- create
computation
argument, encourageapprox
/use_annoy
/use_faiss
users to move to it - simplify per-batch query logic (C. Xu)
- add
key_added
argument to mirrorscanpy.pp.neighbors()
- update setup to flit, add missing pandas dependency
- update sklearn syntax to match >=1.0.2, require appropriate version (Matchy, R. Williams)
- update scipy syntax to match >=1.6.0, require appropriate version
- remove pre-0.7.0 anndata support
- internal coding consistency improvements and readthedocs modernisation
- correctly check for pynndescent metric space in
bbknn.bbknn()
- pynndescent support (suggested by G. Palla)
- default metric set to Euclidean
- code refactoring; impactful changes:
- innards relocated to
bbknn.matrix
,bbknn.bbknn_pca_matrix()
is nowbbknn.matrix.bbknn()
n_trees
is nowannoy_n_trees
- innards relocated to
- remove trimming bug which would lead to an asymmetric matrix for cells with a neighbour count very near the threshold (spotted by E. Dann)
- clarify docstring, neighbour counts are there for the initial KNN computation
- add ridge regression preprocessing like in [https://science.sciencemag.org/content/367/6480/eaay3224.abstract](Park, 2020)
- require batches to feature a minimum of
neighbors_within_batch
cells each - rework demo notebook, remove old preprint notebooks to separate repository
- add packaging as a dependency in
setup.py
- fix version parsing for newer versions of
anndata
andpackaging
- store the computed graphs in
.obsp
if anndata is 0.7 or newer - include additional parameter information in
.uns['neighbors']
- return
.uns['neighbor']
-style parameter dictionary as third tuple element frombbknn_pca_matrix
- sklearn is actually scikit-learn for dependency purposes (D. DeTomaso)
- support umap-learn 0.4 connectivity output in a backward compatible manner (I. Driver)
- support for non-PCA dimensionality reductions on input (brought to attention by D. Cittaro)
- further updated scanpy logging compatibility tweak (F. Rost)
- update logging to match scanpy 1.4.4 standards
- rework
.obsm['X_pca']
check to be compatible with new structure (I. Virshup)
- add MANIFEST.in for conda purposes (J. Preußner)
- run trimming by default at 10 times the neighbour count for each cell
- remove scanpy dependency from
bbknn_pca_matrix
; rework setup dependencies as a result - removal of
save_knn
option - creation of diagnostic
extract_cell_connectivity
function for plotting connectivities
- metric sanity check typo fix
- optional faiss support
- default swap to annoy neighbours with angular as the metric, add annoy's
n_trees
parameter - removal of distance scaling
- removal of
n_jobs
parameter - automatically parallelise cKDTree - metric sanity checks
bbknn_pca_matrix
function- code refactoring
- annoy support
- initial release