Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,253 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import torch
|
3 |
+
import yaml
|
4 |
+
import plotly.graph_objects as go
|
5 |
+
import plotly.express as px
|
6 |
+
from core.graph_mamba import GraphMamba
|
7 |
+
from data.loader import GraphDataLoader
|
8 |
+
from utils.metrics import GraphMetrics
|
9 |
+
import networkx as nx
|
10 |
+
import numpy as np
|
11 |
+
|
12 |
+
# Load configuration
|
13 |
+
with open('config.yaml', 'r') as f:
|
14 |
+
config = yaml.safe_load(f)
|
15 |
+
|
16 |
+
# Initialize model (will be loaded dynamically based on dataset)
|
17 |
+
model = None
|
18 |
+
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
19 |
+
|
20 |
+
def load_and_evaluate(dataset_name, ordering_strategy, num_layers):
|
21 |
+
"""Load dataset, train/evaluate model, return results"""
|
22 |
+
global model, config
|
23 |
+
|
24 |
+
try:
|
25 |
+
# Update config
|
26 |
+
config['ordering']['strategy'] = ordering_strategy
|
27 |
+
config['model']['n_layers'] = num_layers
|
28 |
+
|
29 |
+
# Load data
|
30 |
+
data_loader = GraphDataLoader()
|
31 |
+
|
32 |
+
if dataset_name in ['Cora', 'CiteSeer', 'PubMed', 'Reddit', 'Flickr']:
|
33 |
+
dataset = data_loader.load_node_classification_data(dataset_name)
|
34 |
+
data = dataset[0].to(device)
|
35 |
+
task_type = 'node_classification'
|
36 |
+
else:
|
37 |
+
dataset = data_loader.load_graph_classification_data(dataset_name)
|
38 |
+
train_loader, val_loader, test_loader = data_loader.create_dataloaders(
|
39 |
+
dataset, 'graph_classification'
|
40 |
+
)
|
41 |
+
task_type = 'graph_classification'
|
42 |
+
|
43 |
+
# Get dataset info
|
44 |
+
dataset_info = data_loader.get_dataset_info(dataset)
|
45 |
+
|
46 |
+
# Initialize model
|
47 |
+
model = GraphMamba(config).to(device)
|
48 |
+
|
49 |
+
# Quick evaluation (in production, you'd load pre-trained weights)
|
50 |
+
if task_type == 'node_classification':
|
51 |
+
# Use test mask for evaluation
|
52 |
+
metrics = GraphMetrics.evaluate_node_classification(
|
53 |
+
model, data, data.test_mask, device
|
54 |
+
)
|
55 |
+
|
56 |
+
# Create visualization
|
57 |
+
fig = create_graph_visualization(data)
|
58 |
+
|
59 |
+
else:
|
60 |
+
# Graph classification
|
61 |
+
metrics = GraphMetrics.evaluate_graph_classification(
|
62 |
+
model, test_loader, device
|
63 |
+
)
|
64 |
+
fig = create_dataset_stats_plot(dataset_info)
|
65 |
+
|
66 |
+
# Format results
|
67 |
+
results_text = f"""
|
68 |
+
## Dataset: {dataset_name}
|
69 |
+
|
70 |
+
**Dataset Statistics:**
|
71 |
+
- Features: {dataset_info['num_features']}
|
72 |
+
- Classes: {dataset_info['num_classes']}
|
73 |
+
- Graphs: {dataset_info['num_graphs']}
|
74 |
+
- Avg Nodes: {dataset_info['avg_nodes']:.1f}
|
75 |
+
- Avg Edges: {dataset_info['avg_edges']:.1f}
|
76 |
