fix: added balanced datasets

README.md

@@ -14,4 +14,11 @@ configs:
   - split: test
     path: "test.csv"
   default: true
+- config_name: sentiment
+  data_files:
+  - split: complete
+    path: "positive_negative.csv"
+  data_files:
+  - split: balanced
+    path: "positive_negative_balanced.csv"
 ---

convert.py

@@ -1,4 +1,5 @@
 from datetime import datetime
+from random import shuffle
 
 import pandas as pd
 
@@ -61,6 +62,20 @@ def run():
     save(test_rows, "test.csv")
     save(all_rows, "complete.csv")
 
+    # Extract only ratings 1, 2, 3, and 4 (negative reviews), and also
+    # ratings 7, 8, 9, and 10 (positive reviews)
+    negative_rows = [row for row in all_rows if row["rating"] <= 4]
+    positive_rows = [row for row in all_rows if row["rating"] >= 7]
+    positive_negative_rows = negative_rows + positive_rows
+    shuffle(positive_negative_rows)
+    save(positive_negative_rows, "positive_negative.csv")
+
+    # Balance the dataset
+    min_size = min(len(negative_rows), len(positive_rows))
+    balanced_rows = negative_rows[:min_size] + positive_rows[:min_size]
+    shuffle(balanced_rows)
+    save(balanced_rows, "positive_negative_balanced.csv")
+
 
 if __name__ == "__main__":
     run()

positive_negative.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ec6b3ee0b25857f5e6bb83e55b3da85e3304aeaf3796de72aef5cb8ba5549f3b
+size 97868615

positive_negative_balanced.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a7d3030fe9ea8d256276dd52cc7ecffc6985b748447f6c78690bde23cbfb410b
+size 53051421