Spaces:
Running
Running
Update leaderboard display
Browse files
app.py
ADDED
@@ -0,0 +1,803 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import json
|
3 |
+
import pandas as pd
|
4 |
+
import numpy as np
|
5 |
+
import plotly.express as px
|
6 |
+
import plotly.graph_objects as go
|
7 |
+
from plotly.subplots import make_subplots
|
8 |
+
import os
|
9 |
+
import traceback
|
10 |
+
from datetime import datetime
|
11 |
+
from packaging import version
|
12 |
+
|
13 |
+
# Color scheme for charts
|
14 |
+
COLORS = px.colors.qualitative.Plotly
|
15 |
+
|
16 |
+
# Line colors for radar charts
|
17 |
+
line_colors = [
|
18 |
+
"#EE4266",
|
19 |
+
"#00a6ed",
|
20 |
+
"#ECA72C",
|
21 |
+
"#B42318",
|
22 |
+
"#3CBBB1",
|
23 |
+
]
|
24 |
+
|
25 |
+
# Fill colors for radar charts
|
26 |
+
fill_colors = [
|
27 |
+
"rgba(238,66,102,0.05)",
|
28 |
+
"rgba(0,166,237,0.05)",
|
29 |
+
"rgba(236,167,44,0.05)",
|
30 |
+
"rgba(180,35,24,0.05)",
|
31 |
+
"rgba(60,187,177,0.05)",
|
32 |
+
]
|
33 |
+
|
34 |
+
# Define the question categories
|
35 |
+
QUESTION_CATEGORIES = ["simple", "set", "mh", "cond", "comp"]
|
36 |
+
METRIC_TYPES = ["retrieval", "generation"]
|
37 |
+
|
38 |
+
def load_results():
|
39 |
+
"""Load results from the results.json file."""
|
40 |
+
try:
|
41 |
+
# Get the directory of the current script
|
42 |
+
script_dir = os.path.dirname(os.path.abspath(__file__))
|
43 |
+
# Build the path to results.json
|
44 |
+
results_path = os.path.join(script_dir, 'results.json')
|
45 |
+
|
46 |
+
print(f"Loading results from: {results_path}")
|
47 |
+
|
48 |
+
with open(results_path, 'r', encoding='utf-8') as f:
|
49 |
+
results = json.load(f)
|
50 |
+
print(f"Successfully loaded results with {len(results.get('items', {}))} version(s)")
|
51 |
+
return results
|
52 |
+
except FileNotFoundError:
|
53 |
+
# Return empty structure if file doesn't exist
|
54 |
+
print(f"Results file not found, creating empty structure")
|
55 |
+
return {"items": {}, "last_version": "1.0", "n_questions": "0"}
|
56 |
+
except Exception as e:
|
57 |
+
print(f"Error loading results: {e}")
|
58 |
+
print(traceback.format_exc())
|
59 |
+
return {"items": {}, "last_version": "1.0", "n_questions": "0"}
|
60 |
+
|
61 |
+
def filter_and_process_results(results, n_versions, only_actual_versions):
|
62 |
+
"""Filter results by version and process them for display."""
|
63 |
+
if not results or "items" not in results:
|
64 |
+
return pd.DataFrame(), [], [], []
|
65 |
+
|
66 |
+
all_items = results["items"]
|
67 |
+
last_version_str = results.get("last_version", "1.0")
|
68 |
+
last_version = version.parse(last_version_str)
|
69 |
+
|
70 |
+
print(f"Last version: {last_version_str}")
|
71 |
+
|
72 |
+
# Group items by model_name
|
73 |
+
model_groups = {}
|
74 |
+
|
75 |
+
for version_str, version_items in all_items.items():
|
76 |
+
version_obj = version.parse(version_str)
|
77 |
+
for item_id, item in version_items.items():
|
78 |
+
model_name = item.get("model_name", "Unknown")
|
79 |
+
|
80 |
+
if model_name not in model_groups:
|
81 |
+
model_groups[model_name] = []
|
82 |
+
|
83 |
+
# Add version info to the item (both as string and as parsed version object for comparison)
|
84 |
+
item["version_str"] = version_str
|
85 |
+
item["version_obj"] = version_obj
|
86 |
+
model_groups[model_name].append(item)
|
87 |
+
|
88 |
+
rows = []
|
89 |
+
for model_name, items in model_groups.items():
|
90 |
+
# Sort items by version (newest first)
|
91 |
+
items.sort(key=lambda x: x["version_obj"], reverse=True)
|
92 |
+
|
93 |
+
# Filter versions based on selection
|
94 |
+
filtered_items = []
|
95 |
+
|
96 |
+
if only_actual_versions:
|
97 |
+
# Get the n most recent actual dataset versions
|
98 |
+
all_versions = sorted([version.parse(v_str) for v_str in all_items.keys()], reverse=True)
|
99 |
+
# Take at most n_versions
|
100 |
+
versions_to_consider = all_versions[:n_versions] if all_versions else []
|
101 |
+
|
102 |
+
# Filter items that match those versions
|
103 |
+
filtered_items = [item for item in items if any(item["version_obj"] == v for v in versions_to_consider)]
