Add option to show all edges removed

[honghaoli42]

Original report by Marcel Ribeiro-Dantas (Bitbucket: mribeirodantas_, ).

---

We recently changed MIIC so that by default the final result variable has information on all edges. For the removed edges, it keeps both edges removed by marginal independence and edges removed by conditioning on a sep set.

Now, we should bring it back to the way it was before (by default, do not keep info about edges removed by marginal independence) but add a parameter so that, if set, these edges are kept in the final result variable.

[honghaoli42]

@mribeirodantas What's the current status of this? Is the option implemented?

[mribeirodantas]

At first, what I had in mind was to leave this to the webserver. The user could easily show only non-removed edges through the filter widget (type). However, thinking about it now, R-package users could benefit from a parameter that would keep only the retained edges in the final result variable.

We'd have a parameter that, if set, would keep removed edges information in the final variable or, otherwise, they're removed (default behavior). I can implement that. What do you think?

[honghaoli42]

I suppose by "removed edges" you mean those removed with marginal independence (without conditioning)?
From what I see it's already implemented but commented out for some reason:

miic_R_package/R/parseResults.R

Lines 24 to 29 in dd41854

	# List of negative edges with null conditioning set
	indep_null_cond_set_edges <- as.matrix(edges[(is.na(edges$ai.vect)) &
	edges$category != 1, c("x", "y")])
	# Add to summarized edges list
	summarized_edges <- rbind(summarized_edges, conditioned_edges)
	#indep_null_cond_set_edges)

So you can simply add a parameter as a condition, though I don't quite see the interest in having those edges shown because the size of the edge list would be of the order of n_nodes^2

[mribeirodantas]

I had thought of having (a) only retained edges at the end or (b) all edges (retained and removed, either by conditional independence or not). I agree that the number of edges can be very large, I even brought this up to Hervé and his reply was that people could turn off this through the parameter if they wanted so. I don't really mind, to be honest. What do you think?

[honghaoli42]

I think edges removed with conditioning should be kept (as what we have currently) because they have information on the orientation of retained edges.
You can add the parameter, and i agree that it should be turned off by default.