main_DGSA_Reservoir_Sensitivity.m

%% Distance based global sensitivity analysis - Application to reservoir responses
% Part1: Compute main/conditional effects of Prior models

% Author : Jihoon Park (jhpark3@stanford.edu)
% Date : 17 May 2016

clear all; close all; fclose('all'); rng('default');
%% 0. Add directories for path
addpath(genpath(pwd))

%% 1. Specify inputs for DGSA. 

%% 2. Load & process input data
load('PriorResponses'); % load responses
load('DistanceMatrix'); % load distance matrix. Distance matrix can be computed from pdist() function in Matlab. 
                        % For details,refer to main_DGSA_analytic.m and the manual included.
PriorResponses=WWPTbyWell; 

%DGSA=ProcessInputParameters('PriorParameters.dat'); % Process Input parameters
DGSA=ProcessInputParameters('PriorParameters_Spatial.dat'); % This data includes two more parameters - permeability and porosity.
DGSA.D=D;
clear WWPTbyWell;clear D;

%% Loading my own DGSA Matrix
clear all; close all; fclose('all'); rng('default');
addpath(genpath(pwd))

load('C:\Users\Markus Zechner\Documents\GitHub\QUSS\data\evidential\DistanceM.mat','DGSA');



%% 4. Compute & display main effects

% 4.1 Inputs for clustering and display options.
DGSA.Nbcluster=3; % # of clusters
DGSA.MainEffects.Display.ParetoPlotbyCluster=1; % if true, main effects over cluster will be displayed with Pareto plot.
DGSA.MainEffects.Display.StandardizedSensitivity='CI'; 

% if 'CI', the confidence interval will be overlapped on the Pareto plot
% if 'Pareto', Pareto plots will be displayed only
% if 'None' No plot will be generated for standardd main effects
% (default)

% 4.2 Compute main effects from DGSA.
SigLevel_CI=[.95,.9,1]; % This is needed only when you want to display confidence intervals
                        % If you do not need to display confidence intervals,set DGSA.MainEffects.Display.StandardizedSensitivity='Preto' or 'None'     

% 4.3 Perform clustering
DGSA.Clustering=kmedoids(DGSA.D,DGSA.Nbcluster,10); % In this example, K medoid clustering is applied.

% 4.4 Compute Main Effects
DGSA=ComputeMainEffects(DGSA,SigLevel_CI); % If you use Pareto plot or do not want to display main effects, remove SigLevel_CI, otherwise it shows an error.

%% Display cdfs

cdf_MainFactor(DGSA.ParametersValues, DGSA.Clustering, DGSA.ParametersNames,{'owc','oilvis'}); % In this example, CDFs for 'owc' and 'oilvis' will be displayed.
%% 5. Compute & Display conditional effects

% 5.1 Specify additional variables to estimate conditional effects.

DGSA.ConditionalEffects.NbBins=3*ones(1,length(DGSA.ParametersNames));

% 5.2 Compute conditional effects

rng('default'); 
DGSA=ComputeConditionalEffects(DGSA);

% 5.3 Display conditional effcts

% Speicify the method to display standardized conditional effects.
DGSA.ConditionalEffects.Display.SensitivityByClusterAndBins=1; % if true, display pareto plots to visualize sensitivities by bins/clusters. 
DGSA.ConditionalEffects.Display.StandardizedSensitivity='Hplot'; % If omitted Pareto plot will be used. However, this is not recommended when there are lots of parameters

% Visualize conditional effects
DisplayConditionalEffects(DGSA,DGSA.ConditionalEffects.Display.StandardizedSensitivity)

% 5.4 Display class condtional CDFs
cdf_ConditionalEffects('owc','kvkh',DGSA,1)

%% 6. Check the response

% Use MDS plot to check the result of clustering
ClusterColor=DisplayMDSplot(DGSA.D, DGSA.Clustering);

% Specify texts for plots
PlotLabels.xlabtxt='Time (day)'; PlotLabels.ylabtxt='WWPT (stb)'; 
PlotLabels.xlimits=[0, 7300]; PlotLabels.ylimits=[0 7E7];

% Visualize dynamic response by clusters
dynamicbycluster(DGSA.Clustering,PriorResponses, ClusterColor,PlotLabels);


%% Save all variables for futher application
%save('VariablesSaved/DGSA_Completed.mat');
%save('VariablesSaved/DGSA_Spatial_Completed.mat');