Modified PCA analysis script to run on simulated data.
This commit is contained in:
parent
1d178a3d75
commit
e3571c12b0
@ -1,78 +1,210 @@
|
|||||||
%% STEP 1: Generate synthetic image dataset
|
%% STEP 1: Extract Images
|
||||||
clear; close all; clc;
|
clear; close all; clc;
|
||||||
|
|
||||||
% Parameters
|
baseDir = 'D:/Results - Numerics/Data_Full3D/PhaseTransition/DTS/Point2';
|
||||||
numImages = 50; % number of images in the dataset
|
JobNumber = 0;
|
||||||
imgSize = [50 50]; % image dimensions
|
runFolder = sprintf('Run_%03d', JobNumber);
|
||||||
|
savefileName = 'DropletsToStripes'; % Output file name
|
||||||
|
datafileName = './DropletsToStripes.mat';
|
||||||
|
reverseOrder = false; % Set this to true to reverse the theta ordering
|
||||||
|
|
||||||
% Preallocate array
|
TitleString = 'Droplets To Stripes';
|
||||||
images = zeros([imgSize, numImages]);
|
|
||||||
|
|
||||||
% Generate base features: a horizontal stripe + diagonal gradient
|
%%
|
||||||
[x, y] = meshgrid(1:imgSize(2), 1:imgSize(1));
|
baseDir = 'D:/Results - Numerics/Data_Full3D/PhaseTransition/STD/Point2';
|
||||||
|
JobNumber = 0;
|
||||||
|
runFolder = sprintf('Run_%03d', JobNumber);
|
||||||
|
savefileName = 'StripesToDroplets'; % Output file name
|
||||||
|
datafileName = './StripesToDroplets.mat';
|
||||||
|
reverseOrder = true; % Set this to true to reverse the theta ordering
|
||||||
|
|
||||||
for k = 1:numImages
|
TitleString = 'Stripes To Droplets';
|
||||||
% Feature 1: horizontal stripe
|
|
||||||
stripe = double(y > 20 & y < 30);
|
%%
|
||||||
|
folderList = dir(baseDir);
|
||||||
% Feature 2: diagonal gradient
|
isValid = [folderList.isdir] & ~ismember({folderList.name}, {'.', '..'});
|
||||||
gradient = (x + y) / max(x(:) + y(:));
|
folderNames = {folderList(isValid).name};
|
||||||
|
nimgs = numel(folderNames);
|
||||||
% Random variation in stripe intensity
|
|
||||||
stripeIntensity = 0.8 + 0.4*randn(1);
|
% Extract theta values from folder names
|
||||||
|
PolarAngleVals = zeros(1, nimgs);
|
||||||
% Random rotation of gradient strength
|
|
||||||
gradIntensity = 0.5 + 0.2*randn(1);
|
for k = 1:nimgs
|
||||||
|
tokens = regexp(folderNames{k}, 'theta_(\d{3})', 'tokens');
|
||||||
% Combine features + Gaussian noise
|
if isempty(tokens)
|
||||||
images(:,:,k) = stripeIntensity*stripe + gradIntensity*gradient + 0.1*randn(imgSize);
|
warning('No theta found in folder name: %s', folderNames{k});
|
||||||
|
PolarAngleVals(k) = NaN;
|
||||||
|
else
|
||||||
|
PolarAngleVals(k) = str2double(tokens{1}{1});
|
||||||
|
end
|
||||||
end
|
end
|
||||||
|
|
||||||
%% STEP 2: Visualize some sample images
|
% Choose sort direction
|
||||||
figure;
|
sortDirection = 'ascend';
|
||||||
for i = 1:6
|
if reverseOrder
|
||||||
subplot(2,3,i);
|
sortDirection = 'descend';
|
||||||
imagesc(images(:,:,i));
|
|
