apaam/mlpack
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

added get_scores.m to niche/functional

Browse Source
This commit is contained in:
tekhnofiend
2008-01-15 20:06:19 +00:00
parent b525a6a9a1
commit e0ca41303e
3 changed files with 61 additions and 37 deletions
Download Patch File Download Diff File Expand all files Collapse all files
fastlib/u/niche/functional/funcica.m
+9 -5
View File
@@ -1,5 +1,5 @@
function [ic_curves_pos, ic_coef_pos, Y_pos, h_Y_pos, pc_coef, pc_curves, pc_scores, mean_coef, W_pos, whitening_transform] = ...
funcica(t, s, myfd_data, p, basis_curves, myfdPar);
funcica(t, s, myfd_data, p, basis_curves, myfdPar, basis_inner_products);
% funcica() - functional ICA
% first call prelim_funcica
% USAGE: [ic_curves_pos, ic_coef_pos, h_Y_pos] = funcica(t, s, data)
@@ -11,7 +11,11 @@ data_coef = getcoef(myfd_data);
pca_results = pca_fd(myfd_data, p, myfdPar);
pc_coef = getcoef(pca_results.harmfd);
%pc_curves = basis_curves * pc_coef;
pc_scores = pca_results.harmscr;
% pc_scores = pca_results.harmscr;% this doesn't work if lambda > 0
% instead, we do:
pc_scores = get_scores(data_coef, pc_coef', basis_inner_products);
%mean_coef = getcoef(pca_results.meanfd);
%figure(1); plot(t, pc_curves(:,1));
@@ -22,7 +26,7 @@ pc_scores = pca_results.harmscr;
% p_small should be automatically selected according to some
% reconstruction error threshold
%{
total_sum_var = 0;
for i = 1:p
total_sum_var = total_sum_var + sum(pc_scores(:,i).^2);
@@ -36,8 +40,8 @@ for p_small = 1:p
break
end
end
%}
p_small = 2;
% p_small = 2;
p_small
fastlib/u/niche/functional/get_scores.m
+22
View File
@@ -0,0 +1,22 @@
function scores = get_scores(data_coef, pc_coef, ...
basis_inner_products);
% USAGE: scores = get_scores(data_coef, pc_coef, basis_inner_products)
% data_coef is a matrix of size num_pc by N
% pc_coef is a matrix of num_pc by num_basis
% basis_inner_products = full(eval_penalty(mybasis, int2Lfd(0)))
% scores is a matrix of size N by num_pc
[num_pc, num_basis] = size(pc_coef);
N = size(data_coef, 2);
scores = zeros(N, num_pc);
for j = 1:num_pc
pc_coef_j = pc_coef(j,:);
for i = 1:N
scores(i,j) = ...
sum(sum((data_coef(:,i) * ...
pc_coef_j) .* basis_inner_products));
end
end
fastlib/u/niche/functional/test_smooth_funcica.m
+30 -32
View File
@@ -2,31 +2,31 @@
% h_Y2_test_set, h_Y_test_set, ...
% h_P1_train_set, h_P2_train_set, h_P_train_set, h_P1_test_set, ...
% h_P2_test_set, h_P_test_set] = ...
% test_smooth_funcica(data, t, s);
% test_smooth_funcica(data, t, s, basis_inner_products);
N = size(data, 2);
p = 30; % hardcoded for now
mybasis = create_bspline_basis([0 1], p, 4);
basis_curves = eval_basis(t, mybasis);
pen = full(eval_penalty(mybasis, int2Lfd(0)));
basis_inner_products = full(eval_penalty(mybasis, int2Lfd(0)));
myfd_data = data2fd(data, t, mybasis);
cut_fraction = .9;
cut = round(cut_fraction * N);
myfd_data_coef = getcoef(myfd_data);
data_coef = getcoef(myfd_data);
num_tests = 10;
num_tests = 1;
% a simple method for smoothing
%indices = 1:200:1000;
%data = data(indices,:);
%lambda_set = 0;
lambda_set = [0 1e-4 1e-3 5e-3 1e-2];
lambda_set = 1e-4;
%lambda_set = [0 1e-4 1e-3 5e-3 1e-2];
for lambda_i = 1:length(lambda_set)
@@ -48,26 +48,28 @@ for lambda_i = 1:length(lambda_set)
% extract and center train data
data_train = center(data(:, rand_train_indices));
myfd_data_train = data2fd(data_train, t, mybasis);
myfd_data_train_coef = getcoef(myfd_data_train);
data_train_coef = getcoef(myfd_data_train);
% extract and center test data
data_test = center(data(:, rand_test_indices));
myfd_data_test = data2fd(data_test, t, mybasis);
myfd_data_test_coef = getcoef(myfd_data_test);
data_test_coef = getcoef(myfd_data_test);
% alternate way of generating train and test data
% not used because no centering
%{
myfd_data_train_coef = myfd_data_coef(:,rand_train_indices);
myfd_data_test_coef = myfd_data_coef(:,rand_test_indices);
data_train_coef = data_coef(:,rand_train_indices);
data_test_coef = data_coef(:,rand_test_indices);
myfd_data_train = ...
fd(myfd_data_train_coef, getbasis(myfd_data), getnames(myfd_data));
fd(data_train_coef, getbasis(myfd_data), getnames(myfd_data));
myfd_data_test = ...
fd(myfd_data_test_coef, getbasis(myfd_data), getnames(myfd_data));
fd(data_test_coef, getbasis(myfd_data), getnames(myfd_data));
%}
[ic_curves_pos, ic_coef_pos, Y_pos, h_Y_pos, pc_coef, pc_curves, pc_scores, mean_coef, W, whitening_transform] = ...
funcica(t, s, myfd_data_train, p, basis_curves, myfdPar);
funcica(t, s, myfd_data_train, p, basis_curves, myfdPar, ...
basis_inner_products);
@@ -77,34 +79,30 @@ for lambda_i = 1:length(lambda_set)
scores_train = zeros(cut, p_small);
scores_test = zeros(N - cut, p_small);
% rescale the utilized parts of pc_coef such that
% the pc_curves square integrate to 1
%{
for j = 1:p_small
pc_coef_j = pc_coef(j,:);
alpha = sum(sum((pc_coef_j' * pc_coef_j) .* pen));
pc_coef(j,:) = pc_coef(j,:) / sqrt(alpha);
alpha = sqrt(sum(sum((pc_coef_j' * pc_coef_j) .* basis_inner_products)));
pc_coef(j,:) = pc_coef(j,:) / alpha;
end
%}
for j = 1:p_small
pc_coef_j = pc_coef(j,:);
for i = 1:cut
scores_train(i,j) = sum(sum((myfd_data_train_coef(:,i) * ...
pc_coef_j) .* pen));
end
for i = 1:(N - cut)
scores_test(i,j) = sum(sum((myfd_data_test_coef(:,i) * ...
pc_coef_j) .* pen));
end
end
scores_train = scores_train';
scores_test = scores_test';
scores_train = ...
get_scores(data_train_coef, ...
pc_coef(1:p_small,:), ...
basis_inner_products)';
scores_test = ...
get_scores(data_train_coef, ...
pc_coef(1:p_small,:), ...
basis_inner_products)';
Y_scores_train = W * scores_train(1:2,:);
Y_scores_test = W * scores_test(1:2,:);