Homec4science
Diffusion Simon Botrylloides Regeneration Group (master) Browse

SiBoRG/process_results/analyse_identification_keys.m
3aa2120ee410master

Last Change
Raw File

analyse_identification_keys.m
View Options

% ANALYSE_IDENTIFICATION_KEYS loads a characteristics table, use it to classify
% the different species and displays the results in the form of a tree.
%
% Blanchoud Group, UNIFR
% Simon Blanchoud
% 24/06/2020
function analyse_identification_keys
% The file to be loaded
ref_file = '/home/blanchou/Documents/Lionel/car_table_summary.csv';
% Parse the content of the file and extract the various information
data = csv2cell(ref_file);
props = data(2:end, 1); % Name of the characteristics
species = data(1, 3:end); % Name of the species
ids = cell2mat(data(2:end, 2)); % IDs of the characteristics
data = cell2mat(data(2:end, 3:end)); % Groups of the characteristics
ease_indxs = [find(ids>9, 1) find(ids>4, 1)]; % Indexes of the types of char.
% First inspect the data to identify similar species
val = corr(data);
figure;imagesc(val, [-1 1]);
colormap(brewermap(64, 'RdBu'));
colorbar;
% Parameters of the scoring function used in the analysis
% Adapt these values to provide more or less importance to each factor
winfo = 1.0; % Information content, the number of non-0 elements
wrank = 1.3; % Rank of the classification, the number of different groups
wease = 0.5; % How easy is the measure (visual/internal/termporal)
wtrust = -0.5; % Is the value in conflict with other reports?
weights = [winfo wrank wease wtrust]; % A simpler format
% Run the actual classification
[decis, iters] = classify_species(data, @best_discriminant, [weights ease_indxs]);
% And display it
plot_decisions(decis, iters, weights);
return;
end
% This function ranks the characteristics
function score = best_discriminant(data, indxs, weights)
% Not enough data to work
if (size(data, 2) < 2)
score = zeros(size(data));
return;
end
% Get the categories in case of variability
data2 = round(data);
% The categories with varibility issues
trust = sum(data2==5, 2);
% The inferred categories ignoring the variability
tmp = round(mod(data, 1)*10);
data(data2 == 5) = tmp(data2 == 5);
tmp = sort(data,2);
% The number of different categories
rank = sum(diff(tmp, [], 2)>0, 2);
% The information content
info = sum(data>0, 2);
% The ease of measuring the characteristic, based only on their index
ease = ones(size(indxs)) + (indxs<weights(5)) + (indxs<weights(6));
% The final weighted score
score = weights(1)*info + weights(2)*rank + weights(3)*ease + weights(4)*trust;
return;
end
% This function displays a decision tree
function plot_decisions(decis, iters, weights)
% Formatting the input
lin_iters = cat(2, iters{:});
nclass = max(cat(2, lin_iters{:}));
nsteps = length(decis);
sizes = cellfun('numel', iters);
% The total number of elements in case of repeats
ntot = max(cellfun(@(x)sum(cellfun('numel', x)), iters));
% Temporary variables to build the tree
prev_elems = {[1:nclass]};
prev_coords = [ntot/2; nsteps];
prev_cat = [ntot];
% The actual figure
hfig = figure;
haxe = axes('Parent', hfig, 'NextPlot', 'add');
axis(haxe, [-3 ntot+1 -1 nsteps+1], 'off');
% The top of the tree
text(prev_coords(1,1), prev_coords(2, 1)+1.5, num2str(weights), ...
'horizontalalignment', 'center');
plot(haxe, prev_coords(1,[1 1]), prev_coords(2,1)+[0 1]);
% We go through all steps of the classification
for i=1:nsteps
% Display the iteration
text(-2, prev_coords(2, 1), num2str(decis(i)));
plot(haxe, [-1 ntot], prev_coords(2,[1 1]), 'k');
% Prepare the current level
new_cat = cellfun('numel', iters{i});
new_coords = NaN(2, sizes(i));
counts = 0;
min_indx = 1;
unused = ones(1, size(prev_coords,2));
% Go through all the groups
for j=1:sizes(i)
% Check which groups the current element belongs to
hits = cellfun(@(x)all(ismember(iters{i}{j}, x)), prev_elems);
% Get the first hit in case of duplications
indx = find(hits);
min_indx = min(max(indx, min_indx));
unused(min_indx) = 0;
% Plot the links between the groups
new_coords(:, j) = [new_cat(j)/2 + counts; prev_coords(2, min_indx)-1];
plot(haxe, [prev_coords(1, min_indx) new_coords(1, j)], [prev_coords(2, min_indx) new_coords(2, j)]);
counts = counts + new_cat(j);
% Highlight that we identified a category
if (new_cat(j)==1)
scatter(haxe, new_coords(1, j), new_coords(2, j));
text(new_coords(1,j), new_coords(2, j)-0.25, num2str(iters{i}{j}), ...
'horizontalalignment', 'center');
end
end
% Highlight if we have indirectly lost some nodes
unused = prev_coords(:, logical(unused)&(prev_cat~=1));
if (~isempty(unused))
scatter(haxe, unused(1, :), unused(2, :), 'x');
end
% Next iteration
prev_elems = iters{i};
prev_coords = new_coords;
prev_cat = new_cat;
end
return;
end
c4science · Help