1. Load existed ToolBox object if there is one in the saving_path.

2. Return the existed split settings if the data has already been split in ToolBox object, in case of overwriting.

alipy/toolbox.py

@@ -1,6 +1,7 @@
 import copy
 import os
 import pickle
+import warnings
 import inspect
 
 from sklearn.linear_model import LogisticRegression
@@ -168,8 +169,20 @@ def __init__(self, y, X=None, instance_indexes=None,
             self._saving_path = os.path.abspath(saving_path)
             if os.path.isdir(self._saving_path):
                 self._saving_dir = self._saving_path
+                if os.path.exists(os.path.join(saving_path, 'al_settings.pkl')):
+                    with open(os.path.join(saving_path, 'al_settings.pkl'), 'rb') as f:
+                        existed_toolbox = pickle.load(f)
+                        for ke in existed_toolbox.__dict__.keys():
+                            setattr(self, ke, getattr(existed_toolbox, ke))
+                        return
             else:
                 self._saving_dir = os.path.split(self._saving_path)[0]  # if a directory, a dir and None will return.
+                if os.path.exists(saving_path):
+                    with open(os.path.abspath(saving_path), 'rb') as f:
+                        existed_toolbox = pickle.load(f)
+                        for ke in existed_toolbox.__dict__.keys():
+                            setattr(self, ke, getattr(existed_toolbox, ke))
+                        return
             self.save()
 
     def split_AL(self, test_ratio=0.3, initial_label_rate=0.05,
@@ -209,6 +222,11 @@ def split_AL(self, test_ratio=0.3, initial_label_rate=0.05,
 
         """
         # should support other query types in the future
+        if self._split is True:
+            warnings.warn("Data has already been split. Return the existed split in case of overwriting.",
+                          category=RuntimeWarning)
+            return self.train_idx, self.test_idx, self.label_idx, self.unlabel_idx
+
         self.split_count = split_count
         if self._target_type != 'Features':
             if self._target_type != 'multilabel':
@@ -241,6 +259,7 @@ def split_AL(self, test_ratio=0.3, initial_label_rate=0.05,
                 saving_path=self._saving_path
             )
         self._split = True
+        self.save()
         return self.train_idx, self.test_idx, self.label_idx, self.unlabel_idx
 
     def get_split(self, round=None):