[90] | 1 | %LASSOC |
---|
| 2 | % |
---|
| 3 | % W = LASSOC(X, LAMBDA) |
---|
| 4 | % |
---|
| 5 | % Train the LASSO classifier on dataset X. LAMBDA is the regularization |
---|
| 6 | % parameter. |
---|
| 7 | |
---|
[5] | 8 | function w = lassoc(x, lambda) |
---|
| 9 | |
---|
| 10 | mustScale=0; |
---|
| 11 | |
---|
| 12 | if (nargin < 2) |
---|
| 13 | %prwarning(3,'Lambda set to one'); |
---|
| 14 | lambda = 1; |
---|
| 15 | end |
---|
| 16 | if (nargin < 1) | (isempty(x)) |
---|
[90] | 17 | w = prmapping(mfilename,{lambda}); |
---|
[5] | 18 | w = setname(w,'LASSO classifier'); |
---|
| 19 | return |
---|
| 20 | end |
---|
| 21 | |
---|
| 22 | if ~ismapping(lambda) % train the mapping |
---|
| 23 | |
---|
| 24 | % Unpack the dataset. |
---|
| 25 | islabtype(x,'crisp'); |
---|
| 26 | %isvaldset(x,1,2); % at least 1 object per class, 2 classes |
---|
| 27 | [n,k,c] = getsize(x); |
---|
| 28 | |
---|
| 29 | % Is this necessary?? |
---|
| 30 | if mustScale |
---|
| 31 | wsc = scalem(x,'variance'); |
---|
| 32 | x.data = x.data*wsc; |
---|
| 33 | end |
---|
| 34 | |
---|
| 35 | if c ~= 2 % two-class classifier: |
---|
| 36 | error('Only a two-class classifier is implemented'); |
---|
| 37 | end |
---|
| 38 | |
---|
| 39 | beta=-lasso(+x,3-2*getnlab(x),lambda); |
---|
| 40 | |
---|
| 41 | % now find out how sparse the result is: |
---|
| 42 | nr = sum(abs(beta)>1.0e-8); |
---|
| 43 | |
---|
| 44 | % and store the results: |
---|
| 45 | if mustScale |
---|
| 46 | W.wsc = wsc; |
---|
| 47 | end |
---|
| 48 | |
---|
| 49 | W.beta = beta; % the ultimate weights |
---|
| 50 | W.nr = nr; |
---|
[90] | 51 | w = prmapping(mfilename,'trained',W,getlablist(x),size(x,2),c); |
---|
[5] | 52 | w = setname(w,'LASSO classifier'); |
---|
| 53 | |
---|
| 54 | else |
---|
| 55 | % Evaluate the classifier on new data: |
---|
| 56 | W = getdata(lambda); |
---|
| 57 | n = size(x,1); |
---|
| 58 | |
---|
| 59 | % scaling and linear classifier: |
---|
| 60 | if mustScale |
---|
| 61 | x = x*W.wsc; |
---|
| 62 | end |
---|
| 63 | out = x*W.beta; |
---|
| 64 | |
---|
| 65 | % and put it nicely in a prtools dataset: |
---|
| 66 | w = setdat(x,sigm([-out out]),lambda); |
---|
| 67 | |
---|
| 68 | end |
---|
| 69 | |
---|
| 70 | return |
---|