From 2c1d15318912a6d8c124d5cc5c43007da15eeee5 Mon Sep 17 00:00:00 2001
From: Zhiyi Wu <zwu@exscientia.ai>
Date: Tue, 21 Mar 2023 22:17:43 +0000
Subject: [PATCH] Fix the dE method for AMBER (#300)

* fix #299
* Fix the dE method in u_nk2series to use the difference between two
   lambda columns instead of using the next lambda column or the previous
   column for the last window
* change tests (GROMACS results changed)
* Update CHANGES
---
 CHANGES                                    | 10 ++++++++++
 src/alchemlyb/preprocessing/subsampling.py | 10 ++++++++--
 src/alchemlyb/tests/test_preprocessing.py  |  8 ++++----
 3 files changed, 22 insertions(+), 6 deletions(-)

diff --git a/CHANGES b/CHANGES
index 7e55c754..fd23a12b 100644
--- a/CHANGES
+++ b/CHANGES
@@ -13,6 +13,16 @@ The rules for this file:
   * release numbers follow "Semantic Versioning" https://semver.org
 
 ------------------------------------------------------------------------------
+*/*/* xiki-tempula,
+
+  * unreleased
+
+Fixes
+  - Fix the dE method in u_nk2series to use the difference between two
+    lambda columns instead of using the next lambda column or the previous
+    column for the last window (issue #299, PR #300).
+
+
 12/12/2022 xiki-tempula, orbeckst
 
   * 2.0.0
diff --git a/src/alchemlyb/preprocessing/subsampling.py b/src/alchemlyb/preprocessing/subsampling.py
index a659ca09..ffe39704 100644
--- a/src/alchemlyb/preprocessing/subsampling.py
+++ b/src/alchemlyb/preprocessing/subsampling.py
@@ -153,6 +153,10 @@ def u_nk2series(df, method="dE"):
 
 
     .. versionadded:: 1.0.0
+    .. versionchanged:: 2.0.1
+       The `dE` method computes the difference between the current lambda
+       and the next lambda (previous lambda for the last window), instead
+       of using the next lambda or the previous lambda for the last window.
 
     """
 
@@ -196,11 +200,13 @@ def u_nk2series(df, method="dE"):
             # For the case of more than 1 lambda
             index = df.columns.values.tolist().index(key)
             # for the state that is not the last state, take the state+1
+        current_lambda = df.iloc[:, index]
         if index + 1 < len(df.columns):
-            series = df.iloc[:, index + 1]
+            new_lambda = df.iloc[:, index + 1]
             # for the state that is the last state, take the state-1
         else:
-            series = df.iloc[:, index - 1]
+            new_lambda = df.iloc[:, index - 1]
+        series = new_lambda - current_lambda
     else:
         raise ValueError("Decorrelation method {} not found.".format(method))
     return series
diff --git a/src/alchemlyb/tests/test_preprocessing.py b/src/alchemlyb/tests/test_preprocessing.py
index ea8231ab..766bc3df 100644
--- a/src/alchemlyb/tests/test_preprocessing.py
+++ b/src/alchemlyb/tests/test_preprocessing.py
@@ -422,7 +422,7 @@ def test_decorrelate_u_nk_burnin(u_nk):
                 remove_burnin=True,
             )
         )
-        == 2849
+        == 2848
     )
 
 
@@ -493,9 +493,9 @@ class TestU_nk2series:
     @pytest.mark.parametrize(
         "methodargs,reference",  # reference = sum
         [
-            ({}, 9207.80229000283),
+            ({}, 7988.667045),
             ({"method": "all"}, 85982.34668751864),
-            ({"method": "dE"}, 9207.80229000283),
+            ({"method": "dE"}, 7988.667045),
         ],
     )
     def test_u_nk2series(self, u_nk, methodargs, reference):
@@ -507,7 +507,7 @@ def test_u_nk2series(self, u_nk, methodargs, reference):
         "methodargs,reference",  # reference = sum
         [
             ({"method": "dhdl_all"}, 85982.34668751864),
-            ({"method": "dhdl"}, 9207.80229000283),
+            ({"method": "dhdl"}, 7988.667045),
         ],
     )
     def test_u_nk2series_deprecated(self, u_nk, methodargs, reference):