Releases: JaneliaSciComp/JRCLUST
v3.1.7
Bugfix: Feature projection view
Correct y-axis values (PC2) are plotted when splitting on the diagonal
site pairs when PCA feature is selected.
Speed improvement: fft_clean operation (fft_thresh parameter) is performed faster.
New feature: Auto-splitting now supports manual splitting in three PC dimensions.
Auto-split now offers three choices: {'yes', 'no', 'manual'}.
Select "manual" and choose the projection ('PC1 vs PC2', 'PC3 vs PC2', 'PC1 vs PC3').
v3.1.6
[2017/11/20: v3.1.6]
Bugfix: Nested S0 struct is removed when older version is loaded.
Drift correction is improved. Three average waveforms are computed by grouping
spikes from similar center-of-mass. Now using all spikes from the same unit
previously used spikes that were centered at the peak site only.
New parameter: fSpatialMask_clu (1 by default) applies spatial masking to
the features. Half decay length set by the maxDist_site_um.
Added support to import hdf5 dataset from the Ed Boyden lab recordings for
paired intracellular recording validation.
v3.1.5
[2017/11/5: v3.1.5]
Bugfix: 'jrc describe' failed to report clustering time
Bugfix: Probe file backward compatibility added
Failed if 'maxSite' exists but 'nSites_ref' doesn't exist in the probe file (.prb)
Now nSites_ref is computed based on 'maxDist_site_spk_um' (see wiki)
Probe files are first searched in ./prb/ location and elsewhere if not found
Bugfix: 'jrc update' command now copies sub-folders and backup files correctly
Previously subfolders didn't get copied and files were not back-up properly.
Run 'jrc update' twice to get proper updates, since you need to run a new update script.
v3.1.4
JRCLUST repository moved to JanalieSciComp
https://github.com/JaneliaSciComp/JRCLUST
v3.1.3
v3.1.2
v3.1.1
[2017/10/27: v3.1.1]
Bugfix: PSTH time offset error is removed (0.01 s previously)
Manual GUI: Speed improvement when selecting a unit.
Bugfix: raw waveform display fixed for the manual GUI.
Waveforms did not display if you used an older .prm file
New paramter: fInterp_fet = 1 (default), set to 0 to disable this new feature since v3.0.8
This interpolates the spike waveforms 2x and finds the optimal time-lag to project the principal vector.
New paramter: fDrift_merge = 1 (default). Set to 0 to disable this new feature since v3.0.8
This compares mean unit waveforms at three drift locations based on the spike position (CoM).
New paramter: nInterp_merge = 2 (default), set to 1 to disable this new feature since v3.0.8
This compares mean unit waveforms at three drift locations based on the spike position (CoM).
Bugfix: Annotations previously got deleted when deleting clusteres using "delete-auto" command.
Manual GUI: "auto-merge" to automatically merge units based on the waveform correlations.
v3.1.0
Bugfix: raw waveform display fixed for the manual GUI.
Waveforms did not display if you used an older .prm file
'fir1' filter is supported for "vcFilter" parameter.
Finite Impulse Response filter
The frequency range is sett by "freqLim=[f_low, f_high]" parameter.
New default: Detection threshold set to "qqFactor = 5".
Increasing the detection threshold can incorrectly merge large amplitude units with background noise.
New default: Auto-calculation of maxSite and nSites_ref based on the physical distance.
maxSite=[] and nSites_ref=[] will automatically set it based on the site spacing and layout.
The parameter "maxDist_site_um" is used to determine the number of sites used for features (default 50 um)
The parameter "maxDist_site_spk_um" is used to determine the number of reference sites (default 75 um)
v3.0.9
[2017/10/25: v3.0.9]
Faster auto-splitting. Now displaying three principal component axis.
Individual spike waveforms are no longer displayed but mean waveforms are displayed
nSites_ref (reference sites) are not included in the auto-splitting projection
Faster updates for manual GUI. Refresh speed increased when user selects a unit.
Bugfix: Waveform correlation matrix error fixed.
v3.0.8
[2017/10/24: v3.0.8]
Faster manual GUI loading speed. Optimized caching
Improved drift resistance. Three average unit waveforms are computed based on the vertical spike positions,
and waveform similarities are compared for spatial and temporal shifts.
Faster post-hoc automated merging by using parfor (parallel computing toolbox)
Bugfix: manual GUI waveform similiarity matrix calculation.
Bugfix: Removed a horizontal curve from the rho-delta plot.
Bugfix: Manual GUI will clear the unsaved results when user selects "do not save".
Previously cached information was incorrectly shown when the manual GUI was restarted.
New default: Raw spike waveform duration is now the same as the filtered spike waveform length.
spkLim_ms = [-.25, .75]; sets the spike waveform range to 0.25 ms before the negative peak and 0.75 ms after the peak.
Setting spkLim_raw_ms = []; will set spkLim_raw_ms = spkLim_ms.
[2017/10/9]
Default changed: nPcPerChan=1 (previously 3), since other PC dimensions explain less than
2% of the total variance when differentiator filter is used. Using only the first PC is
also more robust to the presence of overlapping spikes.
New default filter: vcFilter = 'ndiff' (convolves with [1,2,-2,-1] if nDiff_filt=2 chosen).
The new differentiator filter is less attenuating than 'sgdiff' for small spikes,
which convolves with [2,1,0,-1,-2] if nDiff_filt=2 is chosen.
Default changed: fRepeat_clu=0 (previously 1)
Repeating clustering is not necessary if lower feature dimension is used (nPcPerChan=1).
Default changed: delta1_cut=0.6 (previously 0.5) if lower feature dimension is used.
Improved auto-merging based on mean unit waveforms using interpolation
Unit waveform similarity matrix is computed by finding the maximum correlation score
by relatively time-shifting waveform pairs after interpolating 4x in time.
Bugfix: Diagonal waveform similarity is calculated correctly after merge and subtract
Reduced the peak memory use by 50%. Spike waveforms are directly saved to the disk.
fRamCache = 1; is set to default, which stores both the raw and filtered spike waveforms in RAM.
Setting "fRamCache = 0" will only load either raw or filtered spike waveforms in RAM to save space.