Model evidence calculation in TI #81
Description
Currently, when approximating model evidence by TI, for each TI step one sample is randomly selected and with these samples the quadrature is evaluated. This is repeated ~1000 times and for the final result the mean of all the quadrature results is evaluated, together with its standard deviation. This was done, to be able to report a standard deviation.
However, the standard deviation is too large and also not really reporting what we are interested in. Therefore, either a new approach needs to be found or the standard deviation should not be reported.