From 4001a5d1571cb622315dd6c7ff6034ad42d6cd17 Mon Sep 17 00:00:00 2001
From: qianyun210603 <qianyun210603@hotmail.com>
Date: Sun, 13 Nov 2022 19:03:23 +0800
Subject: [PATCH] Bug fix for Rank and WMA operators (#1228)

* bug fix: 1) 100 should be used to scale down percentileofscore return to 0-1, not length of array; 2) for (linear) weighted MA(n), weight should be n, n-1, ..., 1 instead of n-1, ..., 0

* use native pandas fucntion for rank

* remove useless import

* require pandas 1.4+

* rank for py37+pandas 1.3.5 compatibility

* lint improvement

* lint black fix

* use hasattr instead of version to check whether rolling.rank is implemented
---
 qlib/data/ops.py | 24 +++++++++---------------
 1 file changed, 9 insertions(+), 15 deletions(-)

diff --git a/qlib/data/ops.py b/qlib/data/ops.py
index 1cbb1d2e6..fe2ebc9f6 100644
--- a/qlib/data/ops.py
+++ b/qlib/data/ops.py
@@ -34,8 +34,6 @@ np.seterr(invalid="ignore")
 
 
 #################### Element-Wise Operator ####################
-
-
 class ElemOperator(ExpressionOps):
     """Element-wise Operator
 
@@ -216,9 +214,7 @@ class Not(NpElemOperator):
 
     Parameters
     ----------
-    feature_left : Expression
-        feature instance
-    feature_right : Expression
+    feature : Expression
         feature instance
 
     Returns
@@ -241,8 +237,6 @@ class PairOperator(ExpressionOps):
         feature instance or numeric value
     feature_right : Expression
         feature instance or numeric value
-    func : str
-        operator function
 
     Returns
     ----------
@@ -1155,9 +1149,13 @@ class Rank(Rolling):
     def __init__(self, feature, N):
         super(Rank, self).__init__(feature, N, "rank")
 
+    # for compatiblity of python 3.7, which doesn't support pandas 1.4.0+ which implements Rolling.rank
     def _load_internal(self, instrument, start_index, end_index, *args):
         series = self.feature.load(instrument, start_index, end_index, *args)
-        # TODO: implement in Cython
+
+        rolling_or_expending = series.expanding(min_periods=1) if self.N == 0 else series.rolling(self.N, min_periods=1)
+        if hasattr(rolling_or_expending, "rank"):
+            return rolling_or_expending.rank(pct=True)
 
         def rank(x):
             if np.isnan(x[-1]):
@@ -1165,13 +1163,9 @@ class Rank(Rolling):
             x1 = x[~np.isnan(x)]
             if x1.shape[0] == 0:
                 return np.nan
-            return percentileofscore(x1, x1[-1]) / len(x1)
+            return percentileofscore(x1, x1[-1]) / 100
 
-        if self.N == 0:
-            series = series.expanding(min_periods=1).apply(rank, raw=True)
-        else:
-            series = series.rolling(self.N, min_periods=1).apply(rank, raw=True)
-        return series
+        return rolling_or_expending.apply(rank, raw=True)
 
 
 class Count(Rolling):
@@ -1341,7 +1335,7 @@ class WMA(Rolling):
         # TODO: implement in Cython
 
         def weighted_mean(x):
-            w = np.arange(len(x))
+            w = np.arange(len(x)) + 1
             w = w / w.sum()
             return np.nanmean(w * x)