Improve selection of units per file considered by SWAN based on all metadata

[mdenker]

Data files typically contain not only perfect single units (ie, units with high SNR coming very likely from one neuron), but often also multi-unit activity or noise. In many data formats, such extra information is not available. E.g., the Blackrock format, especially when channeld through, e.g., Plexon Offline Spikesorter with Blackrock as Output, has only rough indications based on the unit ID being 0 or 255. In essence, for this reason there is no uniform standard to indicate such properties for a given unit/spiketrain.

The current implementation of SWAN is based on an older version of the BlackrockIO of Neo, which labeled channel 255 as noise and channel 0 as `unclassified' according to the file format specification. SWAN discarded those channels. However, this is not valid for other file formats, and more so, this is not indicative of SUA/MUA distrinctions (reasoning being that MUAs should probably be excluded from SWAN analysis).

In the absence of a defined field in Neo to label SUA/MUA/noise/etc, the following would be the suggestion as a flexible way to improve usability and interoperability:

• Remove hard-coded (arbitrary) removal of noise/unclassified units in neodata.py.
• Add a mechanism to see a unit's description, a spiketrain's descritption, and their annotations in the GUI. This would reveal all metainformation available from the file. In particular, using nix files, users could prepare their files with all relevant information.
• Provide a mechanism to manually remove a unit from analysis
• Provide a mechanism to filter units from analysis based on selection criteria (e.g., via a json syntax). This filter mechanism should be closely tied to the revamping of the filter function in neo.

[mdenker]

This issue should be revisted for 0.2.0. Most likely these steps will be moved outside of swan (during data generation)