|
104 |
+
else:
|
105 |
+
# Consider n_versions most recent items for this model
|
106 |
+
filtered_items = items[:n_versions]
|
107 |
+
|
108 |
+
if not filtered_items:
|
109 |
+
continue
|
110 |
+
|
111 |
+
config = filtered_items[0]["config"] # Use config from most recent version
|
112 |
+
|
113 |
+
# Create row with basic info
|
114 |
+
row = {
|
115 |
+
'Model': model_name,
|
116 |
+
'Embeddings': config.get('embedding_model', 'N/A'),
|
117 |
+
'Retriever': config.get('retriever_type', 'N/A'),
|
118 |
+
'Top-K': config.get('retrieval_config', {}).get('top_k', 'N/A'),
|
119 |
+
'Versions': ", ".join([item["version_str"] for item in filtered_items]),
|
120 |
+
'Last Updated': filtered_items[0].get("timestamp", "")
|
121 |
+
}
|
122 |
+
|
123 |
+
# Format timestamp if available
|
124 |
+
if row['Last Updated']:
|
125 |
+
try:
|
126 |
+
dt = datetime.fromisoformat(row['Last Updated'].replace('Z', '+00:00'))
|
127 |
+
row['Last Updated'] = dt.strftime("%Y-%m-%d")
|
128 |
+
except:
|
129 |
+
pass
|
130 |
+
|
131 |
+
# Process metrics based on categories
|
132 |
+
category_metrics = {
|
133 |
+
category: {
|
134 |
+
metric_type: {
|
135 |
+
"avg": 0.0,
|
136 |
+
"count": 0
|
137 |
+
} for metric_type in METRIC_TYPES
|
138 |
+
} for category in QUESTION_CATEGORIES
|
139 |
+
}
|
140 |
+
|
141 |
+
# Collect metrics by category
|
142 |
+
for item in filtered_items:
|
143 |
+
metrics = item.get("metrics", {})
|
144 |
+
for category in QUESTION_CATEGORIES:
|
145 |
+
if category in metrics:
|
146 |
+
for metric_type in METRIC_TYPES:
|
147 |
+
if metric_type in metrics[category]:
|
148 |
+
metric_values = metrics[category][metric_type]
|
149 |
+
avg_value = sum(metric_values.values()) / len(metric_values)
|
150 |
+
|
151 |
+
# Add to the running sum for this category and metric type
|
152 |
+
category_metrics[category][metric_type]["avg"] += avg_value
|
153 |
+
category_metrics[category][metric_type]["count"] += 1
|
154 |
+
|
155 |
+
# Calculate averages and add to row
|
156 |
+
for category in QUESTION_CATEGORIES:
|
157 |
+
for metric_type in METRIC_TYPES:
|
158 |
+
metric_data = category_metrics[category][metric_type]
|
159 |
+
if metric_data["count"] > 0:
|
160 |
+
avg_value = metric_data["avg"] / metric_data["count"]
|
161 |
+
# Add to row with appropriate column name
|
162 |
+
col_name = f"{category}_{metric_type}"
|
163 |
+
row[col_name] = round(avg_value, 4)
|
164 |
+
|
165 |
+
# Calculate overall averages for each metric type
|
166 |
+
for metric_type in METRIC_TYPES:
|
167 |
+
total_sum = 0
|
168 |
+
total_count = 0
|
169 |
+
|
170 |
+
for category in QUESTION_CATEGORIES:
|
171 |
+
metric_data = category_metrics[category][metric_type]
|
172 |
+
if metric_data["count"] > 0:
|
173 |
+
total_sum += metric_data["avg"]
|
174 |
+
total_count += metric_data["count"]
|
175 |
+
|
176 |
+
if total_count > 0:
|
177 |
+
row[f"{metric_type}_avg"] = round(total_sum / total_count, 4)
|
178 |
+
|
179 |
+
rows.append(row)
|
180 |
+
|
181 |
+
# Create DataFrame
|
182 |
+
df = pd.DataFrame(rows)
|
183 |
+
|
184 |
+
# Get lists of metrics for each category
|
185 |
+
category_metrics = []
|
186 |
+
for category in QUESTION_CATEGORIES:
|
187 |
+
metrics = []
|
188 |
+
for metric_type in METRIC_TYPES:
|
189 |
+
col_name = f"{category}_{metric_type}"
|
190 |
+
if col_name in df.columns:
|
191 |
+
metrics.append(col_name)
|
192 |
+
if metrics:
|
193 |
+
category_metrics.append((category, metrics))
|
194 |
+
|
195 |
+
# Define retrieval and generation columns for radar charts
|
196 |
+
retrieval_metrics = [f"{category}_retrieval" for category in QUESTION_CATEGORIES if f"{category}_retrieval" in df.columns]
|
197 |
+
generation_metrics = [f"{category}_generation" for category in QUESTION_CATEGORIES if f"{category}_generation" in df.columns]
|
198 |
+
|
199 |
+
return df, retrieval_metrics, generation_metrics, category_metrics
|
200 |
+
|
201 |
+
def create_radar_chart(df, selected_models, metrics, title):
|
202 |
+
"""Create a radar chart for the selected models and metrics."""
