Opération de maintenance au data center INRAE de Toulouse du jeudi 20 mars à 15h00 au lundi 24 Mars 2025 à 13h
Les services ForgeMia et Mattermost seront donc inaccessibles pendant cette période (plus de détails)
A set of R and Julia scripts for the application of Approximate Bayesian Computation with Deep Learning and Conformal Prediction.
This repository is an add-on of: Meili Baragatti, Bertrand Cloez, David Métivier, Isabelle Sanchez. Approximate Bayesian Computation with Deep Learning and Conformal prediction.
This repository contains the Supplementary Material for the paper, an add-on to: *Approximate Bayesian Computation with Deep Learning and Conformal Prediction* by Meili Baragatti, Bertrand Cloez, David Métivier, and Isabelle Sanchez.
The following links give the codes used to obtain and reproduce the results of the article. The codes are in Julia for the MA(2), Lotka-Volterra and Phytoplankton dynamics in lake examples, and in R for the Gaussian Field 2D example.
The following links provide the code used to obtain and reproduce the results of the article. The code is written in Julia for the MA(2), Lotka-Volterra, and Phytoplankton dynamics in a lake examples, and in R for the 2D Gaussian Field example.
-[2D Gaussian Field example](https://mistea.pages.mia.inra.fr/codes_articles/abcdconformal/R/GaussianFields/Comparaison_ABCD_conv2d_conformal.html)(1 parameter)
-[MA(2) example](https://mistea.pages.mia.inra.fr/codes_articles/abcdconformal/julia/MA2/ABC_Conformal_MA2.html)(2 parameters) → A well-known toy Bayesian example.
-[Discrete Lotka-Volterra example](https://mistea.pages.mia.inra.fr/codes_articles/abcdconformal/julia/LoktaVolterra/ABC_Conformal_LV.html)(3 parameters) → A very challenging case for some extreme parameter values.
-[Phytoplankton dynamics in a lake with a toy model example](https://mistea.pages.mia.inra.fr/codes_articles/abcdconformal/julia/Lake_norm/ABC_Conformal_Lake.html)(9 parameters) → A high-dimensional example.
-[Gaussian Field example](https://mistea.pages.mia.inra.fr/codes_articles/abcdconformal/R/GaussianFields/Comparaison_ABCD_conv2d_conformal.html)(1 parameter)
-[MA(2) example](https://mistea.pages.mia.inra.fr/codes_articles/abcdconformal/julia/MA2/ABC_Conformal_MA2.html)(2 parameters) -> Well known toy Bayesian example.
-[Discrete Lotka-Volterra example](https://mistea.pages.mia.inra.fr/codes_articles/abcdconformal/julia/LoktaVolterra/ABC_Conformal_LV.html)(3 parameters) -> Very challenging for some extreme parameters.
-[Phytoplankton dynamics in Lake with a toy model example](https://mistea.pages.mia.inra.fr/codes_articles/abcdconformal/julia/Lake_norm/ABC_Conformal_Lake.html)(9 parameters) -> High dimensional example.
We are also making the R code available for the MA(2) and Lotka-Volterra examples. The R code is located in the `/R` folder, while the Julia code is in the `/julia` folder.
But we are also making the R codes available, for the MA(2) and Lotka-Volterra examples (the R codes are in the /R folder, while the Julia codesare in the /Julia folder).
You can view all the `html`-rendered reports (in both R and Julia) on the following [website](https://mistea.pages.mia.inra.fr/codes_articles/abcdconformal).
You can visualize ALL the html reports (in R and Julia) in the following [web site](https://mistea.pages.mia.inra.fr/codes_articles/abcdconformal).
Please note that our method was first developed in R, and the examples were later transferred and reworked in Julia. As a result, the Julia code is more sophisticated than the R code, and the results obtained with Julia are sometimes superior to those obtained with R. The speed of Julia allows the examples to be run on much larger datasets than with R.
Please note that our method was first developed in R, and the examples were then transferred and reworked in Julia. As a consequence, the Julia codes are more sophisticated than the R codes, and the results obtained with Julia are sometimes better than those obtained with R. Indeed, the speed of Julia enables the examples to be run on much larger datasets than with R.
**Please take care to check the versions of Julia (> 1.10) and Quarto (> 1.6).**
### Settings
Download the `Project.toml` and `Manifest.toml` and datasets and pretrained models if available. Activate and instantiate the environment.
If you are not familiar with Julia, we recommend downloading the latest version using the command line provided on the [official Julia website](https://julialang.org/downloads/).
