ff_wkz_vf_vecsv.m

%% Risky + Safe Asset Dyna Prog Two-Step Solution (Optimized-Vectorized)
% *back to <https://fanwangecon.github.io Fan>'s
% <https://fanwangecon.github.io/CodeDynaAsset/ Dynamic Assets Repository> 
% Table of Content.*

%%
function result_map = ff_wkz_vf_vecsv(varargin)
%% FF_WKZ_VF_VECSV solve infinite horizon exo shock + endo asset problem
% This program solves the infinite horizon dynamic savings and risky
% capital asset problem with some ar1 shock. This is the efficient vectorized version
% of
% <https://fanwangecon.github.io/CodeDynaAsset/m_akz/solve/html/ff_wkz_vf.html
% ff_wkz_vf>. See that file for more descriptions. 
%
% @param param_map container parameter container
%
% @param support_map container support container
%
% @param armt_map container container with states, choices and shocks
% grids that are inputs for grid based solution algorithm
%
% @param func_map container container with function handles for
% consumption cash-on-hand etc.
%
% @return result_map container contains policy function matrix, value
% function matrix, iteration results, and policy function, value function
% and iteration results tables. 
%
% keys included in result_map:
%
% * mt_val matrix states_n by shock_n matrix of converged value function grid
% * mt_pol_a matrix states_n by shock_n matrix of converged policy function grid
% * ar_val_diff_norm array if bl_post = true it_iter_last by 1 val function
% difference between iteration
% * ar_pol_diff_norm array if bl_post = true it_iter_last by 1 policy
% function difference between iterations
% * mt_pol_perc_change matrix if bl_post = true it_iter_last by shock_n the
% proportion of grid points at which policy function changed between
% current and last iteration for each element of shock
%
% @include
%
% * <https://fanwangecon.github.io/CodeDynaAsset/m_akz/solve/html/ff_wkz_evf.html ff_wkz_evf>
% * <https://fanwangecon.github.io/CodeDynaAsset/m_akz/paramfunc/html/ffs_akz_set_default_param.html ffs_akz_set_default_param>
% * <https://fanwangecon.github.io/CodeDynaAsset/m_akz/paramfunc/html/ffs_akz_get_funcgrid.html ffs_akz_get_funcgrid>
% * <https://fanwangecon.github.io/CodeDynaAsset/m_akz/solvepost/html/ff_akz_vf_post.html ff_akz_vf_post>
%
% @seealso
%
% * concurrent (safe + risky) loop: <https://fanwangecon.github.io/CodeDynaAsset/m_akz/solve/html/ff_akz_vf.html ff_akz_vf>
% * concurrent (safe + risky) vectorized: <https://fanwangecon.github.io/CodeDynaAsset/m_akz/solve/html/ff_akz_vf_vec.html ff_akz_vf_vec>
% * concurrent (safe + risky) optimized-vectorized: <https://fanwangecon.github.io/CodeDynaAsset/m_akz/solve/html/ff_akz_vf_vecsv.html ff_akz_vf_vecsv>
% * two-stage (safe + risky) loop: <https://fanwangecon.github.io/CodeDynaAsset/m_akz/solve/html/ff_wkz_vf.html ff_wkz_vf>
% * two-stage (safe + risky) vectorized: <https://fanwangecon.github.io/CodeDynaAsset/m_akz/solve/html/ff_wkz_vf_vec.html ff_wkz_vf_vec>
% * two-stage (safe + risky) optimized-vectorized: <https://fanwangecon.github.io/CodeDynaAsset/m_akz/solve/html/ff_wkz_vf_vecsv.html ff_wkz_vf_vecsv>
% * two-stage + interpolate (safe + risky) loop: <https://fanwangecon.github.io/CodeDynaAsset/m_akz/solve/html/ff_iwkz_vf.html ff_iwkz_vf>
% * two-stage + interpolate (safe + risky) vectorized: <https://fanwangecon.github.io/CodeDynaAsset/m_akz/solve/html/ff_iwkz_vf_vec.html ff_iwkz_vf_vec>
% * two-stage + interpolate (safe + risky) optimized-vectorized: <https://fanwangecon.github.io/CodeDynaAsset/m_akz/solve/html/ff_iwkz_vf_vecsv.html ff_iwkz_vf_vecsv>
%

