How tony-y/cgnn works
Main Components
a) Data Processing Layer
• Input Processing
• Crystal Structure Data
• Target Values (CSV)
• Graph Data Generation
• Data Splitting (train/val/test)
b) Core CGNN Architecture
• Node Feature Processing
• Graph Convolution Layers
• Fully Connected Layers
• Output Layer
c) Training/Prediction Pipeline
• Model Training System
• Validation System
• Testing/Prediction System
d) Utilities and Tools
• Visualization Tools
• Performance Metrics
• Data Management
Component Relationships
• Data flow from raw input through processing to training
• Model architecture flow
• Training/validation/testing pipeline
• Output generation and visualization
Specific Implementation Details
[Data Sources] → [Data Processing] → [CGNN Core] → [Output/Predictions]
• ↓ ↓ ↓ ↓
[File Formats] [Processing Tools] [Model Layers] [Visualization]
• Blue: Data processing components (data_utils.py)
• Green: Core model components (models.py, layers.py)
• Orange: Training/prediction pipeline (cgnn.py)
• Purple: Utility tools (tools directory)
Detailed Component Breakdown
Data Processing (src/data_utils.py):
[Crystal Structure] → [Graph Generation] → [Node Vectors]
• ↓ ↓ ↓
[targets.csv] → [graph_data.npz] → [config.json]
Model Architecture (src/models.py, src/layers.py):
[Node Features] → [Graph Convolution] → [Hidden Layers] → [Output]
• ↓ ↓ ↓ ↓
[n_node_feat] [n_conv layers] [n_hidden_feat] [Predictions]