|
203 |
+
if not metrics or len(selected_models) == 0:
|
204 |
+
# Return empty figure if no metrics or models selected
|
205 |
+
fig = go.Figure()
|
206 |
+
fig.update_layout(
|
207 |
+
title=title,
|
208 |
+
title_font_size=16,
|
209 |
+
height=400,
|
210 |
+
width=500,
|
211 |
+
margin=dict(l=30, r=30, t=50, b=30)
|
212 |
+
)
|
213 |
+
return fig
|
214 |
+
|
215 |
+
# Filter dataframe for selected models
|
216 |
+
filtered_df = df[df['Model'].isin(selected_models)]
|
217 |
+
|
218 |
+
if filtered_df.empty:
|
219 |
+
# Return empty figure if no data
|
220 |
+
fig = go.Figure()
|
221 |
+
fig.update_layout(
|
222 |
+
title=title,
|
223 |
+
title_font_size=16,
|
224 |
+
height=400,
|
225 |
+
width=500,
|
226 |
+
margin=dict(l=30, r=30, t=50, b=30)
|
227 |
+
)
|
228 |
+
return fig
|
229 |
+
|
230 |
+
# Limit to top 5 models for better visualization (similar to inspiration file)
|
231 |
+
if len(filtered_df) > 5:
|
232 |
+
filtered_df = filtered_df.head(5)
|
233 |
+
|
234 |
+
# Prepare data for radar chart
|
235 |
+
categories = [m.split('_', 1)[0] for m in metrics] # Get category name (simple, set, etc.)
|
236 |
+
|
237 |
+
fig = go.Figure()
|
238 |
+
|
239 |
+
# Process in reverse order to match inspiration file
|
240 |
+
for i, (_, row) in enumerate(filtered_df.iterrows()):
|
241 |
+
values = [row[m] for m in metrics]
|
242 |
+
# Close the loop for radar chart
|
243 |
+
values.append(values[0])
|
244 |
+
categories_loop = categories + [categories[0]]
|
245 |
+
|
246 |
+
fig.add_trace(go.Scatterpolar(
|
247 |
+
name=row['Model'],
|
248 |
+
r=values,
|
249 |
+
theta=categories_loop,
|
250 |
+
showlegend=True,
|
251 |
+
mode="lines",
|
252 |
+
line=dict(width=2, color=line_colors[i % len(line_colors)]),
|
253 |
+
fill="toself",
|
254 |
+
fillcolor=fill_colors[i % len(fill_colors)]
|
255 |
+
))
|
256 |
+
|
257 |
+
fig.update_layout(
|
258 |
+
font=dict(size=13, color="black"),
|
259 |
+
template="plotly_white",
|
260 |
+
polar=dict(
|
261 |
+
radialaxis=dict(
|
262 |
+
visible=True,
|
263 |
+
gridcolor="black",
|
264 |
+
linecolor="rgba(0,0,0,0)",
|
265 |
+
gridwidth=1,
|
266 |
+
showticklabels=False,
|
267 |
+
ticks="",
|
268 |
+
range=[0, 1] # Ensure consistent range for scores
|
269 |
+
),
|
270 |
+
angularaxis=dict(
|
271 |
+
gridcolor="black",
|
272 |
+
gridwidth=1.5,
|
273 |
+
linecolor="rgba(0,0,0,0)"
|
274 |
+
),
|
275 |
+
),
|
276 |
+
legend=dict(
|
277 |
+
orientation="h",
|
278 |
+
yanchor="bottom",
|
279 |
+
y=-0.35,
|
280 |
+
xanchor="center",
|
281 |
+
x=0.4,
|
282 |
+
itemwidth=30,
|
283 |
+
font=dict(size=13),
|
284 |
+
entrywidth=0.6,
|
285 |
+
entrywidthmode="fraction",
|
286 |
+
),
|
287 |
+
margin=dict(l=0, r=16, t=30, b=30),
|
288 |
+
autosize=True,
|
289 |
+
)
|
290 |
+
|
291 |
+
return fig
|
292 |
+
|
293 |
+
def create_summary_df(df, retrieval_metrics, generation_metrics):
|
294 |
+
"""Create a summary dataframe with averaged metrics for display."""
