function[optparams,min_cvError,CVerr_Grid]= do_CV(params,nfolds)

x=params.x;

y=params.y;

MAX_ITR=params.MAX_ITR;

epsilon=params.epsilon;

lambda_list=params.lambda_list;

gamma_list=params.gamma_list;

Bind=params.Bind;

CVerr_Grid=zeros(length(gamma_list),length(lambda_list));

Num_Tasks=size(y,2);

Fold_indices = Generate_CV_Folds(x,nfolds)>0;

cX_train = arrayfun(@(idx_fold) x(Fold_indices(:,idx_fold),:),1:nfolds, 'Unif', 0);

cY_train = arrayfun(@(idx_fold) y(Fold_indices(:,idx_fold),:),1:nfolds, 'Unif', 0);

cX_test = arrayfun(@(idx_fold) x(~Fold_indices(:,idx_fold),:),1:nfolds, 'Unif', 0);

cY_test = arrayfun(@(idx_fold) y(~Fold_indices(:,idx_fold),:),1:nfolds, 'Unif', 0);

parfor k=1:length(lambda_list)

lambda=lambda_list(k);

mse = zeros(nfolds,Num_Tasks);

CV_err = zeros(1,length(gamma_list)) ;

for p=1:length(gamma_list)

gamma = gamma_list(p);

% Train model over N folds

for idx_fold = 1:size(Fold_indices,2)

% Creat fold data andd

X_train = cX_train{idx_fold};

y_train = cY_train{idx_fold};

X_test = cX_test{idx_fold};

y_test = cY_test{idx_fold};

coff = proxgrad(X_train,y_train,MAX_ITR, lambda,gamma,Bind,epsilon);

% Compute prediction error for each task

mse(idx_fold,:) = mean((X_test * coff- y_test).^2);

end

%In_Fold_Error{1,p} = tmp;

CV_err(p) = mean(mse(:));

end

% CVErr_twoparam{1,k} = In_Fold_Error;

CVerr_Grid(:,k)=CV_err;

[opt_lambda, opt_gamma,min_cvError]= modelSelection(CVerr_Grid,lambda_list,gamma_list);

optparams.opt_lambda = opt_lambda;

optparams.opt_gamma = opt_gamma;