From 71ee535e60e005270fd66cc594ea0290c3157faf Mon Sep 17 00:00:00 2001
From: Pieter Van Trappen <vtpieter@gmail.com>
Date: Wed, 6 May 2020 15:40:10 +0200
Subject: [PATCH] changed fillna for first lines from zeroes to backfill, as
done last year; probably best for the model no to have large unreal jumps in
data values
---
src/preprocessing/preprocess_features.py | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/src/preprocessing/preprocess_features.py b/src/preprocessing/preprocess_features.py
index cba33a33..ff1d01c1 100755
--- a/src/preprocessing/preprocess_features.py
+++ b/src/preprocessing/preprocess_features.py
@@ -46,8 +46,8 @@ if __name__ == "__main__":
missing_values.append(col)
print('Forward-filling columns {}'.format(missing_values))
features = features.fillna(method='ffill')
- # additional fill with zeros as a dirty fix for first rows missing values
- features.fillna(value=0, inplace=True)
+ # additional backfill as a dirty fix for remaining first rows missing values
+ features = features.fillna(method='bfill')
print('Forward-filled {} columns'.format(len(missing_values)))
print('\nNot filled columns are {}'.format(
[c for c in features.columns.values.tolist() if not c in missing_values]))
--
GitLab