|
295 |
+
if df.empty:
|
296 |
+
return pd.DataFrame()
|
297 |
+
|
298 |
+
summary_df = df.copy()
|
299 |
+
|
300 |
+
# Add retrieval average
|
301 |
+
if retrieval_metrics:
|
302 |
+
retrieval_avg = summary_df[retrieval_metrics].mean(axis=1).round(4)
|
303 |
+
summary_df['Retrieval (avg)'] = retrieval_avg
|
304 |
+
|
305 |
+
# Add generation average
|
306 |
+
if generation_metrics:
|
307 |
+
generation_avg = summary_df[generation_metrics].mean(axis=1).round(4)
|
308 |
+
summary_df['Generation (avg)'] = generation_avg
|
309 |
+
|
310 |
+
# Add total score if both averages exist
|
311 |
+
if 'Retrieval (avg)' in summary_df.columns and 'Generation (avg)' in summary_df.columns:
|
312 |
+
summary_df['Total Score'] = summary_df['Retrieval (avg)'] + summary_df['Generation (avg)']
|
313 |
+
summary_df = summary_df.sort_values('Total Score', ascending=False)
|
314 |
+
|
315 |
+
# Select columns for display
|
316 |
+
summary_cols = ['Model', 'Embeddings', 'Retriever', 'Top-K']
|
317 |
+
if 'Retrieval (avg)' in summary_df.columns:
|
318 |
+
summary_cols.append('Retrieval (avg)')
|
319 |
+
if 'Generation (avg)' in summary_df.columns:
|
320 |
+
summary_cols.append('Generation (avg)')
|
321 |
+
if 'Total Score' in summary_df.columns:
|
322 |
+
summary_cols.append('Total Score')
|
323 |
+
if 'Versions' in summary_df.columns:
|
324 |
+
summary_cols.append('Versions')
|
325 |
+
if 'Last Updated' in summary_df.columns:
|
326 |
+
summary_cols.append('Last Updated')
|
327 |
+
|
328 |
+
return summary_df[summary_cols]
|
329 |
+
|
330 |
+
def create_category_df(df, category, retrieval_col, generation_col):
|
331 |
+
"""Create a dataframe for a specific category with detailed metrics."""
|
332 |
+
if df.empty or retrieval_col not in df.columns or generation_col not in df.columns:
|
333 |
+
return pd.DataFrame()
|
334 |
+
|
335 |
+
category_df = df.copy()
|
336 |
+
|
337 |
+
# Calculate total score for this category
|
338 |
+
category_df[f'{category} Score'] = category_df[retrieval_col] + category_df[generation_col]
|
339 |
+
|
340 |
+
# Sort by total score
|
341 |
+
category_df = category_df.sort_values(f'{category} Score', ascending=False)
|
342 |
+
|
343 |
+
# Select columns for display
|
344 |
+
category_cols = ['Model', 'Embeddings', 'Retriever', retrieval_col, generation_col, f'{category} Score']
|
345 |
+
|
346 |
+
# Rename columns for display
|
347 |
+
category_df = category_df[category_cols].rename(columns={
|
348 |
+
retrieval_col: 'Retrieval',
|
349 |
+
generation_col: 'Generation'
|
350 |
+
})
|
351 |
+
|
352 |
+
return category_df
|
353 |
+
|
354 |
+
# Load initial data
|
355 |
+
results = load_results()
|
356 |
+
last_version = results.get("last_version", "1.0")
|
357 |
+
n_questions = results.get("n_questions", "100")
|
358 |
+
date_title = results.get("date_title", "---")
|
359 |
+
|
360 |
+
# Initial data processing
|
361 |
+
df, retrieval_metrics, generation_metrics, category_metrics = filter_and_process_results(
|
362 |
+
results, n_versions=1, only_actual_versions=True
|
363 |
+
)
|
364 |
+
|
365 |
+
# Pre-generate charts for initial display
|
366 |
+
default_models = df['Model'].head(5).tolist() if not df.empty else []
|
367 |
+
initial_gen_chart = create_radar_chart(df, default_models, generation_metrics, "Performance on Generation Tasks")
|
368 |
+
initial_ret_chart = create_radar_chart(df, default_models, retrieval_metrics, "Performance on Retrieval Tasks")
|
369 |
+
|
370 |
+
# Create summary dataframe
|
371 |
+
summary_df = create_summary_df(df, retrieval_metrics, generation_metrics)
|
372 |
+
|
373 |
+
with gr.Blocks(css="""
|
374 |
+
.title-container {
|
375 |
+
text-align: center;
|
376 |
+
margin-bottom: 10px;
|
377 |
+
}
|
378 |
+
.description-text {
|
379 |
+
text-align: left;
|
380 |
+
padding: 10px;
|
381 |
+
margin-bottom: 0px;
|
382 |
+
}
|
383 |
+
.version-info {
|
384 |
+
text-align: center;
|
385 |
+
padding: 10px;
|
386 |
+
background-color: #f0f0f0;
|
387 |
+
border-radius: 8px;
|
388 |
+
margin-bottom: 15px;
|
389 |
+
}
|
390 |
+
.version-selector {
|
391 |
+
padding: 15px;
|
392 |
+
border: 1px solid #ddd;
|
393 |
+
border-radius: 8px;
|
394 |
+
margin-bottom: 20px;
|
395 |
+
background-color: #f9f9f9;
|
396 |
+
height: 100%;
