Simplifying loop in mata

[sungwook kim]

Dear all,

I'm running iteration in Mata until the difference between the sum of previous weight matrix and current weight matrix is close zero. Then, I want to stop the iteration and extract the weight matrix using

st_matrix("weight", weight)

.

However, I don't want to write all the repeated process in my code chunk- I want to simplify this code.

    
clear mata 
    mata
  ind_cat = 0,1,1,0\0,1,1,0\1,0,1,0\0,1,0,1\1,0,0,1
  weight = J(5,3,1)  
  cons = 8,4,6,6\2,8,4,6\7,4,3,8 
  ind_agg = J(3,4,1)  
    A=rows(ind_cat) 

    B=rows(cons)
    C=cols(cons)
  weight = J(A,B,1) 
  w1 = sum(weight)
  ind_agg = J(B,C,1)  
 sum_ind_cat = colsum(ind_cat)
 
mat_ind_cat= J(B,C,1)  
ind_agg0= sum_ind_cat:*mat_ind_cat

    ind_agg1=J(B,C,1) 
 for (j=1;j<=B;j++) {
   for (i=1;i<=C;i++) {
    index= select( 1::rows(ind_cat) ,ind_cat[., i])
    weight[index,j] = weight[index,j] * cons[j, i]/ ind_agg0[j,i]
    
    ind_agg1[j,] = colsum(ind_cat:*weight[,j])


   weight[index,j] = weight[index,j]* cons[j , i] / ind_agg1[j,i]
  w2= sum(weight)
    }
    }
    weight = weight  
 for (j=1;j<=B;j++) {
   for (i=1;i<=C;i++) {
    index= select( 1::rows(ind_cat) ,ind_cat[., i])
    weight[index,j] = weight[index,j] * cons[j, i]/ ind_agg0[j,i]
    
    ind_agg1[j,] = colsum(ind_cat:*weight[,j])
   weight[index,j] = weight[index,j]* cons[j , i] / ind_agg1[j,i]
  w3 = sum(weight)
    }
}
    weight = weight  //this is important
 for (j=1;j<=B;j++) {
   for (i=1;i<=C;i++) {
    index= select( 1::rows(ind_cat) ,ind_cat[., i])
    weight[index,j] = weight[index,j] * cons[j, i]/ ind_agg0[j,i]
    
    ind_agg1[j,] = colsum(ind_cat:*weight[,j])
   weight[index,j] = weight[index,j]* cons[j , i] / ind_agg1[j,i]
  w4 = sum(weight)
    }
        //weight1 = weight

}
    weight = weight  //this is important
 for (j=1;j<=B;j++) {
   for (i=1;i<=C;i++) {
    index= select( 1::rows(ind_cat) ,ind_cat[., i])
    weight[index,j] = weight[index,j] * cons[j, i]/ ind_agg0[j,i]
    
    ind_agg1[j,] = colsum(ind_cat:*weight[,j])
   weight[index,j] = weight[index,j]* cons[j , i] / ind_agg1[j,i]
  w5 = sum(weight)
    }
}
diff1 = w2 -w1
diff2 = w3- w2
diff3 = w4 - w3
diff4 = w5 - w4


diff1
diff2
diff3
diff4

end

.

That is, each diff becomes zero (or close to zero) I would like to stop the looping. For actual data, I did this 40 times, writing the same code 40 times.

Many thanks in advance!

Kind regards,

Kim

[daniel klein]

I am having trouble following both the verbal description and the code. For example, you code

...
  ind_agg = J(3,4,1)  
    A=rows(ind_cat)

    B=rows(cons)
    C=cols(cons)
  weight = J(A,B,1)
  w1 = sum(weight)
  ind_agg = J(B,C,1)

Why do you initialize ind_agg when you replace it a couple of lines later without ever using it before?

Anyway, iteration until convergence is usually coded as

difference = initaial_difference
while (difference > cut_off_criteria) {
    differecnce = ...
}

Best
Daniel

[sungwook kim]

Thanks for your reply. Re your question, replacing ind_agg was done for using actual data. However, I added a simple data here for reference rather than using actual data.
Regards,
Kim