Updated README to reference Python data and streamlined some of the t…

…estscript. Currently a problem with the fit_new_function for the logistic FD

MT_baseclass.m

@@ -95,6 +95,8 @@
             obj.parallel = invarargin(varargin,'parallel');
             if isempty(obj.parallel)
                 obj.parallel = 0;
+            else
+                fprintf('[MT base] Attempting parallel implementation with %d cores\n',obj.parallel);
             end
             obj.prior = struct();
             obj.prior.lambda = 1;

README.md

@@ -57,7 +57,8 @@ There is an enormous space of possibilties for how this framework can be extende
 Please feel free (indeed, urged) to let me know through the issues feature whether something is not working and I will be happy to fix them as soon as I can. If preferrable, feel free also to send mail to vjayaram@tue.mpg.de 
 
 # Python
-Python version coming shortly...
+For the python Python version please check out our related [page](https://github.com/bibliolytic/pyMTL). 
+
 
 # Citations:
 

lambdaCV.m

@@ -13,7 +13,7 @@
 % Optional Arguments
 %       n:             Number of CV loops (default 5)
 %       parallel:    Parallel loops (<num cores> | none)
-%       lrange:     Vector of lambda values (default exp(-6:10))
+%       lrange:     Vector of lambda values (default [exp(-6),exp(-1:0.1:1),exp(6)])
 %       verbose:             boolean, verbose (default 0)
 %       bootstrap: boolean, bootstrap to equalize classes (default 1)
 
@@ -43,7 +43,7 @@
 
 lrange = invarargin(varargin,'lrange');
 if isempty(lrange)
-    lrange=[exp(-6),exp(-2:0.2:2),exp(6)];
+    lrange=[exp(-6),exp(-1:0.1:1),exp(6)];
 end
 
 %% Main code

testscript.m

@@ -23,44 +23,44 @@
     disp('Confirm prior computation switches: ');
     linear_model{i}.printswitches;
 
-    % Code to fit the prior
+    % Code to fit the prior (training on the first 4)
     disp('Training L2 loss prior...')
-    linear_model{i}.fit_prior(T_X2d, T_y);
+    linear_model{i}.fit_prior(T_X2d(1:4), T_y(1:4));
     disp('Training logistic loss prior...')
-    log_model{i}.fit_prior(T_X2d, T_y);
+    log_model{i}.fit_prior(T_X2d(1:4), T_y(1:4));
 
-    % Code that computes prior accuracy on the training data
-    pacc_lin = mean(linear_model{i}.prior_predict(X2d_s) == y_s);
-    pacc_log = mean(log_model{i}.prior_predict(X2d_s) == y_s);
-    fprintf('prior accuracies: \n Linear: %.2f\n Logistic: %.2f\n', pacc_lin, pacc_log);
+    % Code that computes prior accuracy on the held-out session data
+    pacc_lin = mean(linear_model{i}.prior_predict(T_X2d{5}) == T_y{5});
+    pacc_log = mean(log_model{i}.prior_predict(T_X2d{5}) == T_y{5});
+    fprintf('Prior accuracies on held-out session: \n Linear: %.2f\n Logistic: %.2f\n', pacc_lin, pacc_log);
 
     % Code to fit the new task (with cross-validated lambda)
-    fitted_new_linear_task = linear_model{i}.fit_new_task(X2d_s, y_s, 'ml', 1);
-    fitted_new_log_task = log_model{i}.fit_new_task(X2d_s, y_s, 'ml', 1);
+    fitted_new_linear_task = linear_model{i}.fit_new_task(T_X2d{5}, T_y{5}, 'ml', 0);
+    fitted_new_log_task = log_model{i}.fit_new_task(T_X2d{5},T_y{5}, 'ml', 0);
 
     % Classifying after the new task update
         fprintf('New task *training set* accuracy: \n Linear: %.2f\nLogistic: %.2f\n',...
-        mean(fitted_new_linear_task.predict(X2d_s) == y_s), ...
-        mean(fitted_new_log_task.predict(X2d_s) == y_s));
+        mean(fitted_new_linear_task.predict(T_X2d{5}) == T_y{5}), ...
+        mean(fitted_new_log_task.predict(T_X2d{5}) == T_y{5}));
 end
 
 %%
 %%%%%%%%%%%%%%%%%%%%%%%%%%
 % How to use the bilinear version of this approach
 %%%%%%%%%%%%%%%%%%%%%%%%%%%
 
-type = {'linear'};%,'logistic'};
+type = {'linear', 'logistic'};
 
 for i = 1:length(type)
     disp(['********************FD ', type{i},'**********************']);
-    FD{i} = MT_FD_model(type{i},'n_its',5,'verbose',1);
+    FD{i} = MT_FD_model(type{i},'n_its',5,'verbose',0);
     FD{i}.printswitches;
-    FD{i}.fit_prior(T_X, T_y);
-    acc = mean(FD{i}.prior_predict(X_s) == y_s);
-fprintf('Prior accuracy: %.2f\n', acc*100);
-out = FD{i}.fit_new_task(X_s, y_s, 'ml', 1);
-acc = mean(out.predict(X_s) == y_s);
-fprintf('New task training accuracy: %.2f\n', acc*100);
+    FD{i}.fit_prior(T_X(1:4), T_y(1:4));
+    acc = mean(FD{i}.prior_predict(T_X{5}) == T_y{5});
+    fprintf('Prior accuracy on held-out data: %.2f\n', acc*100);
+    out = FD{i}.fit_new_task(T_X{5}, T_y{5}, 'ml', 0);
+    acc = mean(out.predict(T_X{5}) == T_y{5});
+    fprintf('New task training accuracy: %.2f\n', acc*100);
 end
 
 fprintf('Script finished!\n');