**Please ensure that you are using Julia (> 1.10) and Quarto (> 1.6).**
For each example, download the `Project.toml` and `Manifest.toml` files (as well as datasets and pretrained models if available) into the associated folder. Then, activate and instantiate the environment in the correct folder:
```julia
cd(@__DIR__)# cd inside the notebook folder where is stored the `Project.toml` (and `Manifest.toml`)
cd(@__DIR__)# Navigate to the notebook folder where `Project.toml` and `Manifest.toml` are stored
importPkg
Pkg.activate(".")# activate the environment (you can specify another path if needed)
Pkg.pkg"registry add https://github.com/dmetivie/LocalRegistry"# see comment bellow
Pkg.instantiate()# => install all pkgs listed
Pkg.activate(".")# Activate the environment (you can specify another path if needed)
Pkg.pkg"registry add https://github.com/dmetivie/LocalRegistry"# To access ABCMethods.jl and ConcreteDropoutLayer.jl
Pkg.instantiate()# Install all listed packages
```
Then you can run cell by cell the code or use Quarto render function (for example: `quarto render .\julia\Lake_norm\ABC_Conformal_Lake.qmd --to html`) to run the whole notebook at once.
> [!WARNING]
> In Julia, the first time you execute a command, it might take some time. This is particularly true when installing and precompiling all required packages.
For this project, we developed two packages that are not yet in the official Julia registry
-[ABCMethods.jl](https://github.com/dmetivie/ABCMethods.jl): ABC methods used in the paper.
-[ConcreteDropoutLayer.jl](https://github.com/dmetivie/ConcreteDropoutLayer.jl): `Lux.jl` version of the Concrete Dropout layer. Inspired by this [Python version](https://github.com/aurelio-amerio/ConcreteDropout).
The line `Pkg.pkg"registry add https://github.com/dmetivie/LocalRegistry"` (only needed once) tells your package manager where to find these two packages.
### GPUs and reproducibility
> [!WARNING]
> We trained all the Neural Network models (ABC-CNN and ABC-Conformal) with the [`Lux.jl`](https://lux.csail.mit.edu/stable/) deep learning package using GPUs with the extansion `LuxCuda.jl` on NVIDIA GPUs. This significantly improved training times (by at least 20x for larger applications).
> 1) Make sure you do have compatible GPUs or consult the [Lux GPU page](https://lux.csail.mit.edu/stable/#Want-GPU-Support?) to use other type of models. If you do not have GPUs just comment out the line `using LuxCUDA` and set `dev = cpu_device()`.
> 2) Despite trying our best to make the notebooks 100% reproducible from scratch -- providing the `Manifest.toml` of packages version, setting the initial random seed and configuring `Lux.jl` to “seed” its layers -- GPU operations can still exhibit some non-determinism (e.g. [PyTorch doc](https://pytorch.org/docs/stable/generated/torch.use_deterministic_algorithms.html#torch.use_deterministic_algorithms) for discussion and reference about this) due to optimizations that prioritize speed over reproducibility. As a result, training the models across different sessions, software versions, GPUs, or computers may yield slightly varying outcomes.
To accelerate notebooks, we provide in for the [Discrete Lotka-Volterra example](https://mistea.pages.mia.inra.fr/codes_articles/abcdconformal/julia/LoktaVolterra/ABC_Conformal_LV.html) and [Lake toy model example](https://mistea.pages.mia.inra.fr/codes_articles/abcdconformal/julia/Lake_norm/ABC_Conformal_Lake.html) the pre-trained neural networks models as well as the datasets (even if they are fast to regenerate).
The code to train them from scratch is also provided.
In the [MA(2) example](https://mistea.pages.mia.inra.fr/codes_articles/abcdconformal/julia/MA2/ABC_Conformal_MA2.html), we do not provide this pretrained model as it is the most lightweight example.
### Running the notbooks
You can run cell by cell the code or use Quarto render function (for example: `quarto render julia/Lake_norm/ABC_Conformal_Lake.qmd --to html`) to run the whole notebook at once.
We tested it works on Linux and Windows. However, we did notice on one of our other computers an error when using Quarto render function `ERROR: Malt.TerminatedWorkerException()` that we could not yet debug. If you encounter this, try deactivating the GPU package or running the notebook cell by cell.
## For R with R markdown notebooks
...
...
@@ -88,7 +119,7 @@ ABCDconformal
|-> GaussianFields: contains codes for GaussianFields example
|-> Concrete_Dropout_YGal_fix_tf_2_David.R
|-> Spatial_Concrete_Dropout_YGal_tf_2_fix.R
|-> Julia
|-> julia
|-> MA2: contains codes for MA2 example
|-> LoktaVoltera: contains codes for Lokta Voltera