||||||
axis image off; colormap gray;
|
|
||||||
title(sprintf('Image %d', i));
|
|
||||||
end
|
end
|
||||||
sgtitle('Sample Synthetic Images');
|
|
||||||
|
% Sort folderNames based on polar angle
|
||||||
|
[~, sortIdx] = sort(PolarAngleVals, sortDirection);
|
||||||
|
folderNames = folderNames(sortIdx);
|
||||||
|
PolarAngleVals = PolarAngleVals(sortIdx); % Optional: if you still want sorted list
|
||||||
|
imgs = cell(1, nimgs);
|
||||||
|
alphas = zeros(1, nimgs);
|
||||||
|
|
||||||
|
for k = 1:nimgs
|
||||||
|
folderName = folderNames{k};
|
||||||
|
SaveDirectory = fullfile(baseDir, folderName, runFolder);
|
||||||
|
|
||||||
|
% Extract alpha (theta) again from folder name
|
||||||
|
tokens = regexp(folderName, 'theta_(\d{3})', 'tokens');
|
||||||
|
alpha_val = str2double(tokens{1}{1});
|
||||||
|
alphas(k) = alpha_val;
|
||||||
|
|
||||||
|
matPath = fullfile(SaveDirectory, 'psi_gs.mat');
|
||||||
|
if ~isfile(matPath)
|
||||||
|
warning('Missing psi_gs.mat in %s', SaveDirectory);
|
||||||
|
continue;
|
||||||
|
end
|
||||||
|
|
||||||
|
try
|
||||||
|
Data = load(matPath, 'psi', 'Params', 'Transf', 'Observ');
|
||||||
|
catch ME
|
||||||
|
warning('Failed to load %s: %s', matPath, ME.message);
|
||||||
|
continue;
|
||||||
|
end
|
||||||
|
|
||||||
|
Params = Data.Params;
|
||||||
|
Transf = Data.Transf;
|
||||||
|
Observ = Data.Observ;
|
||||||
|
|
||||||
|
psi = Data.psi;
|
||||||
|
if isgpuarray(psi)
|
||||||
|
psi = gather(psi);
|
||||||
|
end
|
||||||
|
if isgpuarray(Observ.residual)
|
||||||
|
Observ.residual = gather(Observ.residual);
|
||||||
|
end
|
||||||
|
|
||||||
|
% Axes and projection
|
||||||
|
z = Transf.z * Params.l0 * 1e6;
|
||||||
|
dz = z(2)-z(1);
|
||||||
|
|
||||||
|
if k == 1
|
||||||
|
x = Transf.x * Params.l0 * 1e6;
|
||||||
|
y = Transf.y * Params.l0 * 1e6;
|
||||||
|
% Calculate frequency increment (frequency axes)
|
||||||
|
Nx = length(x); % grid size along X
|
||||||
|
Ny = length(y); % grid size along Y
|
||||||
|
dx = mean(diff(x)); % real space increment in the X direction (in micrometers)
|
||||||
|
dy = mean(diff(y)); % real space increment in the Y direction (in micrometers)
|
||||||
|
dvx = 1 / (Nx * dx); % reciprocal space increment in the X direction (in micrometers^-1)
|
||||||
|
dvy = 1 / (Ny * dy); % reciprocal space increment in the Y direction (in micrometers^-1)
|
||||||
|
|
||||||
|
% Create the frequency axes
|
||||||
|
vx = (-Nx/2:Nx/2-1) * dvx; % Frequency axis in X (micrometers^-1)
|
||||||
|
vy = (-Ny/2:Ny/2-1) * dvy; % Frequency axis in Y (micrometers^-1)
|
||||||
|
|
||||||
|
% Create the Wavenumber axes
|
||||||
|
kx_full = 2*pi*vx; % Wavenumber axis in X
|
||||||
|
ky_full = 2*pi*vy; % Wavenumber axis in Y
|
||||||
|
end
|
||||||
|
|
||||||
|
n = abs(psi).^2;
|
||||||
|
nxy = squeeze(trapz(n * dz, 3));
|
||||||
|
|
||||||
|
