apop_model_fix_params.c File Reference

Data Structures

• struct apop_model_fixed_params_settings

Functions

• apop_model * apop_model_fix_params (apop_model *model_in)

---

Detailed Description

Set some of the parameters of a model to fixed values.

---

Function Documentation

apop_model* apop_model_fix_params

(

apop_model *

model_in

)

Produce a model based on another model, but with some of the parameters fixed at a given value.

As well as a pointer to the model whose parameters are to be fixed, I need two sets of data.

For the fixed parameters, I need to know the values at which they'll be fixed. Send me an apop_data set that has the same shape as the parameters of your model; at the positions of the fixed parameters, give the values to which they will be fixed. For the free parameters, I (mostly) don't care what value they have. This set of parameters can be either set as the model_in->parameters element, or as an argument to the model. [If you give me both, I will use the one explicitly sent in rather than the one attached to the input model.]

I also need to know which parameters to fix, which requires a mask that I can hold over the paramete set. Again, the mask is an apop_data set of the same size and shape as your data. Where there is a nonzero marker, I will fix the parameter.

You again have two options for giving me this information. You can use the parameter matrix as the mask: just set parameters to be left free to a nonzero value (including GSL_NAN). Or, you can explicitly send in a mask, with ones at params to be fixed and zero elsewhere. Again, I will try the explicit mask first.

You also need to input the base model, which I will copy (along with its settings groups) to form the new model.

The output is an apop_model that can be estimated, Bayesian updated, et cetera.

• The estimate method always uses an MLE, and it never calls the base model's estimate method.

• If the input model has MLE-style settings attached, I'll use them for the estimate method. Otherwise, I'll set my own.

• If the parameter input has non-NaN values at the free parameters, then I'll use those as the starting point for any search; else I'll start from 1 as usual.

Here is a sample program. It produces a few thousand draws from a multivariate normal distribution, and then tries to recover the means given a var/covar matrix fixed at the correct variance.

#include <apop.h>

int main(){
  gsl_rng *r    = apop_rng_alloc(10);
  size_t    i, ct = 1e4;
  apop_data *d  = apop_data_alloc(0,ct,2);
  double    draws[2];
  apop_data *params = apop_data_alloc(2,2,2);
  apop_data_fill(params, 8,  1, 0.5,
                         2,  0.5, 1);
  apop_model *pvm = apop_model_copy(apop_multivariate_normal);
  pvm->parameters = apop_data_copy(params);
    for(i=0; i< ct; i++){
        apop_draw(draws, r, pvm);
        apop_data_set(d, i, 0, draws[0]);
        apop_data_set(d, i, 1, draws[1]);
    }
    gsl_vector_set_all(pvm->parameters->vector, GSL_NAN);
    apop_model *mep1   = apop_model_fix_params(pvm);
    apop_model *e1  = apop_estimate(d, *mep1);
    printf("original params: ");
    apop_vector_show(params->vector);
    printf("estimated params: ");
    apop_vector_show(e1->parameters->vector);
    return 0;
}

Parameters:

model_in

The base model

Returns:

a model that can be used like any other, with the given params fixed or free.