added intercept and verbosity

Author: JanBenisek

cobra/cobra.py

@@ -35,6 +35,8 @@ class COBRA(object):
     
     ***ATTRIBUTES***
     :_partition_dict:           Dict with partitioned DFs X/Y train/selection/validation
+    :_headers_dict:             Dict of 4 lists with header names (object, numeric, bool, other)
+    :_partitioning_settings:    Dict with train/sel/valid sets with their size
     ----------------------------------------------------
     __init__: contains variables which are established with the object. 
               If some of them is changed, then the whole process must be redone(call the class again),
@@ -90,6 +92,9 @@ def transform(self):
 
         df_trans = dtrans.transform()
 
+        self._headers_dict = dpc._headers_dict
+        self._partitioning_settings = dpc._partitioning_settings
+        
         return df_trans
 
 
@@ -108,7 +113,7 @@ def fit_univariate(self, df_t, preselect_auc=0.53, preselect_overtrain=5):
 
         return df_sel, df_corr
 
-    def fit_model(self, df_t, df_us, modeling_nsteps=30, forced_vars=None, excluded_vars=None, name=None, verbose=False):
+    def fit_model(self, df_t, df_us, modeling_nsteps=30, forced_vars=None, excluded_vars=None, name=None, verbose=False, positive_only=True):
         '''
         Method fits and finds best model
         Returns dataframe with all the info - forward selection, AUC, importance...
@@ -120,9 +125,10 @@ def fit_model(self, df_t, df_us, modeling_nsteps=30, forced_vars=None, excluded_
         excluded_vars: list with variables to be excluded in the model
         name: name of the model
         verbose: whether immediate steps of the procedure should be printed to the console
+        positive_only: whether only positive coeficients should be considered (recommended to stay so)
         ---------------------------------------------------- 
         '''
-        modsel = ms.ModelSelection(verbose=verbose)
+        modsel = ms.ModelSelection(verbose=verbose, positive_only=positive_only)
 
         df_models = modsel.fit(df_t, 
                                df_us,

cobra/model_selection.py

@@ -26,11 +26,13 @@ class ModelSelection(object):
     ***ATTRIBUTES***
     :_partition_dict:             Dict with partitioned DFs X/Y train/selection/validation
     :_optimal_nvars:              Optimal number of variables
+    :positive_only:               Whether all coeficients should be positivr
     ---------------------------------------------------- 
     '''
 
-    def __init__(self, verbose):
+    def __init__(self, verbose, positive_only):
         self.verbose = verbose
+        self.positive_only = positive_only
 
 
     def fit(self, df_trans, df_unisel, modeling_nsteps, forced_vars, excluded_vars, name):
@@ -93,7 +95,7 @@ def _getTrainSelectValidXY(self, df):
 
         return dict_out
 
-    def _forwardSelection(self, df_sel, forced_vars, excluded_vars, positive_only=True):
+    def _forwardSelection(self, df_sel, forced_vars, excluded_vars):
         '''
         Method performs forward selection
         Returns DF with performance
@@ -121,6 +123,7 @@ def _forwardSelection(self, df_sel, forced_vars, excluded_vars, positive_only=Tr
         df_forward_selection = pd.DataFrame(None,columns=[
                                                           'step',
                                                           'coef',
+                                                          'intercept',
                                                           'all_coefs_positive',
                                                           'auc_train',
                                                           'auc_selection',
@@ -178,6 +181,7 @@ def _forwardSelection(self, df_sel, forced_vars, excluded_vars, positive_only=Tr
                 df_forward_selection.loc[row] = [
                                                  step,                      #Step
                                                  logit.coef_,               #coef
+                                                 logit.intercept_,          #intercept
                                                  all_coefs_positive,        #all_coefs_positive
                                                  AUC_train,                 #auc_train
                                                  AUC_selection,             #auc_selection
@@ -194,12 +198,14 @@ def _forwardSelection(self, df_sel, forced_vars, excluded_vars, positive_only=Tr
                 row +=1
 
             #Only positive coefs
-            if positive_only:
+            if self.positive_only:
                 all_coefs_negative = len(df_forward_selection[(df_forward_selection['all_coefs_positive'] == True) & (df_forward_selection['step'] == step)]) == 0
 
                 if all_coefs_negative:
                     if self.verbose:
-                        print('No models with only positive coefficients, step skipped.')
+                        
+                        print('No models with only positive coefficients, following step skipped.')
+                        print(df_forward_selection[(df_forward_selection['all_coefs_positive'] == True) & (df_forward_selection['step'] == step)])                        
                         #Skip the next steps and go the next iteration
                         #The fitted models are not of interest if the user explicitly
                         # says positive_only=True