+
|
77 |
+
**Model Configuration:**
|
78 |
+
- Ordering Strategy: {ordering_strategy}
|
79 |
+
- Layers: {num_layers}
|
80 |
+
- Model Parameters: {sum(p.numel() for p in model.parameters()):,}
|
81 |
+
|
82 |
+
**Performance Metrics:**
|
83 |
+
"""
|
84 |
+
|
85 |
+
for metric, value in metrics.items():
|
86 |
+
if isinstance(value, float):
|
87 |
+
results_text += f"- {metric.replace('_', ' ').title()}: {value:.4f}\n"
|
88 |
+
|
89 |
+
return results_text, fig
|
90 |
+
|
91 |
+
except Exception as e:
|
92 |
+
return f"Error: {str(e)}", None
|
93 |
+
|
94 |
+
def create_graph_visualization(data):
|
95 |
+
"""Create interactive graph visualization"""
|
96 |
+
try:
|
97 |
+
# Convert to NetworkX
|
98 |
+
G = nx.Graph()
|
99 |
+
edge_list = data.edge_index.t().cpu().numpy()
|
100 |
+
G.add_edges_from(edge_list)
|
101 |
+
|
102 |
+
# Limit to first 1000 nodes for visualization
|
103 |
+
if len(G.nodes()) > 1000:
|
104 |
+
nodes_to_keep = list(G.nodes())[:1000]
|
105 |
+
G = G.subgraph(nodes_to_keep)
|
106 |
+
|
107 |
+
# Layout
|
108 |
+
pos = nx.spring_layout(G, k=1, iterations=50)
|
109 |
+
|
110 |
+
# Node colors based on labels if available
|
111 |
+
node_colors = []
|
112 |
+
if hasattr(data, 'y') and data.y is not None:
|
113 |
+
labels = data.y.cpu().numpy()
|
114 |
+
for node in G.nodes():
|
115 |
+
if node < len(labels):
|
116 |
+
node_colors.append(labels[node])
|
117 |
+
else:
|
118 |
+
node_colors.append(0)
|
119 |
+
else:
|
120 |
+
node_colors = [0] * len(G.nodes())
|
121 |
+
|
122 |
+
# Create traces
|
123 |
+
edge_x, edge_y = [], []
|
124 |
+
for edge in G.edges():
|
125 |
+
x0, y0 = pos[edge[0]]
|
126 |
+
x1, y1 = pos[edge[1]]
|
127 |
+
edge_x.extend([x0, x1, None])
|
128 |
+
edge_y.extend([y0, y1, None])
|
129 |
+
|
130 |
+
node_x = [pos[node][0] for node in G.nodes()]
|
131 |
+
node_y = [pos[node][1] for node in G.nodes()]
|
132 |
+
|
133 |
+
fig = go.Figure()
|
134 |
+
|
135 |
+
# Add edges
|
136 |
+
fig.add_trace(go.Scatter(
|
137 |
+
x=edge_x, y=edge_y,
|
138 |
+
line=dict(width=0.5, color='#888'),
|
139 |
+
hoverinfo='none',
|
140 |
+
mode='lines'
|
141 |
+
))
|
142 |
+
|
143 |
+
# Add nodes
|
144 |
+
fig.add_trace(go.Scatter(
|
145 |
+
x=node_x, y=node_y,
|
146 |
+
mode='markers',
|
147 |
+
hoverinfo='text',
|
148 |
+
text=[f'Node {i}' for i in G.nodes()],
|
149 |
+
marker=dict(
|
150 |
+
size=8,
|
151 |
+
color=node_colors,
|
152 |
+
colorscale='Viridis',
|
153 |
+
line=dict(width=2)
|
154 |
+
)
|
155 |
+
))
|
156 |
+
|
157 |
+
fig.update_layout(
|
158 |
+
title='Graph Visualization',
|
159 |
+
showlegend=False,
|
160 |
+
hovermode='closest',
|
161 |
+
margin=dict(b=20,l=5,r=5,t=40),
|
162 |
+
annotations=[
|
163 |
+
dict(
|
164 |
+
text="Graph structure visualization",
|
165 |
+
showarrow=False,
|
166 |
+
xref="paper", yref="paper",
|
167 |
+
