files for friday

programming/exercises/STA/sta.m

@@ -0,0 +1,18 @@
+function [sta, std_sta, valid_spikes]= sta(stimulus, spike_times, count, sampling_rate)
+
+snippets = zeros(numel(spike_times), 2*count);
+valid_spikes = 1;
+for i = 1:numel(spike_times)
+    t = spike_times(i);
+    index = round(t*sampling_rate);
+    if index < count || (index + count) > length(stimulus)
+        continue
+    end
+    snippets(valid_spikes,:) = stimulus(index-count:index+count-1);
+    valid_spikes = valid_spikes + 1;
+end
+
+snippets(end-(end-valid_spikes):end,:) = [];
+
+sta = mean(snippets, 1);
+std_sta = std(snippets,[],1);

programming/exercises/STA/sta_script.m

@@ -0,0 +1,62 @@
+clear 
+clc
+
+%% load data
+load p-unit_spike_times.mat
+load p-unit_stimulus.mat
+sample_rate = 20000;
+
+%% calculate STA
+all_times = [];
+for i = 1:length(spike_times)
+    all_times = cat(1, all_times, spike_times{i});
+end
+
+[sta, sta_sd, num] = sta(stimulus_strong(:,2), all_times, 1000, sample_rate);
+fig = figure();
+set(fig, 'PaperUnits', 'centimeters');
+set(fig, 'PaperSize', [11.7 9.0]);
+set(fig, 'PaperPosition',[0.0 0.0 11.7 9.0]);
+set(fig,'Color', 'white')
+plot(((1:length(sta))-1000)/sample_rate, sta)
+xlabel('time [s]')
+ylabel('stimulus')
+
+%% reverse reconstruction
+
+% make binary representation of the spike times
+binary_spikes = zeros(size(stimulus_strong, 1), length(spike_times));
+estimated_stims = zeros(size(binary_spikes));
+for i = 1:length(spike_times)
+    binary_spikes(round(spike_times{i}*sample_rate), i) = 1;
+    estimated_stims(:,i) = conv(binary_spikes(:,i), sta, 'same');
+end
+
+fig = figure();
+set(fig, 'PaperUnits', 'centimeters');
+set(fig, 'PaperSize', [11.7 9.0]);
+set(fig, 'PaperPosition',[0.0 0.0 11.7 9.0]);
+set(fig,'Color', 'white')
+plot(stimulus_strong(:,1), stimulus_strong(:,2), 'displayname','original')
+hold on
+plot(stimulus_strong(:,1), mean(estimated_stims,2), 'r', 'displayname', 'reconstruction')
+xlabel('time [s]')
+ylabel('stimulus')
+legend show
+
+%% calculate STC
+
+% we need to downsample the data otherwise the covariance matrixs gets too
+% large 20Khz to 1kHz
+
+downsampled_binary = zeros(size(stimulus_strong, 1)/20, length(spike_times));
+downsampled_stim = zeros(size(downsampled_binary,1),1);
+
+for i = 1:length(spike_times)
+    binary_spikes(round(spike_times{i}*1000), i) = 1;
+end
+for i = 1:length(downsampled_stim)
+    start_index = (i-1) * 1000 + 1;
+    downsampled_stim(i) = mean(stimulus_strong()*20))
+end
+