%% Default
% * it_param_set = 1: quick test
% * it_param_set = 2: benchmark run
% * it_param_set = 3: benchmark profile
% * it_param_set = 4: press publish button

it_param_set = 4;
bl_input_override = true;
[param_map, support_map] = ffs_akz_set_default_param(it_param_set);

% Note: param_map and support_map can be adjusted here or outside to override defaults
% param_map('it_w_n') = 50;
% param_map('it_z_n') = 15;

% get armt and func map
[armt_map, func_map] = ffs_akz_get_funcgrid(param_map, support_map, bl_input_override); % 1 for override
default_params = {param_map support_map armt_map func_map};

%% Parse Parameters 1

% if varargin only has param_map and support_map,
params_len = length(varargin);
[default_params{1:params_len}] = varargin{:};
param_map = [param_map; default_params{1}];
support_map = [support_map; default_params{2}];
if params_len >= 1 && params_len <= 2
    % If override param_map, re-generate armt and func if they are not
    % provided
    bl_input_override = true;
    [armt_map, func_map] = ffs_akz_get_funcgrid(param_map, support_map, bl_input_override);
else
    % Override all
    armt_map = [armt_map; default_params{3}];
    func_map = [func_map; default_params{4}];
end

% append function name
st_func_name = 'ff_wkz_vf_vecsv';
support_map('st_profile_name_main') = [st_func_name support_map('st_profile_name_main')];
support_map('st_mat_name_main') = [st_func_name support_map('st_mat_name_main')];
support_map('st_img_name_main') = [st_func_name support_map('st_img_name_main')];

%% Parse Parameters 2

% armt_map
params_group = values(armt_map, {'ar_w', 'ar_z'});
[ar_w, ar_z] = params_group{:};
params_group = values(armt_map, {'ar_a_meshk', 'ar_k_mesha', 'mt_coh_wkb', 'it_ameshk_n'});
[ar_a_meshk, ar_k_mesha, mt_coh_wkb, it_ameshk_n] = params_group{:};
% func_map
params_group = values(func_map, {'f_util_log', 'f_util_crra', 'f_cons', 'f_coh'});
[f_util_log, f_util_crra, f_cons, f_coh] = params_group{:};
% param_map
params_group = values(param_map, {'fl_r_save', 'fl_r_borr', 'fl_w',...
    'it_z_n', 'fl_crra', 'fl_beta', 'fl_c_min'});
[fl_r_save, fl_r_borr, fl_wage, it_z_n, fl_crra, fl_beta, fl_c_min] = params_group{:};
params_group = values(param_map, {'it_maxiter_val', 'fl_tol_val', 'fl_tol_pol', 'it_tol_pol_nochange'});
[it_maxiter_val, fl_tol_val, fl_tol_pol, it_tol_pol_nochange] = params_group{:};
% support_map
params_group = values(support_map, {'bl_profile', 'st_profile_path', ...
    'st_profile_prefix', 'st_profile_name_main', 'st_profile_suffix',...
    'bl_time', 'bl_display', 'it_display_every', 'bl_post'});
[bl_profile, st_profile_path, ...
    st_profile_prefix, st_profile_name_main, st_profile_suffix, ...
    bl_time, bl_display, it_display_every, bl_post] = params_group{:};

