-
Notifications
You must be signed in to change notification settings - Fork 6
/
MT_FD_model.m
249 lines (220 loc) · 10.2 KB
/
MT_FD_model.m
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
classdef MT_FD_model < MT_baseclass
properties(GetAccess = 'public', SetAccess = 'private')
% remember what sort of model you are
Type
% weight vector for classification
labels
model_spec
model_spat
% internal label representation
classid
% parameters for convergence
maxItVar % maximum variation between iterations before convergence
maxNumVar % maximum number of dimensions allowed to not converge
end
methods
function obj = MT_FD_model(type, varargin)
% Constructor for multitask linear regression.
%
% Input:
% d: number of spectral features
% k: number of spatial features
% type: type of model used for spatial/spectral priors
obj@MT_baseclass(varargin{:})
obj.maxItVar = invarargin(varargin,'max_it_var');
if isempty(obj.maxItVar)
obj.maxItVar = 1e-2;
end
obj.maxNumVar = invarargin(varargin,'max_pct_var');
if isempty(obj.maxNumVar)
obj.maxNumVar = 1e-2;
end
% Initialize models for each dimension
assert(strcmp(type, 'linear') || strcmp(type, 'logistic'), ...
'type has to be linear or logistic.');
if strcmp(type, 'linear')
obj.model_spat = MT_linear('dim_reduce', 0, 'n_its', 1, 'prior_init_val', 1);
obj.model_spec = MT_linear('dim_reduce', 0, 'n_its', 1, 'prior_init_val', 0);
obj.classid=[1;-1];
elseif strcmp(type, 'logistic')
obj.model_spat = MT_logistic('dim_reduce', 0, 'n_its', 1, 'prior_init_val', 1);
obj.model_spec = MT_logistic('dim_reduce', 0, 'n_its', 1, 'prior_init_val', 0);
obj.classid=[1;0];
else
fprintf('Unknown model type, something went terribly wrong!\n');
end
obj.Type = type;
end
%function [] = init_prior(obj)
% obj.prior.spec = obj.spec_model.prior;
% obj.prior.spat = obj.spat_model.prior;
%end
function [] = init_prior(obj, d, k)
obj.model_spat.init_prior(d,1);
obj.model_spec.init_prior(k,0);
obj.prior.spec = obj.model_spec.prior;
obj.prior.spat = obj.model_spat.prior;
end
function prior = fit_prior(obj, Xcell, ycell, varargin)
% sanity checks
assert(length(Xcell) == length(ycell), 'unequal data and labels arrays');
assert(length(Xcell) > 1, 'only one dataset provided');
for i = 1:length(Xcell)
assert(size(Xcell{i},3) == length(ycell{i}), 'number of datapoints and labels differ');
ycell{i} = reshape(ycell{i},[],1);
end
lambda = invarargin(varargin,'lambda');
if isempty(lambda)
lambda = NaN;
end
cv = invarargin(varargin, 'cv');
% flag to get around infinite recursion...
if isempty(cv)
cv = 0;
end
if ~cv
assert(length(unique(cat(1,ycell{:}))) == 2, 'more than two classes present in the data');
obj.labels = [unique(cat(1,ycell{:})),obj.classid];
% replace labels for algorithm
for i = 1:length(ycell)
ycell{i} = MT_baseclass.swap_labels(ycell{i}, obj.labels, 'to');
end
obj.model_spat.labels = repmat(obj.classid,1,2);
obj.model_spec.labels = repmat(obj.classid,1,2);
obj.init_prior(size(Xcell{1},1),size(Xcell{1},2));
obj.prior.spat.W = zeros(size(obj.prior.spat.mu,1),length(Xcell));
obj.prior.spec.W = zeros(size(obj.prior.spec.mu,1),length(Xcell));
prior = fit_prior@MT_baseclass(obj, Xcell, ycell, 'lambda', lambda);
else
obj.init_prior(size(Xcell{1},1),size(Xcell{1},2));
obj.prior.spat.W = zeros(size(obj.prior.spat.mu,1),length(Xcell));
obj.prior.spec.W = zeros(size(obj.prior.spec.mu,1),length(Xcell));
prior = fit_prior@MT_baseclass(obj, Xcell, ycell, 'lambda', lambda);
end
end
function [b, conv] = convergence(obj, prior, prev_prior)
[b_spec, conv_spec] = obj.model_spec.convergence(prior.spec, prev_prior.spec);
[b_spat, conv_spat] = obj.model_spat.convergence(prior.spat, prev_prior.spat);
b = b_spec && b_spat;
conv = conv_spec + conv_spat;
end
function [w, error] = fit_model(obj, X, y, lambda)
num_chans = size(X,1);
%w{2} = ones(chans,1);
%w{1} = zeros(features,1); % I'm not convinced that a cell array is best here...but.....does it make sense to store things
% as rank-1 matrices
% and SVD them every
% time we want the
% components?
