12345678910111213141516171819202122232425262728293031323334353637383940414243444546 |
- function [LH, probSpike, V, mean_predictedSpikes, RPE] = ott_RW_V_base_asymm(startValues, spikeCounts, rewards, timeLocked)
- % reward is 0: mal, 1: suc, 2: water
- alphaPPE = startValues(1);
- alphaNPE = startValues(2);
- slope = startValues(3);
- intercept = startValues(4);
- rho = startValues(5); % how valuable is maltodextrin on a water -> malto -> sucrose scale
- Vinit = 2/3*(alphaPPE / (alphaPPE + alphaNPE)) + 1/3*rho;
- water_ind = rewards == 2;
- mal_ind = rewards == 0;
- rewards(water_ind) = 0;
- rewards(mal_ind) = rho; % scale mal between 0 and 1
- trials = length(rewards);
- V = zeros(trials + 1, 1);
- RPE = zeros(trials, 1);
- V(1) = Vinit;
- % Call learning rule
- for t = 1:trials
- RPE(t) = rewards(t) - V(t);
- if RPE(t) >= 0
- V(t + 1) = V(t) + alphaPPE*RPE(t);
- else
- V(t + 1) = V(t) + alphaNPE*RPE(t);
- end
- end
- rateParam = exp(slope*V(1:trials) + intercept);
- rateParam(rateParam < 0) = 0.1; % set rate param to zero if it goes below; might consider a better rule in the future
- probSpike = poisspdf(spikeCounts, rateParam(timeLocked)); % mask rateParam to exclude trials where the animal didn't lick fast enough
- mean_predictedSpikes = rateParam(timeLocked);
- V = V(1:trials);
- V = V(timeLocked);
- RPE = RPE(timeLocked);
- if any(isinf(log(probSpike)))
- LH = 1e9;
- else
- LH = -1 * sum(log(probSpike));
- end
|