I am trying to do causal discovery on my dataset by following the steps in csuite_example. However, I cannot get the reasonable result. The key difference between my use case and the example is the prior setting.
I only set 0 in the expert matrix (EM), for example, there shouldn't be any cause to gender so if j is gender, then EM[i,j] = 0. I have set a lot of edges to be 0 in expert matrix and else will be 1.
For relevance mask (RM), it is the inverse of expert matrix, EM[i,j] == 0 then RM[i,j] = 1 and EM[i,j] == 1 then RM[i,j] =0.
For confidence matrix (CR), it cloned the RM however I think CM can be either 0 or 1 for edges set to be 0 in EM.
All the setting is based on my understanding from below formula:
self._expert_graph_container.mask
(A - self._expert_graph_container.confidence * self._expert_graph_container.dag)
In the results, I don't have any causes to my dependent variables (I have 6 dependent variables and total 40+ variables). I am wondering:
Is my prior setting wrong? since I only saw the example on setting the edge to be 1 in EM, my use cause is totally different.
If the i is the parent node and j is the child node?
Maybe there is other issue like adjacency_dist, I totally followed the csuite_example for other parts.
Hope you can help on above questions and thanks in advance!
Hi team,
I am trying to do causal discovery on my dataset by following the steps in csuite_example. However, I cannot get the reasonable result. The key difference between my use case and the example is the prior setting.
All the setting is based on my understanding from below formula: self._expert_graph_container.mask
In the results, I don't have any causes to my dependent variables (I have 6 dependent variables and total 40+ variables). I am wondering:
Hope you can help on above questions and thanks in advance!