%% Initialize Output Matrixes

mt_val_cur = zeros(length(ar_a_meshk),length(ar_z));
mt_val = mt_val_cur - 1;
mt_pol_a = zeros(length(ar_a_meshk),length(ar_z));
mt_pol_a_cur = mt_pol_a - 1;
mt_pol_k = zeros(length(ar_a_meshk),length(ar_z));
mt_pol_k_cur = mt_pol_k - 1;
mt_pol_idx = zeros(length(ar_a_meshk),length(ar_z));
mt_ev_condi_z_max_kp = zeros(length(ar_w),length(ar_z));
mt_ev_condi_z_max_kp_cur = mt_ev_condi_z_max_kp - 1;

% We did not need these in ff_oz_vf or ff_oz_vf_vec
% see
% <https://fanwangecon.github.io/M4Econ/support/speed/partupdate/fs_u_c_partrepeat_main.html
% fs_u_c_partrepeat_main> for why store using cells.
cl_u_c_store = cell([it_z_n, 1]);
cl_c_valid_idx = cell([it_z_n, 1]);

%% Initialize Convergence Conditions

bl_vfi_continue = true;
it_iter = 0;
ar_val_diff_norm = zeros([it_maxiter_val, 1]);
ar_pol_diff_norm = zeros([it_maxiter_val, 1]);
mt_pol_perc_change = zeros([it_maxiter_val, it_z_n]);

%% Iterate Value Function
% Loop solution with 4 nested loops
%
% # loop 1: over exogenous states
% # loop 2: over endogenous states
% # loop 3: over choices
% # loop 4: add future utility, integration--loop over future shocks
%

% Start Profile
if (bl_profile)
    close all;
    profile off;
    profile on;
end

% Start Timer
if (bl_time)
    tic;
end

% Value Function Iteration
while bl_vfi_continue
    it_iter = it_iter + 1;
    
    %% Solve Second Stage Problem k*(w,z)
    % This is the key difference between this function and
    % <https://fanwangecon.github.io/CodeDynaAsset/m_akz/paramfunc/html/ffs_akz_set_functions.html
    % ffs_akz_set_functions> which solves the two stages jointly
    % Across iterations, only some k*(w,z) is changing, we only update
    % consumption matrix when k*(w,z) is changing
    % note in ff_wkz_evf a = b
    
    bl_input_override = true;
    [mt_ev_condi_z_max, ~, mt_ev_condi_z_max_kp, mt_ev_condi_z_max_bp] = ...
        ff_wkz_evf(mt_val_cur, param_map, support_map, armt_map, bl_input_override);
    
    %% Find which k choice differ across iterations?
    
    mt_w_kstar_diff_idx = (mt_ev_condi_z_max_kp_cur ~= mt_ev_condi_z_max_kp);
    
    %% Solve First Stage Problem w*(z) given k*(w,z)
        
    % loop 1: over exogenous states
    for it_z_i = 1:length(ar_z)

        % State Array fixed
        ar_coh_z = mt_coh_wkb(:,it_z_i);
        
        % Get 2nd Stage Choice Arrays
        % Update rows where opti k given w=k'+b' is changing
        ar_w_kstar_diff_idx = mt_w_kstar_diff_idx(:, it_z_i);
        ar_w_kstar_z = mt_ev_condi_z_max_kp(ar_w_kstar_diff_idx, it_z_i);
        ar_w_astar_z = mt_ev_condi_z_max_bp(ar_w_kstar_diff_idx, it_z_i);        
        
        % Consumption Update
        % Note that compared to
        % <https://fanwangecon.github.io/CodeDynaAsset/m_akz/paramfunc/html/ffs_akz_set_functions.html
        % ffs_akz_set_functions> the mt_c here is much smaller the same
        % number of columns (states) as in the ffs_akz_set_functions file,
        % but the number of rows equal to ar_w length.            
        mt_c = f_cons(ar_coh_z', ar_w_astar_z, ar_w_kstar_z);
                
        % EVAL current utility: N by N, f_util defined earlier
        if (fl_crra == 1)
            mt_utility_update = f_util_log(mt_c);
            fl_u_neg_c = f_util_log(fl_c_min);            
        else
            mt_utility_update = f_util_crra(mt_c);
            fl_u_neg_c = f_util_crra(fl_c_min);            
        end
        
