nappMATLAB-toolbox/import/save_table_as_hdf5.m

function success = save_table_as_hdf5(input_table, filename)

% This function takes input_table (of table type), structured with
% Thorsten's FileTable format, and stores it as .h5 file with name
% filename. Each row is treated as a group, where columns can be attributes
% of the group or datasets inside the group. 

% Last update: 11.08.2023
% By: Juan F. Florez Ospina
% Email: juan.florez-ospina@psi.ch

if istable(input_table)
    
    string_columns = input_table(:,vartype("string")).Properties.VariableNames;
    input_table = convertvars(input_table,string_columns,'cellstr');
    % split multicolumn variables
    input_table = splitvars(input_table);
    column_names = input_table.Properties.VariableNames;
else
    success = 0;
    return
end

% Create array with indices of column_names
column_names_idx = 1 : length(column_names);

% Divide input_table into tables with numeric and cellstr variables/columns
numeric_table = input_table(:,vartype('numeric')); % group attributes of a timestamped measurement
cellstr_table = input_table(:,vartype('cellstr')); % group datasets of a timestamped measurement

% Record columns/vars original index in reference table
numeric_columns = numeric_table.Properties.VariableNames;
numeric_columns_idx = argwhere_var_in_varset(numeric_columns,column_names);

str_columns = cellstr_table.Properties.VariableNames;
str_columns_idx = argwhere_var_in_varset(str_columns,column_names);

% Based on homogeneous columns/cell arrays, we identify the remaining columns
% with possibly heterogeneous entries. Ideally, these should store numeric
% n-dimensional arrays/datasets.

rest_columns = {};
cntr = 1;
for i = 1 : length(input_table.Properties.VariableNames)
    variable = input_table.Properties.VariableNames{i};

    is_str = sum(strcmp(variable,str_columns))>0;
    is_num = sum(strcmp(variable,numeric_columns))>0;
    if ~(is_num || is_str)
        rest_columns{cntr} = variable;
        disp(variable)
        cntr = cntr + 1;
    end
end

rest_columns_idx = argwhere_var_in_varset(rest_columns,column_names);

% Create an hdf5 file, where rows are seen as measurements/groups with
% attributes given by numeric and str variables and datasets given by the 
% remaining variables as long as they are numeric n-dimensional arrays.

cell_table = input_table(:,rest_columns);
[rows, cols] = size(input_table);

for i = 1 : rows
    group_name = strcat("/MEAS_",string(i),"/");
    if ~isempty(rest_columns)
        for j = 1 : length(rest_columns)
            dataset_name = strcat("DS",string(rest_columns_idx(j)));
    
            % Some table entries do not provide direct access to the object
            % of interest, but may contain either {object} or {{object}}. 
            % The following loop solves that problem. TODO: is this generic enough? 
    
            data = cell_table{i,column_names{rest_columns_idx(j)}}{1}; % <=> data = cell_table{i,rest_columns{j}}{1};
    
            while iscell(data)
                data = data{1};
            end
            % when there is no data, we still need to provide a valid size
            if ~isempty(data)
                sz = size(data);
            else
                sz = [1,1];
            end
            %%%%
            % Group's dataset: Create dataset of size sz in a certain group 
            h5create(filename, strcat(group_name,dataset_name),sz);
            % Dataset's attribute: Add attribute with dataset_name equal to column_name
            attr = "column_name";        
            attr_value = column_names{rest_columns_idx(j)}; % <=> rest_columns{j};
            h5writeatt(filename,strcat(group_name,dataset_name),attr,attr_value)
    
            %attr = "column_name_idx";        
            %attr_value = rest_columns_idx(j); % <=> rest_columns{j};
            %h5writeatt(filename,strcat(group_name,dataset_name),attr,attr_value)
    
            % Assign 'data' to the dataset. 'data' has to be of admissible type. 
            try            
                h5write(filename,strcat(group_name,dataset_name),data)
            catch exception
                disp(exception)
                %h5write(filename,strcat(group_name,dataset_name),cell_table{i,rest_columns{j}})
            end
        end
    else
        dataset_name = "DS_EMPTY";
        sz = [1,1];
        h5create(filename, strcat(group_name,dataset_name),sz);
    end

    % Group's attribute: Assign group attributes
    for attr_idx = 1 : length(numeric_columns) 
        attr = column_names{numeric_columns_idx(attr_idx)};% <=> attr = numeric_columns{attr_idx};
        attr_value = numeric_table{i,attr};

        column_idx_tag = ['ATTR',num2str(numeric_columns_idx(attr_idx)),'_'];
        h5writeatt(filename,group_name,[column_idx_tag,attr],attr_value)

        %attr = "column_name_idx";        
        %attr_value = numeric_columns_idx(attr_idx); % <=> rest_columns{j};

        %h5writeatt(filename,group_name,attr,attr_value)
    end

    for attr_idx = 1 :  length(str_columns)
    
        attr = column_names{str_columns_idx(attr_idx)}; % <=> attr = str_columns{attr_idx}
        attr_value = cellstr_table{i,attr};

        column_idx_tag = ['ATTR',num2str(str_columns_idx(attr_idx)),'_'];
        h5writeatt(filename,group_name,[column_idx_tag,attr],attr_value)

        %attr = "column_name_idx";        
        %attr_value = str_columns_idx(attr_idx); % <=> rest_columns{j};

        %h5writeatt(filename,group_name,attr,attr_value)
    end
    %h5create(filename,"/g2/DS2",[20 Inf],"Chunksize",[5 5]);
end

success = 1;

function idx = argwhere_var_in_varset(var,varset)
    ref_idx = 1 : length(varset);
    if iscellstr(var)
        idx = [];
        for i = 1 : length(var)
            logical_idx = strcmp(var{i},varset); 
            idx = [idx,ref_idx(logical_idx)];
        end
    else
        logical_idx = strcmp(var,varset);    
        idx = ref_idx(logical_idx);
    end