How antoine311200/bsde-tensor-networks works
Project Type and Purpose
• This is a scientific computing library and pipeline for solving high-dimensional parabolic PDEs via tensor-train methods.
• It provides a core Python package (src/bsde_solver) that implements:
• Basis and differential operators (calculus)
• Tensor-train data structures and operations
• Optimization algorithms (ALS, MALS, SALSA)
• Stochastic path generation
• End-to-end pipelines for explicit and implicit discretization
• On top of the core library, there are:
• CLI scripts (src/scripts) for running pipelines and benchmarking experiments
• Jupyter notebooks for exploration and visualization
• Automated tests under src/bsde_solver/tests
High-Level Architecture
• The system is layered:
1. User Interface Layer
– Jupyter notebooks (notebooks/…)
– Command-line scripts (src/scripts)
2. Pipeline Layer
– src/bsde_solver/pipeline.py (explicit)
– src/scripts/pipeline_implicit.py (implicit iterative)
3. Core Library Layer (src/bsde_solver)
a. Core Modules
• calculus: basis, derivative, Hessian
• tensor: core tensor operations, tensor_train, tensor_network
• optimisation: ALS, MALS, solve routines
• stochastic: Monte Carlo path generation
b. Auxiliary
• loss functions (loss.py)
• utility functions (utils.py)
4. Infrastructure & Configuration
– pyproject.toml (package metadata, dependencies)
– .gitignore
• Tests exercise each core submodule and ensure correctness.