-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_overall.m
36 lines (31 loc) · 1.27 KB
/
train_overall.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
function [w,b,lambda_1,maxiter,mv] = train_overall( tr_cell,val_cell,label_set_child,level_mark)
%UNTITLED3 Summary of this function goes here
% Detailed explanation goes here
[X_tr,Y_tr]=gen_train_data(tr_cell,label_set_child);
[X_val,Y_val]=gen_train_data(val_cell,label_set_child);
[lambda_1,maxiter,mv]=findparams_bayesopt(X_tr,Y_tr,X_val,Y_val,level_mark);
%now overall training
%aggregate the training and validation data
train_data_overall=[];
label_data_overall=[];
num_class=size(1,size(label_set_child,2));
for i=1:num_class
label_set_curr=label_set_child{i};
train_temp=[];
for j=1:size(label_set_curr,2)
train_temp=[train_temp,tr_cell{label_set_curr(j)},val_cell{label_set_curr(j)}];
end
train_data_overall=[train_data_overall,train_temp];
if(i==1)
label_data_overall=[label_data_overall,1*ones(1,size(train_temp,2))];
else
label_data_overall=[label_data_overall,-1*ones(1,size(train_temp,2))];
end
end
if(level_mark==1)
[w,b]=SGD_hinge_normal(train_data_overall,label_data_overall,lambda_1,maxiter);
else
%node of the other level
[w,b]=SGD_hinge_normal(train_data_overall,label_data_overall,lambda_1,maxiter);
end
end