get_from_AnnData
function- option to link nodes of 2 different sets in
build_rdn
node_adaptive_trimming
reconstruction methoddelaunay
method tolink_solitaries
functionbuild_lattice
reconstruction method- use pynndescent for larger network reconstruction with knn
- upgrade compatibility with AnnData
- fix legend in plot_network functions
In v0.5.0 3D network reconstruction is officially supported. The core functions were actually already compatible with n-dimensional networks. Here we added additional support in some data wrangling functions, and a tutorial to show what 3D nets look like in Napari.
v0.4.0 adds functions to help using Napari for interactive visualization and annotation, which comes with an additional tutorial. Plus some code refactoring and more detailed instructions in tutorials and README.
v0.3.0 implements 2 new methods, corrects a bug and improves the project organization.
New features
- The
build_contacting_nn
method allows to link segmented objects that are close to each other, then to link the remaining "lonely" objects to their nearest neighbors in order to avoid too many unconnect nodes in the resulting network - The
build_contacting
method has a parallel version implemented with Dask, in which big images are splitted into an optimal numberof tiles depending on the number of cores.
Fixed bugs:
- The
build_knn
method was actually building networks with (k-1) neighbors as the first k was 'oneself' and was discarded. Now it build networks with k neighboring nodes.
Project organization
- Added poetry project configuration files
pyproject.toml
andpoetry.lock
for better reproducibility, see here for some discussion. A guide for easy installation with all commands coming soon :) - This very CHANGELOG file!
v0.2.0 implement a new method for more accurate network reconstruction, makes the build_contacting
method faster, and add the first functions to help using Napari as an ineractive tool to add and modify network annotations (tutorial coming soon).
New features
- The
build_contacting
method is 19 times faster thanks to OpenCV! - New
percentile_size
method to find the optimal trimming distance for Delaunay triangulation. It is based on the proportion of border edges given the number of nodes, and the proportion of artifactual edges in these border nodes. It was shown to perform very well on realistic simulated images, and it was the best performing method on a real bioimage, see the publcation - Add a set of utilities to help using the Napari interactive image visualization library to add and modify nodes and edges annotations, clean edges data while being able to pan, zoom and hide or vary image channels intensities.
Fixed bugs:
- The
build_contacting
method was not compatibe with images with some missing objects, like segmented objects deleted after filtering by size or other criteria. Now it's solved.