Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -0,0 +1,102 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import json
|
3 |
+
import requests
|
4 |
+
import numpy as np
|
5 |
+
import matplotlib.pyplot as plt
|
6 |
+
import seaborn as sns
|
7 |
+
from io import BytesIO
|
8 |
+
|
9 |
+
# -------------------------------
|
10 |
+
# 1. Configuration and Data Loading
|
11 |
+
# -------------------------------
|
12 |
+
# URL to the JSON file (the URL below resolves to the raw file)
|
13 |
+
DATA_URL = "https://huggingface.co/spaces/alielfilali01/3C3H-HeatMap/resolve/main/files/aragen_v1_results.json"
|
14 |
+
|
15 |
+
# Define the metrics order (6 dimensions)
|
16 |
+
METRICS = ["Correctness", "Completeness", "Conciseness", "Helpfulness", "Honesty", "Harmlessness"]
|
17 |
+
|
18 |
+
def load_data(url=DATA_URL):
|
19 |
+
response = requests.get(url)
|
20 |
+
data = response.json()
|
21 |
+
# Filter out any non-model entries (e.g. timestamp entries)
|
22 |
+
model_data = [entry for entry in data if "Meta" in entry]
|
23 |
+
return model_data
|
24 |
+
|
25 |
+
# Load the JSON data once when the app starts
|
26 |
+
DATA = load_data()
|
27 |
+
|
28 |
+
# Extract model names for the dropdown based on the JSON "Meta" field
|
29 |
+
def get_model_names(data):
|
30 |
+
model_names = [entry["Meta"]["Model Name"] for entry in data]
|
31 |
+
return model_names
|
32 |
+
|
33 |
+
MODEL_NAMES = get_model_names(DATA)
|
34 |
+
|
35 |
+
# -------------------------------
|
36 |
+
# 2. Heatmap Generation Functions
|
37 |
+
# -------------------------------
|
38 |
+
def generate_heatmap_image(model_entry):
|
39 |
+
"""
|
40 |
+
Given a model entry from the JSON data, this function extracts the 6 metrics,
|
41 |
+
computes a 6x6 similarity matrix using the definition: similarity = 1 - |v_i - v_j|,
|
42 |
+
and returns the heatmap image as bytes.
|
43 |
+
"""
|
44 |
+
scores = model_entry["claude-3.5-sonnet Scores"]["3C3H Scores"]
|
45 |
+
# Create a vector with the metrics in the defined order
|
46 |
+
v = np.array([scores[m] for m in METRICS])
|
47 |
+
# Compute the 6x6 similarity matrix
|
48 |
+
matrix = 1 - np.abs(np.subtract.outer(v, v))
|
49 |
+
|
50 |
+
# Create a mask for the upper triangle (diagonal remains visible)
|
51 |
+
mask = np.triu(np.ones_like(matrix, dtype=bool), k=1)
|
52 |
+
|
53 |
+
plt.figure(figsize=(6, 5))
|
54 |
+
ax = sns.heatmap(matrix,
|
55 |
+
mask=mask,
|
56 |
+
annot=True,
|
57 |
+
fmt=".2f",
|
58 |
+
cmap="viridis",
|
59 |
+
xticklabels=METRICS,
|
60 |
+
yticklabels=METRICS,
|
61 |
+
cbar_kws={"label": "Similarity"})
|
62 |
+
plt.title(f"Confusion Matrix for Model: {model_entry['Meta']['Model Name']}")
|
63 |
+
plt.xlabel("Metrics")
|
64 |
+
plt.ylabel("Metrics")
|
65 |
+
plt.tight_layout()
|
66 |
+
|
67 |
+
# Save the figure to a bytes buffer
|
68 |
+
buf = BytesIO()
|
69 |
+
plt.savefig(buf, format="png")
|
70 |
+
plt.close()
|
71 |
+
buf.seek(0)
|
72 |
+
return buf.read()
|
73 |
+
|
74 |
+
def generate_heatmaps(selected_model_names):
|
75 |
+
"""
|
76 |
+
Filters the global DATA for entries matching the selected model names,
|
77 |
+
generates a heatmap for each one, and returns a list of image bytes.
|
78 |
+
"""
|
79 |
+
filtered_entries = [entry for entry in DATA if entry["Meta"]["Model Name"] in selected_model_names]
|
80 |
+
images = []
|
81 |
+
for entry in filtered_entries:
|
82 |
+
img_bytes = generate_heatmap_image(entry)
|
83 |
+
images.append(img_bytes)
|
84 |
+
return images
|
85 |
+
|
86 |
+
# -------------------------------
|
87 |
+
# 3. Build the Gradio Interface
|
88 |
+
# -------------------------------
|
89 |
+
with gr.Blocks() as demo:
|
90 |
+
gr.Markdown("## 3C3H Heatmap Generator")
|
91 |
+
gr.Markdown("Select the models you want to compare and generate their heatmaps below.")
|
92 |
+
|
93 |
+
with gr.Row():
|
94 |
+
model_dropdown = gr.Dropdown(choices=MODEL_NAMES, label="Select Model(s)", multiselect=True, value=MODEL_NAMES[:3])
|
95 |
+
|
96 |
+
generate_btn = gr.Button("Generate Heatmaps")
|
97 |
+
gallery = gr.Gallery(label="Heatmaps").style(grid=[2], height="auto")
|
98 |
+
|
99 |
+
generate_btn.click(fn=generate_heatmaps, inputs=model_dropdown, outputs=gallery)
|
100 |
+
|
101 |
+
# Launch the Gradio app
|
102 |
+
demo.launch()
|