|
397 |
+
}
|
398 |
+
.citation-block {
|
399 |
+
padding: 15px;
|
400 |
+
border: 1px solid #ddd;
|
401 |
+
border-radius: 8px;
|
402 |
+
margin-bottom: 20px;
|
403 |
+
background-color: #f9f9f9;
|
404 |
+
font-family: monospace;
|
405 |
+
font-size: 14px;
|
406 |
+
overflow-x: auto;
|
407 |
+
height: 100%;
|
408 |
+
}
|
409 |
+
.flex-row-container {
|
410 |
+
display: flex;
|
411 |
+
justify-content: space-between;
|
412 |
+
gap: 20px;
|
413 |
+
width: 100%;
|
414 |
+
}
|
415 |
+
.charts-container {
|
416 |
+
display: flex;
|
417 |
+
gap: 20px;
|
418 |
+
margin-bottom: 20px;
|
419 |
+
}
|
420 |
+
.chart-box {
|
421 |
+
flex: 1;
|
422 |
+
border: 1px solid #eee;
|
423 |
+
border-radius: 8px;
|
424 |
+
padding: 10px;
|
425 |
+
background-color: white;
|
426 |
+
min-height: 550px; /* Increased height to accommodate legend at bottom */
|
427 |
+
}
|
428 |
+
.metrics-table {
|
429 |
+
border: 1px solid #eee;
|
430 |
+
border-radius: 8px;
|
431 |
+
padding: 15px;
|
432 |
+
background-color: white;
|
433 |
+
}
|
434 |
+
.info-text {
|
435 |
+
font-size: 0.9em;
|
436 |
+
font-style: italic;
|
437 |
+
color: #666;
|
438 |
+
margin-top: 5px;
|
439 |
+
}
|
440 |
+
footer {
|
441 |
+
text-align: center;
|
442 |
+
margin-top: 30px;
|
443 |
+
font-size: 0.9em;
|
444 |
+
color: #666;
|
445 |
+
}
|
446 |
+
/* Style for selected rows */
|
447 |
+
table tbody tr.selected {
|
448 |
+
background-color: rgba(25, 118, 210, 0.1) !important;
|
449 |
+
border-left: 3px solid #1976d2;
|
450 |
+
}
|
451 |
+
/* Add this class via JavaScript */
|
452 |
+
.gr-table tbody tr.selected td:first-child {
|
453 |
+
font-weight: bold;
|
454 |
+
color: #1976d2;
|
455 |
+
}
|
456 |
+
.category-tab {
|
457 |
+
padding: 10px;
|
458 |
+
}
|
459 |
+
.chart-title {
|
460 |
+
font-size: 1.2em;
|
461 |
+
font-weight: bold;
|
462 |
+
margin-bottom: 10px;
|
463 |
+
text-align: center;
|
464 |
+
}
|
465 |
+
.clear-charts-button {
|
466 |
+
display: flex;
|
467 |
+
justify-content: center;
|
468 |
+
margin-top: 10px;
|
469 |
+
margin-bottom: 20px;
|
470 |
+
}
|
471 |
+
""") as demo:
|
472 |
+
# Title
|
473 |
+
with gr.Row(elem_classes=["title-container"]):
|
474 |
+
gr.Markdown("# 🐙 Dynamic RAG Benchmark")
|
475 |
+
|
476 |
+
# Version info
|
477 |
+
with gr.Row(elem_classes=["description-text"]):
|
478 |
+
gr.Markdown(f"На этом лидерборде можно сравнить RAG системы в разрезе генеративных и поисковых метрик моделей по вопросам разного типа (простые вопросы, сравнения, multi-hop, условные и др.). <li>Вопросы автоматичеки генерируются на основе новостных источников.</li><li>Обновление датасета с вопросами происходит регулярно, при этом пересчитываются все метрики для открытых моделей.</li><li>Для пользовательских сабмитов учитываются последние посчитанные для них метрики.</li><li>Чтобы посчитать ранее отправленную конфигурацию на последней версии данных, используйте submit_id, полученный при первой отправке через клиент (см. инструкцию ниже).</li>")
|
479 |
+
|
480 |
+
# Version info
|
481 |
+
with gr.Row(elem_classes=["version-info"]):
|
482 |
+
gr.Markdown(f"## Версия {last_version} → {n_questions} вопросов, сгенерированных по новостным источникам → {date_title}")
|
483 |
+
|
484 |
+
# Radar Charts
|
485 |
+
with gr.Row(elem_classes=["charts-container"]):
|
486 |
+
with gr.Column(elem_classes=["chart-box"]):
|
487 |
+
gr.Markdown("### Генеративные метрики", elem_classes=["chart-title"])
|
488 |
+
generation_chart = gr.Plot(value=initial_gen_chart)
|
489 |
+
|
490 |
+
with gr.Column(elem_classes=["chart-box"]):
|
491 |
+
gr.Markdown("### Метрики поиска", elem_classes=["chart-title"])
|
492 |
+
retrieval_chart = gr.Plot(value=initial_ret_chart)
|
493 |
+
|
494 |
+
# Clear Charts Button
|
495 |
+
with gr.Row(elem_classes=["clear-charts-button"]):
|
496 |
+
clear_charts_btn = gr.Button("Очистить графики", variant="secondary")
|
497 |
+
|
498 |
+
# Metrics table with tabs
|
499 |
+
with gr.Tabs(elem_classes=["metrics-table"]) as metrics_tabs:
|
500 |
+
with gr.TabItem("Общая таблица"):
|
501 |
+
selected_models = gr.State(default_models)
|
502 |
+
|
503 |
+
# If dataframe is empty, show a message
|
504 |
+
if df.empty:
|
505 |
+
gr.Markdown("No data available. Please submit some results.")
