say the dag

X1 --> X2
X2 --> X3
X3 --> X4

when converted into the pdag
X1 --> X2
X2 --> X3
X3 --- X4

meek is fine

but
X1 --> X2
X2 --> X3
X4 --- X3

makes meek not halt

I also need to figure out how to make deletions fast.

Other than that, C refactor is pretty much done

When an object is converted into a cgraph, it must be cast first to a dag before being cast to a pattern. Should fix

Running a bootstrapped tetradrunner FCI with a high enough alpha value (0.1) will sometimes throw the error:

Error in cgraph(graph$nodes, new_edges) : Unrecognized edge type o--.

Printing the graph's edge lists doesn't show that edge type in the list, and running it again will sometimes not throw the error and proceed normally.

currently, is.cylic only really works when the edge set of the input is directed.

doing meek rules may help with this situation, but I suspect the topological sort algorithm may need need a slight rethink/

test case:
the edge list:
X1 X2 ---
X2 X3 ---
X3 X1 ---

should (need to check) be incorrectly labeled as "cyclic"

This will have to be done before i can close the c-refactor branch

This is already done.

(from @tk27182)
I was able to install causality on MacOS Mojave 10.14.5 by doing the following:

Opening up a terminal, then running the command open /Library/Developer/CommandLineTools/Packages/macOS_SDK_headers_for_macOS_10.14.pkg. If you've already installed that, you can ignore this step. This package wasn't automatically installed for me.

Since causality uses gcc to compile, you can (re)install gcc by typing brew install gcc in the terminal and make sure it's added to the /usr/local/bin/ path. You can check by using the which gcc command. For me, brew installed gcc-9 and that's the version I used.

Next edit the R Makevars file, which on Mac is found in the /Library/Frameworks/R.framework/Resources/etc folder and is called Makeconf. I used vim Makeconf.

Open the file and the variables CC, CXX, and CXX11. They should be

CC=/usr/local/bin/gcc-9
CXX=/usr/local/bin/g++-9 -std=gnu++11
CXX11=/usr/local/bin/g++-9

or whatever version of gcc you're using.

From there you can open up an R console and install the package like you would install any package from Github. I did library(devtools); install_github('causality')

This affects a few of the as.foo functions

Need to implement this algorithm to resolve this issue

Very rarely, bootstrapping a real data set may result in a data set whose correlations are weak enough that no edges are identified. When this happens, Tetrad outputs a graph object that contains nodes but whose list of edges is empty. When this happens as.cgraph throws an error, exiting whatever loop it's inside of. Ideally, as.cgraph should just convert this to a cgraph with the same nodes but no edges.

need a function that returns the actual Delta BIC for adding an edge, that just takes in the dataset ,and an edge.

delta_score := BIC( y+z,x) - BIC(y,x)

1. X2 --- X3
2. X2 --- X4
3. X2 --- X5
4. X3 --> X5
5. X4 --> X5

is not oriented according to meek3.

unclear on whether or not this is related to possible unoriented edge problemos

This may cause memory leaks when functions are interrupted. However, fixing this is low priority. This needs to be fixed before submitting to CRAN

according to meek rules, this pdag, A

1 --> 2
2 --> 3
3 --> 4
4 --- 1

should be oriented to the
1 --> 2
2 --> 3
3 --> 4
4 --> 1

which is not a pdag.

I'm guessing this has to do with the fact that changing the below DAG to A destroys an unshielded collider,
1 --> 2
2 --> 3
3 --> 4
1 --> 4

My question is: why isn't A a PDAG?

This paper (referenced in Pearl) contains a definition
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.74.5538&rep=rep1&type=pdf

There are numerous technical problems that might be easier to solve via some changes to how undirected edges are stored

summary.causality.graph -- give global stats on connectedness etc
equivalent -- test whether or not two causality.graphs are observationally equivalent.
identical (probably different name needed). test if two graphs are EXACTLY THE SAME.

X1 --> X2
X2 --> X3
X3 --- X4

results in

X1 --> X2
X2 --> X3
X3 --> X4

X1 --> X2
X2 --> X3
X4 --- X3

results in

X1 --> X2
X2 --> X3
X4 --> X3

is this good?

solution to is maximally orient pdag? aka meek?
but, #3 is present so this can't be tested

check chickering to see if is his algorithm can be used to turn pdags into patterns; I suspect yes.

Graph Nodes:
X2,X3,X4,X5

Graph Edges:

1. X2 --- X3
2. X2 --- X4
3. X2 --- X5
4. X5 --> X3
5. X4 --> X5

doesn't orient X2 --> X3