To replicate the paper main experiment, you will need to download the data.
- ACIC 2016: We use the R package associated with the data. You will a version of R installed and perform the following steps:
Load one acic 2016 simulation based on their R package,
[aciccomp2016](https://github.com/vdorie/aciccomp)
Pre-requisite:
```
if (require("remotes", quietly = TRUE) == FALSE) {
install.packages("remotes")
require("remotes")
}
remotes::install_github("vdorie/aciccomp/2016")
-
ACIC 2018: Download only the scaling subset of the data at https://www.synapse.org/#!Synapse:syn11738963
-
TWINS: The twins dataset can be downloaded with the utilitary function
load_twins
fromcaussim.data.loading.py
-
Caussim Dataset: This dataset is simulation based only. The generator is the class
CausalSimulator
fromcaussim.data.simulations.py
The experiences.zip
link on zenodo contain the results of the main experience of the
paper. These results have been obtained by launching as is, the script
/scripts/experiences/causal_scores_evaluation.py
.
The csv contained in the zip should be sufficient to replicate the Figure 5.,
using
/scripts/reports/causal_scores_evaluation.py
.