Using gCastle With mixed type data: how?
jbdatascience opened this issue · comments
What is the best way to use gCastle on mixed type data, so for datasets with a combination of numerical and categorical (nominal and/of ordinal)?
Is any of the causal discovery algorithms specifically appropriate for that?
Or is the only possibility to prepare such datasets so that it only contains numerical data and than using gCastle on it?
Hello,
Sorry for the late reply. Currently, gCastle only considers numerical data and makes no distinctions for categorical/discrete data. There are some algorithms that specifically run experiments on discrete data for example DAG-GNN, you can take a look to see if it satisfies your requirements. You may take a look at the code here: https://github.com/fishmoon1234/DAG-GNN (as our implementation does not contain the code for the discrete use-case).
Hi I have a similar question related to working whit mixed data types, in the event I have binary data and numerical data, can I still used the algorithm for DAG-GNN you have implemented? This is because the treatment and outcome in the dataset are both binary and I do not wish to remove them. Thanks!