|
506 |
+
metrics_table = gr.DataFrame()
|
507 |
+
else:
|
508 |
+
metrics_table = gr.DataFrame(
|
509 |
+
value=summary_df,
|
510 |
+
headers=summary_df.columns.tolist(),
|
511 |
+
datatype=["str"] * len(summary_df.columns),
|
512 |
+
row_count=(min(10, len(summary_df)) if not summary_df.empty else 0),
|
513 |
+
col_count=(len(summary_df.columns) if not summary_df.empty else 0),
|
514 |
+
interactive=False,
|
515 |
+
wrap=True
|
516 |
+
)
|
517 |
+
|
518 |
+
with gr.TabItem("По типам вопросов"):
|
519 |
+
# Create tabs for each category
|
520 |
+
category_tabs = gr.Tabs()
|
521 |
+
category_tables = {}
|
522 |
+
|
523 |
+
# Dictionary to map category codes to display names
|
524 |
+
category_display_names = {
|
525 |
+
"simple": "Simple Questions",
|
526 |
+
"set": "Set-based",
|
527 |
+
"mh": "Multi-hop",
|
528 |
+
"cond": "Conditional",
|
529 |
+
"comp": "Comparison"
|
530 |
+
}
|
531 |
+
|
532 |
+
with category_tabs:
|
533 |
+
for category, _ in category_metrics:
|
534 |
+
if f"{category}_retrieval" in df.columns and f"{category}_generation" in df.columns:
|
535 |
+
with gr.TabItem(category_display_names.get(category, category.capitalize()), elem_classes=["category-tab"]):
|
536 |
+
# Create dataframe for this category
|
537 |
+
category_df = create_category_df(df, category, f"{category}_retrieval", f"{category}_generation")
|
538 |
+
|
539 |
+
if category_df.empty:
|
540 |
+
gr.Markdown(f"No data available for {category_display_names.get(category, category)} category.")
|
541 |
+
category_tables[category] = gr.DataFrame()
|
542 |
+
else:
|
543 |
+
gr.Markdown(f"#### Performance on {category_display_names.get(category, category)}")
|
544 |
+
category_tables[category] = gr.DataFrame(
|
545 |
+
value=category_df,
|
546 |
+
headers=category_df.columns.tolist(),
|
547 |
+
datatype=["str"] * len(category_df.columns),
|
548 |
+
row_count=(min(10, len(category_df)) if not category_df.empty else 0),
|
549 |
+
col_count=(len(category_df.columns) if not category_df.empty else 0),
|
550 |
+
interactive=False,
|
551 |
+
wrap=True
|
552 |
+
)
|
553 |
+
|
554 |
+
# Version selector and Citation block in a flex container
|
555 |
+
with gr.Row():
|
556 |
+
# Citation block (left side)
|
557 |
+
with gr.Column(scale=1, elem_classes=["citation-block"]):
|
558 |
+
gr.Markdown("### Цитирование")
|
559 |
+
gr.Markdown("""
|
560 |
+
```
|
561 |
+
@article{dynamic-rag-benchmark,
|
562 |
+
title={Dynamic RAG Benchmark},
|
563 |
+
author={RAG Benchmark Team},
|
564 |
+
journal={arXiv preprint},
|
565 |
+
year={2024},
|
566 |
+
url={https://github.com/rag-benchmark}
|
567 |
+
}
|
568 |
+
```
|
569 |
+
|
570 |
+
Шаблон для цитирования нашего бенча.
|
571 |
+
""")
|
572 |
+
|
573 |
+
# Version selector (right side)
|
574 |
+
with gr.Column(scale=1, elem_classes=["version-selector"]):
|
575 |
+
gr.Markdown("### Выбор версий")
|
576 |
+
with gr.Column():
|
577 |
+
with gr.Row():
|
578 |
+
with gr.Column(scale=3):
|
579 |
+
only_actual_versions = gr.Checkbox(
|
580 |
+
label="Только актуальные версии",
|
581 |
+
value=True,
|
582 |
+
info="Считать, начиная с актуальной версии датасета"
|
583 |
+
)
|
584 |
+
with gr.Column(scale=5):
|
585 |
+
n_versions_slider = gr.Slider(
|
586 |
+
minimum=1,
|
587 |
+
maximum=5,
|
588 |
+
value=1,
|
589 |
+
step=1,
|
590 |
+
label="Взять n последних версий",
|
591 |
+
info="Количество версий для подсчета метрик"
|
592 |
+
)
|
593 |
+
with gr.Row():
|
594 |
+
filter_btn = gr.Button("Применить фильтр", variant="primary")
|
595 |
+
|
596 |
+
gr.Markdown(
|
597 |
+
"Кликайте на модели в таблице, чтобы добавить их в графики",
|
598 |
+
elem_classes=["info-text"]
|
599 |
+
)
|
600 |
+
|
601 |
+
# Footer
|
602 |
+
with gr.Row():
|
603 |
+
gr.Markdown("""
|
604 |
+
<footer>Dynamic RAG Benchmark Leaderboard</footer>
|
605 |
+
""")
|
606 |
+
|
607 |
+
# Handle row selection for radar charts
|
608 |
+
def update_charts(evt: gr.SelectData, selected_models):
|
609 |
+
try:
|
610 |
+
# Get current data with the latest filters
|
611 |
+
current_df, current_ret_metrics, current_gen_metrics, _ = filter_and_process_results(
|
612 |
+
results, n_versions=n_versions_slider.value, only_actual_versions=only_actual_versions.value
|
613 |
+
)
|
614 |
+
|
615 |
+
# Debug info
|
616 |
+
print(f"Selection event: {evt}, type: {type(evt)}")
|
617 |
+
|
618 |
+
selected_model = None
|
619 |
+
|
620 |
+