        % Eliminate Complex Numbers
        mt_it_c_valid_idx = (mt_c <= fl_c_min);
        mt_utility_update(mt_it_c_valid_idx) = fl_u_neg_c;        
        
        % Update Storage
        if (it_iter == 1)
            cl_u_c_store{it_z_i} = mt_utility_update;
            cl_c_valid_idx{it_z_i} = mt_it_c_valid_idx;
        else
            cl_u_c_store{it_z_i}(ar_w_kstar_diff_idx,:) = mt_utility_update;                
            cl_c_valid_idx{it_z_i}(ar_w_kstar_diff_idx,:) = mt_it_c_valid_idx; 
        end
                
        % EVAL add on future utility, N by N + N by 1
        ar_evzp_ak_condi_z = mt_ev_condi_z_max(:, it_z_i);
        mt_utility = cl_u_c_store{it_z_i} + fl_beta*ar_evzp_ak_condi_z;
        
        % Index update
        % using the method below is much faster than index replace
        % see <https://fanwangecon.github.io/M4Econ/support/speed/index/fs_subscript.html fs_subscript>
        mt_it_c_valid_idx = cl_c_valid_idx{it_z_i};        
        mt_utility = mt_utility.*(~mt_it_c_valid_idx) + fl_u_neg_c*(mt_it_c_valid_idx);
        
        % Optimization: remember matlab is column major, rows must be
        % choices, columns must be states
        % <https://en.wikipedia.org/wiki/Row-_and_column-major_order COLUMN-MAJOR>
        [ar_opti_val1_z, ar_opti_idx_z] = max(mt_utility);
        mt_val(:,it_z_i) = ar_opti_val1_z;
        mt_pol_a(:,it_z_i) = mt_ev_condi_z_max_bp(ar_opti_idx_z, it_z_i);
        mt_pol_k(:,it_z_i) = mt_ev_condi_z_max_kp(ar_opti_idx_z, it_z_i);        
        
        %% Obtain Choice Index for Vectorized/Analytical Distribution Programs
        % For deriving distributions using vectorized and semi-analytical
        % methods, at convergence, what index do optimal choices correspond
        % to in terms of the rows of mt_val and mt_pol_a, and mt_pol_k. For
        % the LHS matrixes here, each column a different optimal choice,
        % each row a different element of the a_meshk and k_mesha vectors.
        % For each column (one optimal choice), which row has that optimal
        % choice's k' and b' values. Note the code is shorter than code in
        % <https://fanwangecon.github.io/CodeDynaAsset/m_akz/solve/html/ff_wkz_vf_vec.html
        % ff_wkz_vf_vec>.
        
        if (it_iter == (it_maxiter_val + 1))           
            [~, ar_opti_fullakvec_idx_z] = max((ar_a_meshk == mt_pol_a(:,it_z_i)').*(ar_k_mesha == mt_pol_k(:,it_z_i)'));
            mt_pol_idx(:,it_z_i) = ar_opti_fullakvec_idx_z;
        end

    end
    
    %% Check Tolerance and Continuation
    
    % Difference across iterations
    ar_val_diff_norm(it_iter) = norm(mt_val - mt_val_cur);
    ar_pol_diff_norm(it_iter) = norm(mt_pol_a - mt_pol_a_cur) + norm(mt_pol_k - mt_pol_k_cur);
    ar_pol_a_perc_change = sum((mt_pol_a ~= mt_pol_a_cur))/(it_ameshk_n);
    ar_pol_k_perc_change = sum((mt_pol_k ~= mt_pol_k_cur))/(it_ameshk_n);    
    mt_pol_perc_change(it_iter, :) = mean([ar_pol_a_perc_change;ar_pol_k_perc_change]);
    
