def _forwardImplementation(self, inbuf, outbuf):
"""Calculate layer outputs (Gaussian parameters etc., not function
values!) from given activations """
K = self.nGaussians
# Mixing parameters and stddevs
outbuf[0:K * 2] = safeExp(inbuf[0:K * 2])
outbuf[0:K] /= sum(outbuf[0:K])
# Means
outbuf[K * 2:] = inbuf[K * 2:]
def _forwardImplementation(self, inbuf, outbuf):
"""Calculate layer outputs (Gaussian parameters etc., not function
values!) from given activations """
K = self.nGaussians
# Mixing parameters and stddevs
outbuf[0:K*2] = safeExp(inbuf[0:K*2])
outbuf[0:K] /= sum(outbuf[0:K])
# Means
outbuf[K*2:] = inbuf[K*2:]
def _forwardImplementation(self, inbuf, outbuf):
outbuf[:] = safeExp(inbuf)
outbuf.shape = scipy.size(outbuf) / self.slicelength, self.slicelength
s = outbuf.sum(axis=1)
outbuf = (outbuf.T / s).T.flatten()
def _forwardImplementation(self, inbuf, outbuf):
outbuf[:] = safeExp(inbuf)
outbuf /= sum(outbuf)
def transformAlphas(self, base):
#print 'base', base
return safeExp(base) / sum(safeExp(base))
def _forwardImplementation(self, inbuf, outbuf):
outbuf[:] = safeExp(inbuf)
outbuf.shape = scipy.size(outbuf) / self.slicelength, self.slicelength
s = outbuf.sum(axis=1)
outbuf = (outbuf.T / s).T.flatten()
def _forwardImplementation(self, inbuf, outbuf):
outbuf[:] = safeExp(inbuf)
outbuf /= sum(outbuf)
def transformAlphas(self, base):
#print 'base', base
return safeExp(base)/sum(safeExp(base))
