Python evaluator module fix (#863)

* Python evaluator module fix

Remove redundant and unnecessary sortings
Refactor get_top_k_items to return DataFrame with 'rank' column
    same as pyspark's

* Update test to catch corner case

Author: loomlike

reco_utils/evaluation/python_evaluation.py

@@ -363,7 +363,7 @@ def merge_ranking_true_pred(
 
     Returns:
         pd.DataFrame, pd.DataFrame, int:
-            DataFrame of recommendation hits
+            DataFrame of recommendation hits, sorted by `col_user` and `rank`
             DataFrmae of hit counts vs actual relevant items per user
             number of unique user ids
     """
@@ -390,9 +390,6 @@ def merge_ranking_true_pred(
         col_rating=col_prediction,
         k=top_k,
     )
-    df_hit["rank"] = df_hit.groupby(col_user)[col_prediction].rank(
-        method="first", ascending=False
-    )
     df_hit = pd.merge(df_hit, rating_true_common, on=[col_user, col_item])[
         [col_user, col_item, "rank"]
     ]
@@ -559,7 +556,7 @@ def ndcg_at_k(
     # relevance in this case is always 1
     df_dcg["dcg"] = 1 / np.log1p(df_dcg["rank"])
     # sum up discount gained to get discount cumulative gain
-    df_dcg = df_dcg.groupby(col_user, as_index=False).agg({"dcg": "sum"})
+    df_dcg = df_dcg.groupby(col_user, as_index=False, sort=False).agg({"dcg": "sum"})
     # calculate ideal discounted cumulative gain
     df_ndcg = pd.merge(df_dcg, df_hit_count, on=[col_user])
     df_ndcg["idcg"] = df_ndcg["actual"].apply(
@@ -625,8 +622,8 @@ def map_at_k(
         return 0.0
 
     # calculate reciprocal rank of items for each user and sum them up
-    df_hit_sorted = df_hit.sort_values([col_user, "rank"])
-    df_hit_sorted["rr"] = (df_hit.groupby(col_user).cumcount() + 1) / df_hit["rank"]
+    df_hit_sorted = df_hit.copy()
+    df_hit_sorted["rr"] = (df_hit_sorted.groupby(col_user).cumcount() + 1) / df_hit_sorted["rank"]
     df_hit_sorted = df_hit_sorted.groupby(col_user).agg({"rr": "sum"}).reset_index()
 
     df_merge = pd.merge(df_hit_sorted, df_hit_count, on=col_user)
@@ -651,14 +648,17 @@ def get_top_k_items(
         k (int): number of items for each user
 
     Returns:
-        pd.DataFrame: DataFrame of top k items for each user
+        pd.DataFrame: DataFrame of top k items for each user, sorted by `col_user` and `rank`
     """
-
-    return (
+    # Sort dataframe by col_user and (top k) col_rating
+    top_k_items = (
         dataframe.groupby(col_user, as_index=False)
         .apply(lambda x: x.nlargest(k, col_rating))
         .reset_index(drop=True)
     )
+    # Add ranks
+    top_k_items["rank"] = top_k_items.groupby(col_user, sort=False).cumcount() + 1
+    return top_k_items
 
 
 """Function name and function mapper.

tests/unit/test_python_evaluation.py

@@ -35,10 +35,8 @@
 def rating_true():
     return pd.DataFrame(
         {
-            DEFAULT_USER_COL: [1, 1, 1, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3],
+            DEFAULT_USER_COL: [1, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 1, 1,],
             DEFAULT_ITEM_COL: [
-                1,
-                2,
                 3,
                 1,
                 4,
@@ -55,8 +53,10 @@ def rating_true():
                 12,
                 13,
                 14,
+                1,
+                2,
             ],
-            DEFAULT_RATING_COL: [5, 4, 3, 5, 5, 3, 3, 1, 5, 5, 5, 4, 4, 3, 3, 3, 2, 1],
+            DEFAULT_RATING_COL: [3, 5, 5, 3, 3, 1, 5, 5, 5, 4, 4, 3, 3, 3, 2, 1, 5, 4,],
         }
     )
 
@@ -65,10 +65,8 @@ def rating_true():
 def rating_pred():
     return pd.DataFrame(
         {
-            DEFAULT_USER_COL: [1, 1, 1, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3],
+            DEFAULT_USER_COL: [1, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 1, 1,],
             DEFAULT_ITEM_COL: [
-                3,
-                10,
                 12,
                 10,
                 3,
@@ -85,10 +83,10 @@ def rating_pred():
                 2,
                 11,
                 14,
+                3,
+                10,
             ],
             DEFAULT_PREDICTION_COL: [
-                14,
-                13,
                 12,
                 14,
                 13,
@@ -105,8 +103,10 @@ def rating_pred():
                 7,
                 6,
                 5,
+                14,
+                13,
             ],
-            DEFAULT_RATING_COL: [5, 4, 3, 5, 5, 3, 3, 1, 5, 5, 5, 4, 4, 3, 3, 3, 2, 1],
+            DEFAULT_RATING_COL: [3, 5, 5, 3, 3, 1, 5, 5, 5, 4, 4, 3, 3, 3, 2, 1, 5, 4,],
         }
     )
 
@@ -115,11 +115,9 @@ def rating_pred():
 def rating_nohit():
     return pd.DataFrame(
         {
-            DEFAULT_USER_COL: [1, 1, 1, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3],
+            DEFAULT_USER_COL: [1, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 1, 1,],
             DEFAULT_ITEM_COL: [100] * 18,
             DEFAULT_PREDICTION_COL: [
-                14,
-                13,
                 12,
                 14,
                 13,
@@ -136,6 +134,8 @@ def rating_nohit():
                 7,
                 6,
                 5,
+                14,
+                13,
             ],
         }
     )