imgs{k} = nxy;
|
||||||
|
end
|
||||||
|
|
||||||
|
%% STEP 2: Visualize images
|
||||||
|
figure(1);
|
||||||
|
set(gcf,'Position',[100 100 950 750])
|
||||||
|
font = 'Bahnschrift';
|
||||||
|
t = tiledlayout(4, 3, 'TileSpacing', 'compact', 'Padding', 'compact'); % 1x4 grid
|
||||||
|
|
||||||
|
for i = 1:nimgs
|
||||||
|
ax1 = nexttile;
|
||||||
|
imagesc(x, y, imgs{i}')
|
||||||
|
% Define normalized positions (relative to axis limits)
|
||||||
|
x_offset = 0.025; % 5% offset from the edges
|
||||||
|
y_offset = 0.025; % 5% offset from the edges
|
||||||
|
axis square;
|
||||||
|
hcb = colorbar;
|
||||||
|
colormap(ax1, Colormaps.plasma());
|
||||||
|
set(gca, 'FontSize', 14); % For tick labels only
|
||||||
|
hL = ylabel(hcb, 'Optical Density');
|
||||||
|
set(hL,'Rotation',-90);
|
||||||
|
set(gca,'YDir','normal')
|
||||||
|
% set(gca, 'YTick', linspace(y_min, y_max, 5)); % Define y ticks
|
||||||
|
% set(gca, 'YTickLabel', flip(linspace(y_min, y_max, 5))); % Flip only the labels
|
||||||
|
hXLabel = xlabel('x (µm)', 'Interpreter', 'tex');
|
||||||
|
hYLabel = ylabel('y (µm)', 'Interpreter', 'tex');
|
||||||
|
hTitle = title('Density', 'Interpreter', 'tex');
|
||||||
|
set([hXLabel, hYLabel, hL], 'FontName', font)
|
||||||
|
set([hXLabel, hYLabel, hL], 'FontSize', 14)
|
||||||
|
set(hTitle, 'FontName', font, 'FontSize', 16, 'FontWeight', 'bold'); % Set font and size for title
|
||||||
|
|
||||||
|
end
|
||||||
|
|
||||||
|
sgtitle('Ground States');
|
||||||
|
|
||||||
%% STEP 3: Reshape images into a 2D data matrix
|
%% STEP 3: Reshape images into a 2D data matrix
|
||||||
% Each image is a row, each pixel is a column
|
% Each image will be a row vector, so size(X) = [numImages × numPixels]
|
||||||
X = reshape(images, [], numImages)'; % size: [numImages × numPixels]
|
|
||||||
|
% First, convert cell array to numeric 3D array
|
||||||
|
allImgs = cat(3, imgs{:}); % size: [Nx × Ny × numImages]
|
||||||
|
|
||||||
|
% Store image size for reshaping later
|
||||||
|
imgSize = size(allImgs(:,:,1)); % [Nx, Ny]
|
||||||
|
|
||||||
|
% Reshape into 2D matrix for PCA
|
||||||
|
X = reshape(allImgs, [], nimgs)'; % [numImages × (Nx*Ny)]
|
||||||
|
|
||||||
%% STEP 4: Perform PCA
|
%% STEP 4: Perform PCA
|
||||||
[coeff, score, latent, tsquared, explained] = pca(X);
|
[coeff, score, latent, tsquared, explained] = pca(X);
|
||||||
|
|
||||||
% coeff : principal component directions (eigenvectors) in pixel space
|
% coeff : principal component directions (eigenvectors) in pixel space
|
||||||
% score : projection of each image onto the principal components
|
% score : projection of each image onto the principal components
|
||||||
% latent : eigenvalues (variance captured by each PC)
|
% latent : eigenvalues (variance captured by each PC)
|
||||||
% explained : percentage variance explained by each PC
|
