Fix bugs and improve cache

column2Vec/Column2Vec.py

@@ -16,10 +16,11 @@ class Cache:
     __cache = pd.DataFrame()
     __read_from_file = False
     __on = True
+    __file = "generated/cache.txt"
 
     def __read(self):
         try:
-            self.__cache = pd.io.parsers.read_csv("generated/cache.txt", index_col=0)
+            self.__cache = pd.io.parsers.read_csv(self.__file, index_col=0)
         except Exception as error:
             pass
 
@@ -61,13 +62,25 @@ def save_persistently(self):
             return
         print(self.__cache.index)
         print(self.__cache.columns)
-        self.__cache.to_csv("generated/cache.txt", index=True)
+        self.__cache.to_csv(self.__file, index=True)
 
     def off(self):
         self.__on = False
     def on(self):
         self.__on = True
 
+    def set_file(self, file: str):
+        self.__file = file
+
+    def clear_cache(self):
+        self.__cache = self.__cache[0:0]
+        self.__read_from_file = False
+    def clear_persistent_cache(self):
+        try:
+            open(self.__file, 'w').close()
+        except FileNotFoundError as e:
+            print(e)
+
 
 cache = Cache()
 
@@ -147,7 +160,7 @@ def weighted_create_embed(column: pd.Series, model: SentenceTransformer, key: st
     """
     res = cache.get_cache(key, function_string)
     if res is not None:
-        return res
+        return res, None
 
     uniq_column = column.value_counts(normalize=True)
     weights = uniq_column.values
@@ -182,8 +195,18 @@ def column2vec_weighted_avg(column: pd.Series, model: SentenceTransformer, key:
     Convert each item in the column to a vector and return the weighted average of all the vectors
     """
     function_string = "column2vec_weighted_avg"
-    encoded_columns, weights = weighted_create_embed(column, model, key, function_string)
-    to_ret = np.average(encoded_columns, axis=0, weights=weights)  # counts weighted average
+    res = cache.get_cache(key, function_string)
+    if res is not None:
+        return res
+    uniq_column = column.value_counts(normalize=True)
+    weights = uniq_column.values
+    column_clean = pd.Series(uniq_column.keys()).apply(lambda x: re.sub("[^(0-9 |a-z)]",
+                                                                        " ", str(x).lower())).values
+    res = model.encode(column_clean)
+
+
+    # encoded_columns, weights = weighted_create_embed(column, model, key, function_string)
+    to_ret = np.average(res, axis=0, weights=weights)  # counts weighted average
     cache.save(key, function_string, to_ret)
     return to_ret
 
@@ -216,6 +239,8 @@ def column2vec_weighted_sum(column: pd.Series, model: SentenceTransformer, key:
     """
     function_string = "column2vec_weighted_sum"
     encoded_columns, weights = weighted_create_embed(column, model, key, function_string)
+    if weights is None:
+        return encoded_columns
     to_ret = 0
     for number, weight in zip(encoded_columns, weights):
         to_ret += number * weight