15minHeadphones

Solution

context_map=csvread('train_X_final.csv');             % Load train input (context matrix)
sim_yield_die_loss=csvread('train_Y_final.csv');      % Load train output
wafer_count=size(context_map,1);                % Get number of wafers

%Build regular WLSR model +++++++++++++++++++++++++++++++++++++++++++++++++
lambda=0.4;                                 %Decay rate on learning
w_diag=fliplr(lambda.^(1:wafer_count));
W=zeros(wafer_count);
for g=1:wafer_count
    W(g,g)=w_diag(g);
end
beta_WLSR=pinv(context_map'*W*context_map)*context_map'*W*sim_yield_die_loss';

context_map_val=csvread('val_X_final.csv');         %Load train input (context matrix)

%Evaluate models
yp_WLSR=context_map_val*beta_WLSR;                 

%Make sure to cap crazy results outside possible die >600 <0
yp_WLSR(yp_WLSR<0)=0;
yp_WLSR(yp_WLSR>600)=600;

%Now write out prediction
csvwrite('WLSR_prediction_04.csv',yp_WLSR)

%Now calc and write out cost
Y_val=csvread('testing/val_Y.csv');
Y_pred=csvread('WLSR_prediction_04.csv');
err=Y_pred-Y_val;
err_over=sum(abs(err(err>=0)));
err_under=sum(abs(err(err<0)))*10;
overall_score_LSR=(err_under+err_over)/wafer_count;
overall_score_LSR