% Init FD model parameters
obj.model_spec.w = obj.model_spec.prior.mu;
obj.model_spat.w = obj.model_spat.prior.mu;
w_prev=zeros(num_chans, 1);
count2=0;
w = {obj.model_spec.w, obj.model_spat.w};
while sum(or(abs(w{2}) > (w_prev+obj.maxItVar*w_prev),abs(w{2}) < (w_prev-obj.maxItVar*w_prev)))>0 && count2< obj.nIts
w_prev = abs(w{2});
aX = dot3d(permute(X, [3, 2, 1]), obj.model_spat.w)';
obj.model_spec.w = obj.model_spec.fit_model(aX, y, lambda);
Xw = dot3d(permute(X, [3, 1, 2]), obj.model_spec.w)';
obj.model_spat.w = obj.model_spat.fit_model(Xw, y, lambda);
w = {obj.model_spec.w, obj.model_spat.w};
% EXPERIMENTAL--norm alpha to 1
%w{2}=w{2}/norm(w{2});
count2=count2+1;
end
error = obj.loss(w, X, y);
end
function out = fit_new_task(obj, X, y, varargin)
% fit_new_task(obj, X, y, varargin)
ML = invarargin(varargin,'ml');
if isempty(ML)
ML = 0;
end
out = struct();
% switch input labels using instance dictionary
y_train = MT_baseclass.swap_labels(y, obj.labels, 'to');
if ML
prev_loss = 0;
out.lambda = 1;
out.loss = 1;
count = 0;
while abs(prev_loss - out.loss) > obj.maxItVar * prev_loss && count < obj.nIts
prev_loss = out.loss;
[out.w, out.loss] = obj.fit_model(X, y_train, out.lambda);
out.lambda = 2*out.loss;
count = count+1;
if obj.verbose
fprintf('[new task fitting] ML lambda Iteration %d, lambda %.4e \n', count, out.lambda);
end
end
else
out.lambda = lambdaCV(@(X,y,lambda)(obj.fit_model(X{1},y{1},lambda)),...
@(w, X, y)(obj.loss(w, X{1}, y{1})),{X},{y_train});
[out.w, out.loss] = obj.fit_model(X, y_train, out.lambda);
end
out.predict = @(X)(obj.predict(out.w, X, obj.labels));
out.training_acc = mean(y == out.predict(X));
end
function [] = update_prior(obj, outputCell)
spec_weights = cell(length(outputCell), 1);
spat_weights = cell(length(outputCell), 1);
for i = 1: length(outputCell)
spec_weights{i} = outputCell{i}{1};
spat_weights{i} = outputCell{i}{2};
end
obj.model_spec.update_prior(spec_weights);
obj.model_spat.update_prior(spat_weights);
obj.prior.spec = obj.model_spec.prior;
obj.prior.spat = obj.model_spat.prior;
%outputCell
%W = [];
%A = [];
%for i = 1:length(outputCell)
% W = cat(2,W,outputCell{i}{1});
% A = cat(2,A, outputCell{i}{2});
%end
%obj.prior.weight = MT_baseclass.update_gaussian_prior(W, obj.trAdjust);
%obj.prior.alpha = MT_baseclass.update_gaussian_prior(A, obj.trAdjust);
end
function L = loss(obj, w, X, y)
Xw = dot3d(permute(X, [3, 1, 2]), w{1})';
aX = dot3d(permute(X, [3, 2, 1]), w{2})';
err1 = obj.model_spec.loss(w{1}, aX, y);
err2 = obj.model_spat.loss(w{2}, Xw, y);
L = err1 + err2;
end
function y = prior_predict(obj, X, varargin)
Xw = dot3d(permute(X, [3, 1, 2]), obj.model_spec.w)';
labels = invarargin(varargin, 'labels');
if isempty(labels)
y = obj.model_spat.prior_predict(Xw, 'labels', obj.labels);
else
y = obj.model_spat.prior_predict(Xw, 'labels', labels);
end
end
function [] = printswitches(obj)
fprintf('[MT FD Model] Model class: %s\n',obj.Type);
printswitches@MT_baseclass(obj);
end
end
methods(Static)
function y = predict(w, X, labels)
y = zeros(size(X,3),1);
for i = 1:length(y)
y(i) = sign(w{2}'*X(:,:,i)*w{1});
end
y = MT_baseclass.swap_labels(y, labels, 'from');
end
function [W] = multi_task_f(obj, Xtrain, Ytrain, lambda)
prior = obj.fit_prior(Xtrain, Ytrain, 'lambda', lambda, 'cv', 1);
W = {prior.spec.W, prior.spat.W};
end
function [loss] = multi_task_loss(obj,W,Xtest,Ytest)
loss = 0;
for i = 1:length(Xtest)
loss = loss + obj.loss({W{1}(:,i),W{2}(:,i)},Xtest{i},Ytest{i});
end
end
end
end