Skip to content

Repository for "Differentiable Causal Discovery from Interventional Data"

License

Notifications You must be signed in to change notification settings

slachapelle/dcdi

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

27 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Differentiable Causal Discovery from Interventional Data

DCDI paper graphics

Repository for "Differentiable Causal Discovery from Interventional Data".

  • The code for DCDI is available here.

  • The datasets used in the paper are available here

The code for the following baselines is also provided:

To run DCDI, you can use a command similar to this one: python ./dcdi/main.py --train --data-path ./data/perfect/data_p10_e10_n10000_linear_struct --num-vars 10 --i-dataset 1 --exp-path exp --model DCDI-DSF --intervention --intervention-type perfect --intervention-knowledge known --reg-coeff 0.5

Here, we assume the zip files in the directory data have been unzipped and that the results will be saved in the directory exp. With this command, you will train the model DCDI-DSF in the perfect known setting (with reasonable default hyperparameters) using the first instance (i_dataset = 1) of the 10-node graph dataset with linear mechanisms and perfect intervention. For further details on other hyperparameters (e.g. the architecture of the networks), see the main.py file where hyperparameters have a description.

About

Repository for "Differentiable Causal Discovery from Interventional Data"

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published