Nol00 commited on
Commit
a881fbe
·
verified ·
1 Parent(s): 3cee0f3

Create leaderboard.py

Browse files
Files changed (1) hide show
  1. crs_arena/leaderboard.py +40 -0
crs_arena/leaderboard.py ADDED
@@ -0,0 +1,40 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """Static leaderboard page."""
2
+
3
+ import pandas as pd
4
+ import streamlit as st
5
+
6
+ # Add button to go back to the CRS Arena
7
+ if st.button("Back to :gun: CRS Arena"):
8
+ st.switch_page("arena.py")
9
+
10
+ st.title(":trophy: CRS Arena Leaderboard")
11
+ st.write(
12
+ "This leaderboard ranks conversational recommender systems present in the CRS Arena based on different metrics, including Elo rating and user satisfaction."
13
+ )
14
+
15
+ st.write("*Last update: 2024-10-04*")
16
+
17
+ df_leaderboard = pd.DataFrame(
18
+ {
19
+ "CRS": [
20
+ "BARCOR_OpenDialKG",
21
+ "BARCOR_ReDial",
22
+ "CRB-CRS_ReDial",
23
+ "ChatGPT_OpenDialKG",
24
+ "ChatGPT_ReDial",
25
+ "KBRD_OpenDialKG",
26
+ "KBRD_ReDial",
27
+ "UniCRS_OpenDialKG",
28
+ "UniCRS_ReDial",
29
+ ],
30
+ "Elo rating*": [968, 1052, 930, 1056, 1036, 966, 1027, 974, 991],
31
+ "% Satisfaction": [17.2, 30.4, 5.4, 50, 58.6, 2.6, 8.6, 9.5, 18.2],
32
+ }
33
+ )
34
+
35
+ # Sort DataFrame by Elo rating
36
+ df_leaderboard = df_leaderboard.sort_values(by="Elo rating*", ascending=False)
37
+
38
+ st.dataframe(df_leaderboard, hide_index=True, use_container_width=True)
39
+
40
+ st.write("*Initial rating is set to 1000 and K-factor is 16")