How mackelab/simformer works
Main Components
The system should be divided into these key components:
a) Core Framework Layer:
• Subcomponents for:
• Custom Primitives
• Interpreters
• JAX Propagation
• Distributions
• Neural Network Components
• Inference Methods
b) Simulation-Based Inference Layer:
• scoresbibm as the main application layer
• Components for:
• Tasks (Various inference problems)
• Methods (Different inference approaches)
• Evaluation
• Configuration Management
Component Relationships
• Dependencies from Methods to Core Framework components
• Configuration management affecting all components
Detailed Component Breakdown
For each major component, include:
a) Core Framework (probjax):
┌─ Core Framework ─────────────────┐
• ┌─ Primitives ┐ ┌─ Interpreters┐│
• RV │ │- Constraints ││
• Custom │ │- Inverse ││
• ┘ └─────────────┘│
• ┌─ Distributions ┐ ┌─ NN ─────┐│
• Continuous │ │- Attention││
• Discrete │ │- Bijective││
• ┘ └───────────┘│
• ┘
b) Inference Layer (scoresbibm):
┌─ Simulation-Based Inference ────┐
• ┌─ Tasks ─┐ ┌─ Methods ───────┐│
• SBIBM │ │- Score SBI ││
• Custom │ │- Transformers ││
• ┘ └────────────────┘│
• ┌─ Evaluation ┐ ┌─ Config ───┐│
• Metrics │ │- Hydra ││
• Analysis │ │- YAML ││
• ┘ └────────────┘│
• ┘
Key Elements to Highlight
• Data flow from simulation tasks through the inference methods
• Transformer architecture's role
• Configuration management's influence
• Evaluation feedback loops