Merge branch 'master' of https://github.com/okbalefthanded/BCI-Baseline

baseline/erp/blda.py

@@ -22,18 +22,26 @@ def fit(self, X, y=None):
         # compute regression targets from class labels 
         # (to do lda via regression)      
         n_instances = y.shape[0]
-
-        y = y.astype(np.float32)
-        y = y.T
-        classes = np.unique(y)
+        yy = y.astype(np.float32)
+        yy = yy.T
+        # y = y.astype(np.float32)
+        # y = y.T
+        # classes = np.unique(y)
+        classes = np.unique(yy)
         if -1 in classes:
             self.neg_class = -1.  
+        '''
         n_posexamples = np.sum(y==self.pos_class)
         n_negexamples = np.sum(y==self.neg_class)
         n_examples = n_posexamples + n_negexamples
         y[y==self.pos_class] = n_examples / n_posexamples
         y[y==self.neg_class] = -n_examples / n_negexamples
-
+        '''
+        n_posexamples = np.sum(yy==self.pos_class)
+        n_negexamples = np.sum(yy==self.neg_class)
+        n_examples = n_posexamples + n_negexamples
+        yy[yy==self.pos_class] = n_examples / n_posexamples
+        yy[yy==self.neg_class] = -n_examples / n_negexamples
         # n_posexamples = np.sum(y==1)
         # n_negexamples = np.sum(y==-1)
         # n_examples = n_posexamples + n_negexamples
@@ -130,5 +138,5 @@ def predict_proba(self,X):
         # return softmax(self.decision_function(X))
         # return self.decision_function(X)
         decision = self.decision_function(X)
-        return expit(decision)
+        return expit(decision).squeeze()
 

baseline/erp/stepwise/stepwise.py

@@ -104,6 +104,7 @@ def stepcalc(allx, y, inmodel):
     # to rows of X that were thrown out.
     b = np.zeros((nin, 1))
     Qb = np.dot(Q.conj().T, y)
+    print(Qb.shape, Qb.min(), Qb.max())
     Qb[abs(Qb) < tol * max(abs(Qb))] = 0
     b[perm] = linalg.solve(R, Qb)