fixed the eMouse script

eMouse_drift/main_eMouse_drift.m

@@ -4,18 +4,18 @@
 sortData = 1;
 runBenchmark = 1; %set to 1 to compare sorted data to ground truth for the simulation
 
-fpath    = 'C:\Users\labadmin\Documents\jic\050320_sim_test\'; % where on disk do you want the simulation? ideally an SSD...
+fpath    = 'G:\Spikes\eMouse\'; % where on disk do you want the simulation? ideally an SSD...
 if ~exist(fpath, 'dir'); mkdir(fpath); end
 
 %KS2 path -- also has the waveforms for the simulation
-KS2path = 'C:\Users\labadmin\Documents\jic\KS2_040920\Kilosort2\';
+KS2path = 'D:\GitHub\KiloSort2\';
 
 % add paths to the matlab path
-addpath(genpath('C:\Users\labadmin\Documents\jic\KS2_040920\Kilosort2\')); % path to kilosort2 folder
-addpath(genpath('C:\Users\labadmin\Documents\jic\npy-matlab-master\'));
+addpath(genpath('D:\GitHub\KiloSort2')) % path to kilosort folder
+addpath('D:\GitHub\npy-matlab') % for converting to Phy
 
 % path to whitened, filtered proc file (on a fast SSD)
-rootH = 'D:\kilosort_datatemp\';
+rootH = 'G:\Spikes\eMouse\';
 
 % path to config file; if running the default config, no need to change.
 pathToYourConfigFile = [KS2path,'eMouse_drift\']; % path to config file
@@ -42,7 +42,7 @@
 end
 %
 % Run kilosort2 on the simulated data
-
+%%
 if( sortData ) 
 
         % common options for every probe
@@ -51,6 +51,10 @@
         ops.NchanTOT    = NchanTOT; % total number of channels in your recording
         ops.trange      = [0 Inf]; % TIME RANGE IN SECONDS TO PROCESS
 
+        ops.sig        = 20;  % spatial smoothness constant for registration
+        ops.fshigh     = 300; % high-pass more aggresively
+        ops.nblocks = 5; % blocks for registration. 0 turns it off, 1 does rigid registration. Replaces "datashift" option.
+
 
         ops.fproc       = fullfile(rootH, 'temp_wh.dat'); % proc file on a fast SSD
 
@@ -64,27 +68,25 @@
         % preprocess data to create temp_wh.dat
         rez = preprocessDataSub(ops);
 
-        % pre-clustering to re-order batches by depth
-        rez = clusterSingleBatches(rez);
+        % NEW STEP TO DO DATA REGISTRATION
+        rez = datashift2(rez, 1); % last input is for shifting data
+%         rez2 = datashift2(rez, 0); % last input is for shifting data
+
 
         % main optimization
-        % learnAndSolve8;
-        rez = learnAndSolve8b(rez);
+        rez = learnAndSolve8b(rez, 1);
 
-
         % final splits
         rez = find_merges(rez, 1);
 
 
         % final splits by SVD
         rez    = splitAllClusters(rez, 1);
 
-        % final splits by amplitudes
-        rez = splitAllClusters(rez, 0);
-
         % decide on cutoff
         rez = set_cutoff(rez);
-
+        rez.good = get_good_units(rez);
+
         % this saves to Phy
         rezToPhy(rez, rootZ);
 
@@ -108,7 +110,7 @@
         fclose(fileID);
 
         % remove temporary file
-        delete(ops.fproc);
+%         delete(ops.fproc);
 end
 
 

eMouse_drift/make_eMouseData_drift.m

@@ -5,8 +5,8 @@ function make_eMouseData_drift(fpath, KS2path, chanMapName, useGPU, useParPool)
 % probe sites.
 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 % you can play with the parameters just below here to achieve a signal more similar to your own data!!! 
-norm_amp  = 16.7; % if 0, use amplitudes of input waveforms; if > 0, set all amplitudes to norm_amp*rms_noise
-mu_mean   = 0.75; % mean of mean spike amplitudes. Incoming waveforms are in uV; make <1 to make sorting harder
+norm_amp  = 1.5 * 16.7; % if 0, use amplitudes of input waveforms; if > 0, set all amplitudes to norm_amp*rms_noise
+mu_mean   = .75; %0.75; % mean of mean spike amplitudes. Incoming waveforms are in uV; make <1 to make sorting harder
 noise_model = 'gauss'; %'gauss' or 'fromData'; 'fromData' requires a noiseModel.mat built by make_noise_model
 rms_noise = 10; % rms noise in uV. Will be added to the spike signal. 15-20 uV an OK estimate from real data
 t_record  = 1200; % duration in seconds of simulation. longer is better (and slower!) (1000)
@@ -24,7 +24,7 @@ function make_eMouseData_drift(fpath, KS2path, chanMapName, useGPU, useParPool)
 drift.y0 = 3800;        %in um, position along probe where motion is largest
                         %y = 0 is the tip of the probe                        
 drift.halfDistance = 1000;   %in um, distance along probe over which the motion decays
-drift.amplitude = 10;        %in um for a sine wave
+drift.amplitude = 5;        %in um for a sine wave
 %                             peak variation is 2Xdrift.amplitude
 drift.halfLife = 2;     %in seconds
 drift.period = 600;      %in seconds

preProcess/datashift2.m

@@ -23,7 +23,7 @@
 npt = floor(xrange/16); % this would come out as 16um for Neuropixels probes, which aligns with the geometry. 
 rez.ops.xup = linspace(xmin, xmax, npt+1); % centers of the upsampled x positions
 
-spkTh = 10; % same as the usual "template amplitude", but for the generic templates
+spkTh = 8; % same as the usual "template amplitude", but for the generic templates
 
 % Extract all the spikes across the recording that are captured by the
 % generic templates. Very few real spikes are missed in this way.