse indexing discards global distance information; only k-nearest distances are preserved, limiting global distance-based analyses.
• CSR format is optimized for row-wise access; column-wise or random (i,j) lookups require conversion or transposition, incurring overhead.
• Large k relative to n_obs reduces sparsity gains; trade-off between neighborhood completeness and memory efficiency must be evaluated per dataset.

Evidence

• [other] Construct CSR sparse matrices for adjacency (adj, using float32 indicator values of 1.0) and distance (dst, using float64 distance values) from the flattened row/column indices and distances.: "Construct CSR sparse matrices for adjacency (adj, using float32 indicator values of 1.0) and distance (dst, using float64 distance values) from the flattened row/column indices and distances."
• [other] Query the NNDescent model for k-nearest neighbors (with k=n_neighs parameter) and extract indices and distances.: "Query the NNDescent model for k-nearest neighbors (with k=n_neighs parameter) and extract indices and distances."
• [other] Set diagonal values: adj diagonal to 1.0 if set_diag is True else existing values; dst diagonal to 0.0.: "Set diagonal values: adj diagonal to 1.0 if set_diag is True else existing values; dst diagonal to 0.0."
• [other] verify adjacency matrix shape matches (n_obs, n_obs), sparsity pattern is consistent between adj and dst, and matrices are in CSR format.: "verify adjacency matrix shape matches (n_obs, n_obs), sparsity pattern is consistent between adj and dst, and matrices are in CSR format."
• [other] Squidpy builds on scanpy and anndata, providing streamlined APIs for feature extraction and spatial statistics that enable scalable analysis of spatial molecular data.: "Squidpy builds on scanpy and anndata, providing streamlined APIs for feature extraction and spatial statistics that enable scalable analysis of spatial molecular data."