How to set the initial value in stanmodel in Matlab

Modeling
1

Hi guys,

I have a hard time setting initial values for parameters in Matlab stanmodel. I defined a structure for the initial value of parameters and brought the defined structure in stanmodel using init. But I have an error. Any suggestion? Thank you in advance.

2

Could you show the code you try to use to set the initial values?

Aki

3

There are some examples on the wiki that may be helpful:

4

Hi,

This is my code:

evol_tree_code = {
‘data {’
’ int<lower=0> T; ‘
’ int<lower=0> N; ‘
’ real<lower=0,upper=1> Fhat[N,T]; ‘
’ matrix<lower=0,upper=1> [N,N] UTree; ‘
’}’
‘parameters {’
’ real<lower= 0,upper=0.3> mu[N]; ‘
’ vector<lower=0> [N] delta_tau; ‘
’}’
‘transformed parameters {’
’ matrix <lower=0> [N,T] numM; ‘
’ matrix <lower=0,upper=1> [N,T] M; ‘
’ matrix <lower=0> [N,T] F; ‘
’ vector<lower=0> [N] tau;’
’ real<lower=0> den[T];’
’ tau = UTree’‘delta_tau; ‘
‘for(t in 1:T){’
’ den[t] = 0.001;//this should be a small value’
’ for (n in 1:N){’
’ if (tau[n] > t - 1){’
’ numM[n,t] = 0.001;’
’ }else{’
’ numM[n,t] = exp(mu[n](t-1 - tau[n]));’
’ }’
’ den[t] = den[t] + numM[n,t];’
’ }’
’ for (n in 1:N){’
’ M[n,t] = numM[n,t]/ den[t];’
’ }’
’}’
’ F = UTree*M; ‘
’}’
’ model {’
’ for(t in 1:T) {’
’ for (n in 1:N){’
’ mu[n] ~ uniform(0, 0.3);’
’ delta_tau[n] ~ exponential(2);’
’ Fhat[n,t] ~ normal(F[n,t],0.01);’
’ }’
’ }’
’}’

};
%%
Fdata =
[1.0000,1.0000,1.0000,1.0000,1.0000,1.0000,1.0000,1.0000,1.0000,1.0000,1.0000
; 0, 0, 0, 0, 0, 0, 0, 0, 0,1.0000, 0
; 0,
0,0.0500,0.2000,0.2600,0.1000,0.2600,0.6000,0.6600,0.8800,0.9700
; 0,0.2800, 0, 0,0.7100,0.9100,0.7600,0.4600,0.4000,0.1300,0.0400
; 0, 0, 0, 0, 0,0.2500,0.3900,0.2200,0.4000,0.1400,0.0300
; 0, 0, 0, 0, 0, 0,0.0800,0.1200,0.1200,0.3400,0.400];

%%
Treedata = [ 0 0 0 0 0 0
;0 0 0 1 1 1
;0 0 0 0 0 0
;1 0 0 0 0 0
;0 0 0 0 0 0
;0 0 1 0 0 0];

UTreedata = [ 1 0 0 0 0 0
;1 1 1 1 1 1
; 0 0 1 0 0 0
; 1 0 0 1 0 0
; 0 0 0 0 1 0
; 0 0 1 0 0 1];

%%
T = size(Fdata,2);
N = size(Fdata,1);

%% initial values

delta_tau_init = ones(N,1);
mu_init = zeros(N,1);

evol_tree_dat = struct(‘T’,T,‘N’,N,‘Fhat’,Fdata,‘UTree’,UTreedata);

evol_tree_initial_values = struct(‘delta_tau’,delta_tau_init,‘mu’,mu_init);

%%
model =
StanModel(‘init’,evol_tree_initial_values,‘file_overwrite’,true,‘verbose’,true,‘method’,‘sample’,‘model_code’,evol_tree_code,‘data’,evol_tree_dat);
%%

model.compile();

fit_vb = model.vb();

print(fit_vb);

Thank you.

5

Switch this to
fit = stan('init',evol_tree_initial_values,'file_overwrite',true,'verbose',true,'method','sample','model_code',evol_tree_code,'data',evol_tree_dat);
and it works (at least no errors). This also uses the MCMC by default and solves your another problem.
It would be better to put the model code in a file as then the compiled model is cached and you don’t need to recompile when refitting with a new data.

Aki

6

Hi Avi,

I tried it but I still have this error: Stan can’t start sampling from this
initial value and rejection of the initial value, Log probability evaluates
to log(0), i.e. negative infinity. Thank you for your help.

7

I don’t have the error when I remove the initialization part but when I add
it, I have this error.

8

This means the initial conditions you are providing are impossible somehow. Usually a constraint is being violated, or a variable is being passed to a function that violates the constraints of that function (like passing a standard deviation less than zero to a normal likelihood).

But! If you can run the model without the initialization, do that. Usually Stan can generate initial conditions automatically, and it’s better to just work with that (cause then you don’t have to worry so much about your initial conditions biasing your results in some weird way).

edit: spelling

9

Thank you.