whobpyt dev discussion 2024-03-13

[JohnGriffiths]

code base discussion

[JohnGriffiths]

Note re this line

whobpyt/whobpyt/datatypes/parameter.py

Line 16 in ef921c3

val : Tensor

if asRand = True
and prior_mean and prior_std are given
and fit_par is True
then it will sample from the distribution

[JohnGriffiths]

lb argument is lower bound

note :in general all params should be in the docstring

[JohnWangDataAnalyst]

whobpyt/whobpyt/datatypes/parameter.py

Line 52 in e37c249

'''

test

[sbastiaens]

whobpyt/whobpyt/datatypes/parameter.py

Line 118 in e37c249

return self.lb + torch.exp(self.val)

For the parameter lb (lower bound), default is equal to 0 now, but shouldn't it be 'None' as some parameters will be negative?

[JohnGriffiths]

if fit_par is specified then it will fit the hyperparameter, i.e. the mean and variance

[JohnGriffiths]

inputs go in as numpy variables, params converts them at the end to tensors

[P-Oveisi]

whobpyt/whobpyt/datatypes/parameter.py

Line 67 in e37c249

if self.asRand == True:

I think there should be an else clause stating that if asRand is false, then set the random seed as a fixed number. (still sampling from the distribution, but now in every run, the same sample would be chosen)