# Extract the selected model based on the row index
|
621 |
+
try:
|
622 |
+
# Get the table component that was clicked
|
623 |
+
component = evt.target
|
624 |
+
|
625 |
+
# Get the row index
|
626 |
+
row_idx = evt.index[0] if isinstance(evt.index, list) else evt.index
|
627 |
+
print(f"Row index: {row_idx}")
|
628 |
+
|
629 |
+
# Determine what type of data we're dealing with and extract model name
|
630 |
+
# First check if it's a summary table
|
631 |
+
if component is metrics_table:
|
632 |
+
# Summary table was clicked
|
633 |
+
if isinstance(summary_df, pd.DataFrame) and 0 <= row_idx < len(summary_df):
|
634 |
+
selected_model = summary_df.iloc[row_idx]['Model']
|
635 |
+
print(f"Selected from summary table: {selected_model}")
|
636 |
+
else:
|
637 |
+
# Check if it's a category table
|
638 |
+
for category, table in category_tables.items():
|
639 |
+
if component is table:
|
640 |
+
# Get the category dataframe
|
641 |
+
category_df = create_category_df(
|
642 |
+
current_df,
|
643 |
+
category,
|
644 |
+
f"{category}_retrieval",
|
645 |
+
f"{category}_generation"
|
646 |
+
)
|
647 |
+
|
648 |
+
if isinstance(category_df, pd.DataFrame) and 0 <= row_idx < len(category_df):
|
649 |
+
selected_model = category_df.iloc[row_idx]['Model']
|
650 |
+
print(f"Selected from {category} table: {selected_model}")
|
651 |
+
break
|
652 |
+
|
653 |
+
# If we still couldn't identify the model, try to get it from the raw data
|
654 |
+
if selected_model is None and hasattr(component, "value"):
|
655 |
+
table_value = component.value
|
656 |
+
if isinstance(table_value, pd.DataFrame) and 0 <= row_idx < len(table_value):
|
657 |
+
selected_model = table_value.iloc[row_idx]['Model']
|
658 |
+
elif isinstance(table_value, list) and 0 <= row_idx < len(table_value):
|
659 |
+
selected_model = table_value[row_idx][0] # Assuming Model is the first column
|
660 |
+
elif isinstance(table_value, dict) and 'data' in table_value and 0 <= row_idx < len(table_value['data']):
|
661 |
+
selected_model = table_value['data'][row_idx][0]
|
662 |
+
except Exception as e:
|
663 |
+
print(f"Error extracting model name: {e}")
|
664 |
+
traceback.print_exc()
|
665 |
+
|
666 |
+
# If we found a model name, toggle its selection
|
667 |
+
if selected_model:
|
668 |
+
print(f"Selected model: {selected_model}")
|
669 |
+
|
670 |
+
# Make sure the model exists in the current dataframe
|
671 |
+
available_models = current_df['Model'].tolist() if not current_df.empty else []
|
672 |
+
|
673 |
+
if selected_model in available_models:
|
674 |
+
# Add to list if not already there, otherwise remove (toggle selection)
|
675 |
+
if selected_model in selected_models:
|
676 |
+
selected_models.remove(selected_model)
|
677 |
+
else:
|
678 |
+
selected_models.append(selected_model)
|
679 |
+
else:
|
680 |
+
print(f"Model {selected_model} not found in current dataframe")
|
681 |
+
|
682 |
+
# Ensure only models from the current dataframe are included
|
683 |
+
available_models = current_df['Model'].tolist() if not current_df.empty else []
|
684 |
+
selected_models = [model for model in selected_models if model in available_models]
|
685 |
+
|
686 |
+
# If no models are selected after filtering, use the first available model
|
687 |
+
if not selected_models and available_models:
|
688 |
+
selected_models = [available_models[0]]
|
689 |
+
|
690 |
+
# Create radar charts using the current dataframe and metrics
|
691 |
+
gen_chart = create_radar_chart(current_df, selected_models, current_gen_metrics, "Performance on Generation Tasks")
|
692 |
+
ret_chart = create_radar_chart(current_df, selected_models, current_ret_metrics, "Performance on Retrieval Tasks")
|
693 |
+
|
694 |
+
return selected_models, gen_chart, ret_chart
|
695 |
+
except Exception as e:
|
696 |
+
print(f"Error in update_charts: {e}")
|
697 |
+
print(traceback.format_exc())
|
698 |
+
return selected_models, generation_chart.value, retrieval_chart.value
|
699 |
+
|
700 |
+
# Use custom event handler for row selection
|
701 |
+
metrics_table.select(
|
702 |
+
fn=update_charts,
|
703 |
+
inputs=[selected_models],
|
704 |
+
outputs=[selected_models, generation_chart, retrieval_chart]
|
705 |
+
)
|
706 |
+
|
707 |
+
# Add selection handlers for category tables too
|
708 |
+
for category_table in category_tables.values():
|
709 |
+
category_table.select(
|
710 |
+
fn=update_charts,
|
711 |
+
