shuffle before partitioning

hwu66.py

@@ -161,6 +161,9 @@ class Hwu66(datasets.GeneratorBasedBuilder):
         # The `key` is here for legacy reason (tfds) and is not important in itself.
 
         df = pd.read_csv(filepath,sep=";")
+        df = df.sample(frac=1.0, random_state=42)
+
+
         df['label'] = df['scenario'] + '_' + df['intent']
         TEST_SIZE = 2000