% explained : percentage variance explained by each PC
|
||||||
|
|
||||||
%% STEP 5: Visualize variance explained
|
%% Visualize cumulative variance
|
||||||
figure;
|
figure(2);
|
||||||
pareto(explained);
|
set(gcf,'Position',[100 100 950 750])
|
||||||
xlabel('Principal Component');
|
cumVar = cumsum(explained);
|
||||||
ylabel('Variance Explained (%)');
|
plot(cumVar, 'o-', 'LineWidth', 1.5);
|
||||||
title('PCA Variance Explained');
|
yline(95, '--r', '95% threshold');
|
||||||
|
xlabel('Number of Principal Components');
|
||||||
|
ylabel('Cumulative Variance (%)');
|
||||||
|
title('Minimum number of PCA components required for a basis');
|
||||||
|
grid on;
|
||||||
|
|
||||||
%% STEP 6: View first few principal components as "eigenimages"
|
%% Visualize first few principal components
|
||||||
numPCsToShow = 4;
|
numPCsToShow = 7;
|
||||||
figure;
|
figure(3);
|
||||||
|
set(gcf,'Position',[100 100 950 750])
|
||||||
|
t = tiledlayout(3, 3, 'TileSpacing', 'compact', 'Padding', 'compact'); % 1x4 grid
|
||||||
for i = 1:numPCsToShow
|
for i = 1:numPCsToShow
|
||||||
pcImage = reshape(coeff(:,i), imgSize); % reshape back to image form
|
pcImage = reshape(coeff(:,i), imgSize); % reshape back to image form
|
||||||
subplot(1,numPCsToShow,i);
|
nexttile;
|
||||||
imagesc(pcImage);
|
imagesc(pcImage)
|
||||||
axis image off; colormap gray;
|
axis image off; colormap(Colormaps.coolwarm());
|
||||||
title(sprintf('PC %d', i));
|
title(sprintf('PC %d (%.1f%%)', i, explained(i)));
|
||||||
end
|
end
|
||||||
sgtitle('First Principal Components (Eigenimages)');
|
sgtitle('First Principal Components (Eigenimages)');
|
||||||
|
|
||||||
%% STEP 7: View projection (scores) in PC space
|
%% Visualize evolution of PCA scores
|
||||||
figure;
|
figure(4);
|
||||||
scatter(score(:,1), score(:,2), 50, 'filled');
|
set(gcf,'Position',[100 100 950 750])
|
||||||
xlabel('PC 1 Score');
|
plot(alphas, score(:,1), '-o', 'LineWidth', 1.5, 'DisplayName', 'PC1 Score');
|
||||||
ylabel('PC 2 Score');
|
hold on;
|
||||||
title('Images in Principal Component Space');
|
plot(alphas, score(:,2), '-o', 'LineWidth', 1.5, 'DisplayName', 'PC2 Score');
|
||||||
grid on;
|
% plot(alphas, score(:,3), '-o', 'LineWidth', 1.5, 'DisplayName', 'PC3 Score');
|
||||||
|
% plot(alphas, score(:,4), '-o', 'LineWidth', 1.5, 'DisplayName', 'PC4 Score');
|
||||||
|
% plot(alphas, score(:,5), '-o', 'LineWidth', 1.5, 'DisplayName', 'PC5 Score');
|
||||||
|
% plot(alphas, score(:,6), '-o', 'LineWidth', 1.5, 'DisplayName', 'PC6 Score');
|
||||||
|
% plot(alphas, score(:,7), '-o', 'LineWidth', 1.5, 'DisplayName', 'PC7 Score');
|
||||||
|
legend('Location', 'northeast');
|
||||||
|
xlabel('Polar Angle (°)'); ylabel('PC Score');
|
||||||
|
title('Evolution of PC Scores');
|
||||||
|
grid on;
|
||||||
Loading…
Reference in New Issue
Block a user