s2_fit_all_cluster.m

%% Set up the dataset and the models we are going to test
% For running on HPC, execute
%       sbatch  hpc_solve_models.sh

%% Parse the hyperparameters
if ~exist('doCross',      'var'), doCross = false;       end
if ~exist('target',       'var'), target  = 'target';    end % 'target' or 'All';
if ~exist('optimizer',    'var'), optimizer = 'classic'; end % 'classic' or 'reparam';
if ~exist('start_idx',    'var'), start_idx = 1;         end % cache 
if ~exist('choose_model', 'var'), choose_model = 'all';  end  

switch doCross
    case false
        cross_valid  = 'one';           % 'one': not cross validate; 'cross_valid': cross validate
        data_folder  = 'noCross';       % save in which folder. value space: 'noCross', 'Cross'
    case true
        cross_valid  = 'cross_valid';   % choose what kind of cross , value space: 'one', 'cross_valid'. 'one' is no cross validation.
        data_folder  = 'Cross';         % save in which folder. value space: 'noCross', .....
end

fittime              = 40;        % how many initialization. value space: Integer
verbose              = 'off';     % show the fit details? 
%% generate save address and choose data 

% save address 
save_address = fullfile(stdnormRootPath, 'Data', data_folder, target,  optimizer);
if ~exist(save_address, 'dir'), mkdir(save_address); end

% create jobs 
T = chooseData(choose_model, optimizer, fittime);

%% Start fit

% assign job 
hpc_job_number = str2double(getenv('SLURM_ARRAY_TASK_ID'));
if isnan(hpc_job_number), hpc_job_number = 2; end
dataset   = T.dataset(hpc_job_number);
roi       = T.roiNum(hpc_job_number);
model_idx = T.modelNum(hpc_job_number);
model     = T.modelLoader{hpc_job_number};

% set the save info, this helps in
% continuing job in the broken pipe 
save_temp = fullfile(save_address, 'temp');
if ~exist(save_temp, 'dir'), mkdir(save_temp); end
save_info.dir       = save_temp;
save_info.roi       = roi;
save_info.model_idx = model_idx;
save_info.dataset   = dataset;
save_info.start_idx = start_idx;

% display information to keep track of fitting 
display = [ 'dataset: ' num2str(dataset), ' roi: ',num2str(roi), ' model: ', num2str(model_idx) ];
disp(display)

% load training label
BOLD_target = dataloader(stdnormRootPath, 'BOLD_target', target, dataset, roi);

% load the input E 
switch model.model_type
    case 'orientation'; which_obj = 'E_ori'; % CE, NOA 
    case 'space'      ; which_obj = 'E_xy'; % SOC, OTS
end

% load contrast energy 
E = dataloader(stdnormRootPath, which_obj, target, dataset, roi);
x = {E};

disp(model.legend)
% get pre-calc normalized energy
switch model.legend
    case 'OTS'
        Z = dataloader(stdnormRootPath, 'Z1', target, dataset, roi);
        x{end + 1} = Z;
    case 'DN'
        Z = dataloader(stdnormRootPath, 'Z2', target, dataset, roi);  
        x{end + 1} = Z;
end

% fit the data without cross validation: knock-1-out
[BOLD_pred, params, Rsquare, model] = ...
model.fit(model, x, BOLD_target, verbose, cross_valid, save_info);

if strcmp(cross_valid, 'one')
    loss_log = model.loss_log;
end

% save data
save(fullfile(save_address , sprintf('parameters_data-%d_roi-%d_model-%d.mat',dataset, roi, model_idx)), 'params');
save(fullfile(save_address , sprintf('prediction_data-%d_roi-%d_model-%d.mat',dataset, roi, model_idx)), 'BOLD_pred');
save(fullfile(save_address , sprintf('Rsquare_data-%d_roi-%d_model-%d.mat',dataset, roi, model_idx))   , 'Rsquare');
if strcmp(cross_valid, 'one')
    save(fullfile(save_address , sprintf('loss_log_data-%d_roi-%d_model-%d.mat',dataset, roi, model_idx)) , 'loss_log');
end