PsiMarginal.py 文件源码

def minEntropyStim(self):
        """Find the stimulus intensity based on the expected information gain.

        Minimum Shannon entropy is used as selection criterion for the stimulus intensity in the upcoming trial.
        """
        self.pdf = self.pdf
        self.nX = len(self.stimRange)
        self.nDims = np.ndim(self.pdf)

        # make pdf the same dims as conditional prob table likelihood
        self.pdfND = np.expand_dims(self.pdf, axis=self.nDims)  # append new axis
        self.pdfND = np.tile(self.pdfND, (self.nX))  # tile along new axis

        # Probabilities of response r (succes, failure) after presenting a stimulus
        # with stimulus intensity x at the next trial, multiplied with the prior (pdfND)
        self.pTplus1success = np.multiply(self.likelihood, self.pdfND)
        self.pTplus1failure = self.pdfND - self.pTplus1success

        # Probability of success or failure given stimulus intensity x, p(r|x)
        self.sumAxes = tuple(range(self.nDims))  # sum over all axes except the stimulus intensity axis
        self.pSuccessGivenx = np.sum(self.pTplus1success, axis=self.sumAxes)
        self.pFailureGivenx = np.sum(self.pTplus1failure, axis=self.sumAxes)

        # Posterior probability of parameter values given stimulus intensity x and response r
        # p(alpha, sigma | x, r)
        self.posteriorTplus1success = self.pTplus1success / self.pSuccessGivenx
        self.posteriorTplus1failure = self.pTplus1failure / self.pFailureGivenx

        # Expected entropy for the next trial at intensity x, producing response r
        self.entropySuccess = self.__entropy(self.posteriorTplus1success)
        self.entropyFailure = self.__entropy(self.posteriorTplus1failure)
        self.expectEntropy = np.multiply(self.entropySuccess, self.pSuccessGivenx) + np.multiply(self.entropyFailure,
                                                                                                 self.pFailureGivenx)
        self.minEntropyInd = np.argmin(self.expectEntropy)  # index of smallest expected entropy
        self.xCurrent = self.stimRange[self.minEntropyInd]  # stim intensity at minimum expected entropy

        self.iTrial += 1
        if self.iTrial == (self.nTrials - 1):
            self.stop = 1