Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Release 0.6 #16

Merged
merged 3 commits into from
Sep 15, 2022
Merged

Release 0.6 #16

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
33f598a
Update README.md
michel2323 Sep 15, 2022
d7d855a
Test against Julia 1.8
michel2323 Sep 15, 2022
9fbcdc3
Bump version and dependencies
michel2323 Sep 15, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion .github/workflows/action.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,7 +18,7 @@ jobs:
strategy:
matrix:
os: [ubuntu-20.04]
julia-version: ['1.7']
julia-version: ['1.7', '1.8']
julia-arch: [x64]

steps:
Expand Down
4 changes: 2 additions & 2 deletions Project.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,7 +1,7 @@
name = "Checkpointing"
uuid = "eb46d486-4f9c-4c3d-b445-a617f2a2f1ca"
authors = ["Michel Schanen <[email protected]>", "Sri Hari Krishna Narayanan <[email protected]>"]
version = "0.5.0"
version = "0.6.0"

[deps]
ChainRulesCore = "d360d2e6-b24c-11e9-a2a3-2a2ae2dbcce4"
Expand All @@ -12,7 +12,7 @@ LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
Serialization = "9e88b42a-f829-5b0c-bbe9-9e923198166b"

[compat]
ChainRulesCore = "1.0"
ChainRulesCore = "1"
DataStructures = "0.18"
Enzyme = "0.10"
HDF5 = "0.16"
Expand Down
10 changes: 6 additions & 4 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,10 +3,11 @@

This package provides checkpointing schemes for adjoint computations using automatic differentiation (AD) of time stepping loops. Currently, we support the macro `@checkpoint_struct`, which differentiates and checkpoints a struct used in the loop. Each loop iteration is differentiated using [Enzyme.jl](https://github.com/EnzymeAD/Enzyme.jl). We rely on [ChainRulesCore.jl](https://github.com/JuliaDiff/ChainRulesCore.jl) to integrate with AD tools applied to the code outside of the loop.

The schemes are agnostic to the ADTool being used and can be easily interfaced with any Julia AD tool. Currently the package provides the following checkpointing schemes:
The schemes are agnostic to the AD tool being used and can be easily interfaced with any Julia AD tool. Currently, the package provides the following checkpointing schemes:

1. Revolve/Binomial checkpointing [1]
2. Periodic checkpointing
3. Online r=2 checkpointing for while loops with a priori unknown number of iterations [2]

## Installation

Expand All @@ -16,7 +17,7 @@ add Checkpointing

## Usage: Example 1D heat equation

We present an example code where Zygote is used to differentiate the implementation of the explicit 1D heat equation. The macro `@checkpointing_struct` covers the transformation of `for` loops with `1:tsteps` ranges where `tsteps=500` is the number of timesteps. As a checkpointing scheme we use Revolve and use a maximum of only 4 snapshots. This implies that instead of requiring to save all 500 temperature fields for the gradient computation, we now only need 4. As a trade-off, recomputation is used to recompute intermediate temperature fields.
We present an example code where Zygote is used to differentiate the implementation of the explicit 1D heat equation. The macro `@checkpointing_struct` covers the transformation of `for` loops with `1:tsteps` ranges where `tsteps=500` is the number of timesteps. As a checkpointing scheme, we use Revolve and use a maximum of only 4 snapshots. This implies that instead of requiring to save all 500 temperature fields for the gradient computation, we now only need 4. As a trade-off, recomputation is used to recompute intermediate temperature fields.

```julia
# Explicit 1D heat equation
Expand Down Expand Up @@ -89,11 +90,12 @@ The following features are planned for development:
* Composition of checkpointing schemes
* Multi-level checkpointing schemes

[1] Andreas Griewank and Andrea Walther. 2000. Algorithm 799: revolve: an implementation of checkpointing for the reverse or adjoint mode of computational differentiation. ACM Trans. Math. Softw. 26, 1 (March 2000), 19–45. DOI:https://doi.org/10.1145/347837.347846
[1] Andreas Griewank and Andrea Walther, Algorithm 799: Revolve: An Implementation of Checkpointing for the Reverse or Adjoint Mode of Computational Differentiation. ACM Trans. Math. Softw. 26, 1 (March 2000), 19–45. DOI: [10.1145/347837.347846](https://doi.org/10.1145/347837.347846)
[2] Philipp Stumm and Andrea Walther, New Algorithms for Optimal Online Checkpointing, 2010, DOI: (10.1137/080742439)[(https://doi.org/10.1137/080742439)

## Funding

This work is supported by the NSF Cyberinfrastructure for Sustained Scientific Innovation (CSSI) program project (DJ4Earth)[https://dj4earth.github.io/]
This work is supported by the NSF Cyberinfrastructure for Sustained Scientific Innovation (CSSI) program project [DJ4Earth](https://dj4earth.github.io/)

[docs-stable-img]: https://img.shields.io/badge/docs-stable-blue.svg
[docs-stable-url]: https://Argonne-National-Laboratory.github.io/Checkpointing.jl/
Expand Down