# HAM10000 QSPICE Pipeline A comprehensive pipeline for training machine learning models on the HAM10000 skin cancer dataset and analyzing the impact of SRAM circuit storage on model performance using QSPICE simulation. ## Project Structure ``` HAM10000/ ├── src/ # Source code │ ├── data/ # Data processing components │ ├── models/ # Model components │ ├── circuits/ # Circuit design and simulation │ ├── analysis/ # Analysis and comparison tools │ ├── config.py # Configuration management │ └── main.py # Main pipeline entry point ├── data/ # Data storage │ ├── raw/ # Raw HAM10000 dataset │ └── processed/ # Processed data ├── models/ # Model storage │ ├── pretrained/ # Pre-trained models │ ├── adapted/ # HAM10000-adapted models │ └── sram/ # SRAM-affected models ├── circuits/ # Circuit files │ ├── qspice/ # QSPICE circuit files │ └── netlists/ # Circuit netlists ├── results/ # Analysis results │ ├── performance/ # Performance metrics │ ├── plots/ # Generated plots │ └── reports/ # Analysis reports ├── tests/ # Test files ├── config/ # Configuration files │ └── pipeline_config.yaml # Main configuration ├── requirements.txt # Python dependencies ├── setup.py # Package setup └── README.md # This file ``` ## Installation 1. Install Python dependencies: ```bash pip install -r requirements.txt ``` 2. Install the package in development mode: ```bash pip install -e . ``` ## Configuration The pipeline is configured through `config/pipeline_config.yaml`. Key configuration sections include: - **data**: Dataset paths and preprocessing parameters - **model**: Model architecture and training parameters - **sram**: SRAM circuit design parameters - **qspice**: QSPICE simulation settings - **analysis**: Analysis and reporting configuration ## Usage Run the complete pipeline: ```bash python src/main.py ``` Run specific stages: ```bash python src/main.py --stage data # Data processing only python src/main.py --stage model # Model training only python src/main.py --stage circuit # Circuit design only python src/main.py --stage analysis # Analysis only ``` Use custom configuration: ```bash python src/main.py --config path/to/custom_config.yaml ``` ## Pipeline Stages 1. **Data Processing**: Load and preprocess HAM10000 dataset 2. **Model Adaptation**: Load pre-trained model and adapt for HAM10000 3. **Weight Extraction**: Extract model weights for circuit storage 4. **Circuit Design**: Design SRAM circuit in QSPICE 5. **Weight Storage**: Simulate storing weights in SRAM circuit 6. **Performance Analysis**: Compare original vs SRAM-affected model performance 7. **Circuit Optimization**: Optimize SRAM design for minimal accuracy loss ## Dependencies - **torch**: PyTorch deep learning framework - **torchvision**: Computer vision utilities - **timm**: Pre-trained model library - **pandas**: Data manipulation - **numpy**: Numerical computing - **matplotlib**: Plotting and visualization - **hypothesis**: Property-based testing - **pytest**: Testing framework ## Testing Run tests: ```bash pytest tests/ ``` Run property-based tests: ```bash pytest tests/ -k "property" ``` ## Development Status This project is currently in development. The basic structure and configuration system are complete. Individual pipeline components will be implemented in subsequent development phases.# HAM10000