Changelog¶

0.5 / 2024-12-11¶

Now compatible with numpy 2.x
Fixed various warnings that occur with pandas 2.x
Minimum supported Python version is now explicitly listed as 3.9
Add missing client arguments in various places instead of using the default. (PR #58 and related commits) This is crucial if multiple clients have been constructed.
fetch_mean_synapses(): Added by_roi option to allow the user to fetch whole-neuron mean synapses
fetch_shorted_paths() allows you to omit filtering entirely using NC()
fetch_neurons(): If no NeuronCriteria is provided, fetch all ``:Neuron``s by default
Added available_datasets to utils.py (PR #60)
Internally generated Cypher now uses backticks for variables/properties that require them. (PR #42 and related commits)
Bug fix in connection_table_to_matrix() (PR #47)
Bug fix in fetch_common_connectivity() (PR #63)
Several other bug fixes
For developers: Added basic pixi configuration

NeuronCriteria now supports many new properties for the MANC v1.0 dataset.
Neuron property columns are determined from cached metadata rather than a full scan of the database.
If more than one Client has been constructed, none of them become automatically become the default client. In that case, you must explicitly pass a client argument to each query function you call. This avoids a common pitfall when dealing with multiple neuprint datasets (and therefore multiple Clients).
SynapseCriteria now uses a default confidence threshold based on the dataset metadata (instead of using 0.0 by default)
Client constructor avoids contacting the database unless it needs to. (Duplicate clients are now cheaper to construct.)
Minor enhancements to skeleton utilities, including a couple new analysis functions.
Added CITATION.cff

In live-updated neuprint databases, it is possible that an edge’s weight can become out-of-sync with its roiInfo totals. That inconsistency triggered an assertion in fetch_adjacencies(), but now it will emit a warning instead.

Implemented a workaround to avoid a pandas bug in certain cases involving empty dataframes.

In fetch_adjacencies() (and fetch_simple_connections()), we now ensure that no 0-weight “connections” are returned.

Note

In recent neuprint databases, some :ConnectsTo relationships may have a weight of 0. In such cases, the relationship will have a non-zero weightHR (high-recall weight), but all of the relevant synapses are low-confidence, hence the “default” weight of 0. We now exclude such connections from our results.

Fixed a crash that could occur if you supplied more than three regular expressions for type or instance.
Fixed a problem involving ‘hidden’ ROIs in the hemibrain v1.0.

Now heal_skeleton() is slightly faster in the case where no healing was necessary.

By default, NeuronCriteria will now guess whether the type and instance contain a regular expression or not, so you don’t need to explicitly pass regex=True. Override the guess by specifying regex=True or regex=False.

[CHANGE IN RETURNED RESULTS] fetch_synapse_connections() now applies ROI filtering criteria to only the post-synaptic points, for consistency with fetch_adjacencies(). (See note in the docs.) This means that the number of synapses returned by fetch_synapse_connections() is now slightly different than it was in previous versions of neuprint-python.
In fetch_neurons(), better handling of old neuprint datasets which lack some fields (e.g. upstream, downstream, mito).

NeuronCriteria has new fields to support upcoming datasets: somaSide, class_, statusLabel, hemilineage, exitNerve.
NeuronCriteria now permits you to search for neurons that contain (or lack) a particular property via a special value NotNull (or IsNull).
fetch_neurons() now returns all neuron properties.
fetch_neurons() now returns special rows for NotPrimary connection counts.
The per-ROI connection counts table returned by fetch_neurons() now includes rows for connections which fall outside of all primary ROIs. These are indicated by the special ROI name NotPrimary.
fetch_synapse_connections() uses a more fine-grained batching strategy, splitting the query across more requests to avoid timeouts.
Fixed a bug in fetch_shortest_paths() which caused it to generate invalid cypher if the intermediate_criteria used a list of bodyIds (or statuses, or rois, etc.) with more than three items.
fetch_output_completeness now accepts a list of statuses to use, rather than assuming only "Traced" neurons are complete.
Added utility function skeleton_segments().

NeuronCriteria now accepts a boolean argument for soma, indicating the presence or absence of a soma on the body.
Added fetch_connection_mitochondria() for finding the nearest mitochondria on both sides of a tbar/psd pair. (#24)
Integrated with Zenodo for DOI generation.

Updated to changes in the neuPrint mitochondria data model. Older versions of neuprint-python cannot query for mitochondria any more.
fetch_neurons(): Added new columns to the roi_counts_df result, for upstream, downstream, mito
fetch_skeletons(): Now supports with_distances option
NeuronCriteria permits lists of strings for type/instance regular expressions. (Previously, lists were only permitted when regex=False.)
Fixed a performance problem in fetch_synapse_connections()
More FAQ entries

SynapseCriteria: Changed the default value of primary_only to True, since it may been counter-intuitive to obtain duplicate results by default.
NeuronCriteria: Added cellBodyFiber parameter. (Philipp Shlegel #13)
Added mitochondria queries

Moved skeleton-related functions into their own module, and added a few more skeleton utilty functions
Simulation: Support Windows
heal_skeleton(): Allow caller to specify a maximum distance for repaired skeleton segments (#12)