How jay-feng2008/phonocardiogram-classification works
This repository is a research implementation and experiment archive for phonocardiogram-based valvular heart disease classification, rather than a packaged application. The documented workflow is to train a ProbSparse self-attention classifier with virtual adversarial training, tune it with Bayesian optimization, and evaluate classification performance. The tree also preserves cross-validation, baseline/model comparisons, and ablation results. Source excerpts confirm the TensorFlow training and attention-layer components; several data-preparation and evaluation connections are documented only broadly, so those links are marked unverified or omitted.
Main entry points: metrics Model Metrics · layers.py ProbSparse Layers · Config-Oct-9/main.py VAT Training · 10-fold-crossvalidation/main.py 10-Fold Evaluation
Components 18
PCG Data
- Raw PCG Data
raw_data - MFCC Features
Config-Oct-9/model_baseline/mfcc.npz
Classification Model
- ProbSparse Layers
layers.py - Transformer Encoder
Config-Oct-9/encoder.py
Training Experiments
- VAT Training
Config-Oct-9/main.py - VAT Baseline
Config-Oct-9/VAT/model_baseline+vat.py - 10-Fold Evaluation
Config-Oct-9/10-fold-crossvalidation/main.py - Updated Cross-Validation
Config-Nov-17-2023/10-fold-cross-validation.py - Bayesian Tuning
Config-Oct-9/Bayesian_optimization/main.py - Hyperparameter Search
Config-Oct-9/grid_search - Baseline Training
Config-Oct-9/model_baseline/model_baseline.py - InceptionTime Comparison
InceptionTime - MLSTM-FCN Comparison
MLSTM-FCN - ResNet Comparison
Resnet
Evaluation Results
- Ablation Study
Ablation Study - Model Metrics
metrics - Result Figures
Figures
Other components
- Researcher
Connections 17
- Researcher → VAT Training: runs
- Raw PCG Data → MFCC Features: feature extraction
- MFCC Features → VAT Training: trains on
- VAT Training → ProbSparse Layers: uses layers
Config-Oct-9/main.py - VAT Baseline → ProbSparse Layers: uses layers
Config-Oct-9/VAT/model_baseline+vat.py - 10-Fold Evaluation → ProbSparse Layers: uses layers
Config-Oct-9/10-fold-crossvalidation/main.py - Updated Cross-Validation → ProbSparse Layers: uses layers
Config-Nov-17-2023/10-fold-cross-validation.py - Bayesian Tuning → ProbSparse Layers: uses layers
Config-Oct-9/Bayesian_optimization/main.py - Baseline Training → ProbSparse Layers: uses layers
Config-Oct-9/model_baseline/model_baseline.py - Bayesian Tuning → VAT Training: tunes training
README.md - VAT Training → Model Metrics: evaluates
README.md - 10-Fold Evaluation → Model Metrics: reports results
- Ablation Study → Model Metrics: compares components
README.md - InceptionTime Comparison → Model Metrics: model comparison
- MLSTM-FCN Comparison → Model Metrics: model comparison
- ResNet Comparison → Model Metrics: model comparison
- Model Metrics → Result Figures: summarized in