torchode 0.2.0__tar.gz → 1.0.0__tar.gz

{torchode-0.2.0 → torchode-1.0.0}/CHANGELOG.md

@@ -5,6 +5,10 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## 1.0.0 - 2024-09-22
+
+After almost a year without any bugs found, we mark this as the first stable release!
+
 ## 0.2.0 - 2023-11-10
 
 ### Changed

torchode-1.0.0/CITATION.cff

@@ -0,0 +1,28 @@
+cff-version: 1.2.0
+title: torchode
+message: >-
+  If you use this software, please cite it using the
+  metadata from this file.
+type: software
+authors:
+  - given-names: Marten
+    family-names: Lienen
+    email: m.lienen@tum.de
+  - given-names: Stephan
+    family-names: Günnemann
+    email: s.guennemann@tum.de
+repository-code: "https://github.com/martenlienen/torchode"
+license: MIT
+preferred-citation:
+  type: conference-paper
+  title: "torchode: A Parallel ODE Solver for PyTorch"
+  authors:
+    - given-names: Marten
+      family-names: Lienen
+      email: m.lienen@tum.de
+    - given-names: Stephan
+      family-names: Günnemann
+      email: s.guennemann@tum.de
+  collection-title: "The Symbiosis of Deep Learning and Differential Equations II, NeurIPS"
+  year: 2022
+  url: "https://openreview.net/forum?id=uiKVKTiUYB0"

{torchode-0.2.0 → torchode-1.0.0}/PKG-INFO

@@ -1,11 +1,11 @@
 Metadata-Version: 2.1
 Name: torchode
-Version: 0.2.0
+Version: 1.0.0
 Summary: A parallel ODE solver for PyTorch
 Author-email: Marten Lienen <m.lienen@tum.de>
 Requires-Python: >= 3.8
 Description-Content-Type: text/markdown
-Classifier: Development Status :: 4 - Beta
+Classifier: Development Status :: 5 - Production/Stable
 Classifier: Intended Audience :: Developers
 Classifier: Intended Audience :: Science/Research
 Classifier: License :: OSI Approved :: MIT License

torchode-1.0.0/docs/backprop.md

@@ -0,0 +1,24 @@
+# Gradients & Backpropagation
+
+There are two ways to compute gradients of the dynamics of an ODE, so the neural network
+in the case of neural ODEs, with respect to the solution of the ODE. The first is to
+backpropagate straight through the solver. After all, an ODE solver is just a series of
+simple operations that define a dynamic computation graph that can be backpropagated
+through with pytorch's autograd. This is implemented in `to.AutoDiffAdjoint`, so called
+because it uses the autodiff/autograd mechanism. In general, this is the preferred method
+as long as you have enough memory, because it is fast and gives accurate gradients.
+
+If you run out of memory, you can compute gradients by solving the so called adjoint
+equations, which basically solve the ODE backwards and track gradients along the way. This
+is implemented in `to.BacksolveAdjoint`. Solving the adjoint equations requires the
+computation of gradients of the model at different steps in time, which
+`to.BacksolveAdjoint` implements with `torch.func`. If your model is not compatible and
+you get errors because of this, you can fall back to `to.JointBacksolveAdjoint`. This
+computes the model gradients with pytorch's usual autograd and should always work but
+comes with two caveats. However, to make this work, `to.JointBacksolveAdjoint` needs to
+solve the `n` independent adjoint equations jointly as one joint system that is jointly
+discretized. This breaks with torchode's approach of solving each ODE completely
+independently, because the joint discretization introduces a subtle coupling between the
+solutions of your batch of ODEs. Therefore, `to.JointBacksolveAdjoint` should be your
+backpropagation choice of last resort. Furthermore, it is only applicable if all ODEs in
+your batch have the same evaluation points.

{torchode-0.2.0 → torchode-1.0.0}/mkdocs.yml

@@ -12,6 +12,7 @@ plugins:
 nav:
   - "Introduction": README.md
   - step-size-controllers.md
+  - backprop.md
   - jit.ipynb
   - extra-args.md
   - extra-stats.ipynb

{torchode-0.2.0 → torchode-1.0.0}/pyproject.toml

@@ -10,7 +10,7 @@ requires-python = ">= 3.8"
 license = {file = "LICENSE"}
 authors = [{name = "Marten Lienen", email = "m.lienen@tum.de"}]
 classifiers = [
-  "Development Status :: 4 - Beta",
+  "Development Status :: 5 - Production/Stable",
   "Intended Audience :: Developers",
   "Intended Audience :: Science/Research",
   "License :: OSI Approved :: MIT License",

{torchode-0.2.0 → torchode-1.0.0}/torchode/__init__.py

@@ -1,6 +1,6 @@
 """A parallel ODE solver for PyTorch"""
 
-__version__ = "0.2.0"
+__version__ = "1.0.0"
 
 from .adjoints import AutoDiffAdjoint, BacksolveAdjoint, JointBacksolveAdjoint
 from .interface import register_method, solve_ivp