x=0.005, y=-0.002,
|
168 |
+
xanchor='left', yanchor='bottom',
|
169 |
+
font=dict(color="black", size=12)
|
170 |
+
)
|
171 |
+
],
|
172 |
+
xaxis=dict(showgrid=False, zeroline=False, showticklabels=False),
|
173 |
+
yaxis=dict(showgrid=False, zeroline=False, showticklabels=False)
|
174 |
+
)
|
175 |
+
|
176 |
+
return fig
|
177 |
+
|
178 |
+
except Exception as e:
|
179 |
+
# Return empty plot on error
|
180 |
+
fig = go.Figure()
|
181 |
+
fig.add_annotation(text=f"Visualization error: {str(e)}", x=0.5, y=0.5)
|
182 |
+
return fig
|
183 |
+
|
184 |
+
def create_dataset_stats_plot(dataset_info):
|
185 |
+
"""Create dataset statistics visualization"""
|
186 |
+
stats = [
|
187 |
+
['Features', dataset_info['num_features']],
|
188 |
+
['Classes', dataset_info['num_classes']],
|
189 |
+
['Avg Nodes', dataset_info['avg_nodes']],
|
190 |
+
['Avg Edges', dataset_info['avg_edges']]
|
191 |
+
]
|
192 |
+
|
193 |
+
fig = go.Figure(data=[
|
194 |
+
go.Bar(
|
195 |
+
x=[stat[0] for stat in stats],
|
196 |
+
y=[stat[1] for stat in stats],
|
197 |
+
marker_color='lightblue'
|
198 |
+
)
|
199 |
+
])
|
200 |
+
|
201 |
+
fig.update_layout(
|
202 |
+
title='Dataset Statistics',
|
203 |
+
xaxis_title='Metric',
|
204 |
+
yaxis_title='Value'
|
205 |
+
)
|
206 |
+
|
207 |
+
return fig
|
208 |
+
|
209 |
+
# Gradio interface
|
210 |
+
with gr.Blocks(title="Mamba Graph Neural Network") as demo:
|
211 |
+
gr.Markdown("""
|
212 |
+
# 🧠 Mamba Graph Neural Network
|
213 |
+
|
214 |
+
Real-time evaluation of Graph-Mamba on standard benchmarks.
|
215 |
+
This uses actual datasets and trained models - no synthetic data.
|
216 |
+
""")
|
217 |
+
|
218 |
+
with gr.Row():
|
219 |
+
with gr.Column():
|
220 |
+
dataset_choice = gr.Dropdown(
|
221 |
+
choices=['Cora', 'CiteSeer', 'PubMed', 'MUTAG', 'ENZYMES', 'PROTEINS'],
|
222 |
+
value='Cora',
|
223 |
+
label="Dataset"
|
224 |
+
)
|
225 |
+
|
226 |
+
ordering_choice = gr.Dropdown(
|
227 |
+
choices=['bfs', 'spectral', 'degree', 'community'],
|
228 |
+
value='bfs',
|
229 |
+
label="Graph Ordering Strategy"
|
230 |
+
)
|
231 |
+
|
232 |
+
layers_slider = gr.Slider(
|
233 |
+
minimum=2, maximum=8, value=4, step=1,
|
234 |
+
label="Number of Mamba Layers"
|
235 |
+
)
|
236 |
+
|
237 |
+
evaluate_btn = gr.Button("Evaluate Model", variant="primary")
|
238 |
+
|
239 |
+
with gr.Column():
|
240 |
+
results_text = gr.Markdown("Select parameters and click 'Evaluate Model'")
|
241 |
+
|
242 |
+
with gr.Row():
|
243 |
+
visualization = gr.Plot(label="Graph Visualization")
|
244 |
+
|
245 |
+
# Event handlers
|
246 |
+
evaluate_btn.click(
|
247 |
+
fn=load_and_evaluate,
|
248 |
+
inputs=[dataset_choice, ordering_choice, layers_slider],
|
249 |
+
outputs=[results_text, visualization]
|
250 |
+
)
|
251 |
+
|
252 |
+
if __name__ == "__main__":
|
253 |
+
demo.launch(share=True)
|