updated readme

README.md

@@ -25,11 +25,15 @@ For example:
 The following Dialogues were excluded:
  - 146 dialogues where a donation of 0 was made at the end of the task but a non-zero amount was pledged by the persuadee in the dialogue, per the following regular expression: `(?:\$(?:0\.)?[1-9]|[1-9][.0-9]*?(?: ?\$| dollars?| cents?))`
  
-Stats:
+Data Info:
  - **Training set**: 587 dialogues, using actual end-task donations as labels
  - **Validation set**: 141 dialogues, using manual donation intention labels from Persuasion For Good 'AnnSet'
  - **Test set**: 143 dialogues, using manual donation intention labels from Persuasion For Good 'AnnSet'
 
-**Test Macro F1**: 0.893
+Training Info:
+ - **Loss**: CrossEntropy with class weights: 1.5447 (class 0) and 0.7393 (class 1). These weights were derived from the training split.
+ - **Early Stopping**: The checkpoint with the highest validation macro f1 was selected. This occurred at step 35 (see training metrics for more detail).
 
-**Test Accuracy**: 0.902
+Testing Info:
+ - **Test Macro F1**: 0.893
+ - **Test Accuracy**: 0.902