inputs=[selected_models],
|
712 |
+
outputs=[selected_models, generation_chart, retrieval_chart]
|
713 |
+
)
|
714 |
+
|
715 |
+
# Handle version filter changes
|
716 |
+
def update_data(n_versions, only_actual, current_selected_models):
|
717 |
+
try:
|
718 |
+
# Get updated data
|
719 |
+
new_df, new_ret_metrics, new_gen_metrics, new_category_metrics = filter_and_process_results(
|
720 |
+
results, n_versions=n_versions, only_actual_versions=only_actual
|
721 |
+
)
|
722 |
+
|
723 |
+
# Get available models
|
724 |
+
available_models = new_df['Model'].tolist() if not new_df.empty else []
|
725 |
+
|
726 |
+
# Filter selected models to only include those that exist in the new dataset
|
727 |
+
filtered_selected_models = [model for model in current_selected_models if model in available_models]
|
728 |
+
|
729 |
+
# If no previously selected models remain, select the top models
|
730 |
+
if not filtered_selected_models and available_models:
|
731 |
+
filtered_selected_models = available_models[:min(5, len(available_models))]
|
732 |
+
|
733 |
+
# Create radar charts
|
734 |
+
gen_chart = create_radar_chart(new_df, filtered_selected_models, new_gen_metrics, "Performance on Generation Tasks")
|
735 |
+
ret_chart = create_radar_chart(new_df, filtered_selected_models, new_ret_metrics, "Performance on Retrieval Tasks")
|
736 |
+
|
737 |
+
# Create summary dataframe
|
738 |
+
summary_df = create_summary_df(new_df, new_ret_metrics, new_gen_metrics)
|
739 |
+
|
740 |
+
# Create category tables dictionary for output
|
741 |
+
category_tables_output = {}
|
742 |
+
|
743 |
+
# First initialize all tables to empty DataFrame
|
744 |
+
for category in category_tables.keys():
|
745 |
+
category_tables_output[category] = pd.DataFrame()
|
746 |
+
|
747 |
+
# Then populate available tables
|
748 |
+
for category, _ in new_category_metrics:
|
749 |
+
if f"{category}_retrieval" in new_df.columns and f"{category}_generation" in new_df.columns:
|
750 |
+
category_df = create_category_df(new_df, category, f"{category}_retrieval", f"{category}_generation")
|
751 |
+
if category in category_tables:
|
752 |
+
category_tables_output[category] = category_df if not category_df.empty else pd.DataFrame()
|
753 |
+
|
754 |
+
# Prepare all outputs
|
755 |
+
outputs = [summary_df, gen_chart, ret_chart, filtered_selected_models]
|
756 |
+
|
757 |
+
# Add category tables to outputs in the same order as in category_tables
|
758 |
+
for category in category_tables.keys():
|
759 |
+
outputs.append(category_tables_output.get(category, pd.DataFrame()))
|
760 |
+
|
761 |
+
# Update global df for later use
|
762 |
+
global df, retrieval_metrics, generation_metrics
|
763 |
+
df = new_df
|
764 |
+
retrieval_metrics = new_ret_metrics
|
765 |
+
generation_metrics = new_gen_metrics
|
766 |
+
|
767 |
+
return outputs
|
768 |
+
except Exception as e:
|
769 |
+
print(f"Error in update_data: {e}")
|
770 |
+
print(traceback.format_exc())
|
771 |
+
# Return original values in case of error
|
772 |
+
empty_tables = [pd.DataFrame() for _ in category_tables]
|
773 |
+
return summary_df, generation_chart.value, retrieval_chart.value, current_selected_models, *empty_tables
|
774 |
+
|
775 |
+
# Define filter button outputs
|
776 |
+
filter_outputs = [metrics_table, generation_chart, retrieval_chart, selected_models]
|
777 |
+
# Add category tables to outputs
|
778 |
+
for category_table in category_tables.values():
|
779 |
+
filter_outputs.append(category_table)
|
780 |
+
|
781 |
+
filter_btn.click(
|
782 |
+
fn=update_data,
|
783 |
+
inputs=[n_versions_slider, only_actual_versions, selected_models],
|
784 |
+
outputs=filter_outputs
|
785 |
+
)
|
786 |
+
|
787 |
+
# Function to clear charts
|
788 |
+
def clear_charts():
|
789 |
+
empty_models = []
|
790 |
+
# Create empty charts
|
791 |
+
empty_gen_chart = create_radar_chart(df, empty_models, generation_metrics, "Performance on Generation Tasks")
|
792 |
+
empty_ret_chart = create_radar_chart(df, empty_models, retrieval_metrics, "Performance on Retrieval Tasks")
|
793 |
+
return empty_models, empty_gen_chart, empty_ret_chart
|
794 |
+
|
795 |
+
# Connect clear charts button
|
796 |
+
clear_charts_btn.click(
|
797 |
+
fn=clear_charts,
|
798 |
+
inputs=[],
|
799 |
+
outputs=[selected_models, generation_chart, retrieval_chart]
|
800 |
+
)
|
801 |
+
|
802 |
+
if __name__ == "__main__":
|
803 |
+
demo.launch()
|