How sacdallago/biotrainer works
Main Components
a. Core Components:
• Input Processing Layer
• Embedding Layer
• Model Layer
• Training Layer
• Inference Layer
• Configuration Management
• Protocol Handler
b. Supporting Components:
• Data Validation
• Utilities
• CLI Interface
• Docker Support
Component Relationships and Architecture
[Input Layer]
• ↓
[Embedding Layer]
• ↓
[Model Layer]
• ↓
[Training/Inference Layer]
Detailed Component Breakdown
a. Input Layer:
• FASTA file processing
• Data validation
• To configuration management system
b. Embedding Layer:
• Multiple embedder options (Hugging Face, One-Hot, Custom)
• Embedding preprocessing
• Embedding storage/caching
c. Model Layer:
• Available model architectures:
• CNN
• FNN
• Light Attention
• LogReg
• Model parameter management
d. Training/Inference Layer:
• Split into two parallel paths:
• Training pipeline with cross-validation, HP management
• Inference pipeline with model loading and prediction
Key Interactions to Highlight
• Data flow from input through embedding to model
• Protocol selection affecting the entire pipeline
• How the solver interfaces with models and training
Implementation Details
biotrainer/
• config/ (Configuration Management)
• datasets/ (Data Processing)
• embedders/ (Embedding Layer)
• models/ (Model Architectures)
• trainers/ (Training Pipeline)
• inference/ (Inference Pipeline)
• solvers/ (Problem Solvers)
Additional Elements
• Version information
• External dependencies (PyTorch, HuggingFace, etc.)
• Configuration file interaction points
Remember to include notes about the system's extensibility, particularly regarding custom embedders and model architectures, as this appears to be a key feature of the system based on the examples provided.