Arbitrary causal graphs

[johndavidbustard]

Is it possible to take as input a table of input values, an arbitrary causal graph representing how these values might cause each other and a table of test values with one or more missing columns and use the dowhy and econml library to infer the values of the missing columns from the test data. We also hope to use dowhy to create measurements testing the validity of the input causal graph. The examples have been very helpful to show how certain types of causal graph can be used to achieve similar goals but we are hoping to tackle the general case.
The application we are applying this to is the prediction of the revenue of different businesses over time based on different business actions such as renovating the business, introducing a new product or changes in tourism or other economic factors etc.
Is this possible with the dowhy library? Is there any existing code or a simple way we can decompose an arbitrary causal graph into multiple prediction problems? Or simply represent the entire graph within the dowhy framework?
Ultimately we also hope to have the table of input values and the test table with the missing columns, without any input causal graph, and produce a set of potentially valid/likely causal graphs and predicted values. Do you have any advice on whether the dowhy framework is suitable for this task or if another approach would be recommended?
We would be very grateful for any help you can provide us,
John

Dr John Bustard
Lecturer in the department of Electrical Engineering and Computer Science
Queen’s University
Belfast

[amit-sharma]

This is a great open-ended question, and I believe will benefit from a longer discussion.
I don't think that DoWhy in its current form supports these questions, but perhaps we can identify which functionalities can be supported in the future.

I'm moving this to the discussions page of the project.