    % Update
    mt_val_cur = mt_val;
    mt_pol_a_cur = mt_pol_a;
    mt_pol_k_cur = mt_pol_k;
    mt_ev_condi_z_max_kp_cur = mt_ev_condi_z_max_kp;
    
    % Print Iteration Results
    if (bl_display && (rem(it_iter, it_display_every)==0))
        fprintf('VAL it_iter:%d, fl_diff:%d, fl_diff_pol:%d\n', ...
            it_iter, ar_val_diff_norm(it_iter), ar_pol_diff_norm(it_iter));
        tb_valpol_iter = array2table([mean(mt_val_cur,1);...
                                      mean(mt_pol_a_cur,1); ...
                                      mean(mt_pol_k_cur,1); ...
                                      mt_val_cur(it_ameshk_n,:); ...
                                      mt_pol_a_cur(it_ameshk_n,:); ...
                                      mt_pol_k_cur(it_ameshk_n,:)]);
        tb_valpol_iter.Properties.VariableNames = strcat('z', string((1:size(mt_val_cur,2))));
        tb_valpol_iter.Properties.RowNames = {'mval', 'map', 'mak', 'Hval', 'Hap', 'Hak'};
        disp('mval = mean(mt_val_cur,1), average value over a')
        disp('map  = mean(mt_pol_a_cur,1), average choice over a')
        disp('mkp  = mean(mt_pol_k_cur,1), average choice over k')
        disp('Hval = mt_val_cur(it_ameshk_n,:), highest a state val')
        disp('Hap = mt_pol_a_cur(it_ameshk_n,:), highest a state choice')
        disp('mak = mt_pol_k_cur(it_ameshk_n,:), highest k state choice')                
        disp(tb_valpol_iter);
    end
    
    % Continuation Conditions:
    % 1. if value function convergence criteria reached
    % 2. if policy function variation over iterations is less than
    % threshold
    if (it_iter == (it_maxiter_val + 1))
        bl_vfi_continue = false;
    elseif ((it_iter == it_maxiter_val) || ...
            (ar_val_diff_norm(it_iter) < fl_tol_val) || ...
            (sum(ar_pol_diff_norm(max(1, it_iter-it_tol_pol_nochange):it_iter)) < fl_tol_pol))
        % Fix to max, run again to save results if needed
        it_iter_last = it_iter;
        it_iter = it_maxiter_val;        
    end
    
end

% End Timer
if (bl_time)
    toc;
end

% End Profile
if (bl_profile)
    profile off
    profile viewer
    st_file_name = [st_profile_prefix st_profile_name_main st_profile_suffix];
    profsave(profile('info'), strcat(st_profile_path, st_file_name));
end

%% Process Optimal Choices

result_map = containers.Map('KeyType','char', 'ValueType','any');
result_map('mt_val') = mt_val;
result_map('mt_pol_idx') = mt_pol_idx;

result_map('cl_mt_pol_coh') = {mt_coh_wkb, zeros(1)};
result_map('cl_mt_pol_a') = {mt_pol_a, zeros(1)};
result_map('cl_mt_pol_k') = {mt_pol_k, zeros(1)};
result_map('cl_mt_pol_c') = {f_cons(mt_coh_wkb, mt_pol_a, mt_pol_k), zeros(1)};
result_map('ar_st_pol_names') = ["cl_mt_pol_coh", "cl_mt_pol_a", "cl_mt_pol_k", "cl_mt_pol_c"];

if (bl_post)
    bl_input_override = true;
    result_map('ar_val_diff_norm') = ar_val_diff_norm(1:it_iter_last);
    result_map('ar_pol_diff_norm') = ar_pol_diff_norm(1:it_iter_last);
    result_map('mt_pol_perc_change') = mt_pol_perc_change(1:it_iter_last, :);
    result_map = ff_akz_vf_post(param_map, support_map, armt_map, func_map, result_map, bl_input_override);
end

end