Rel 0.1.0

Author: goedman

TODO.md

@@ -2,22 +2,10 @@
 
 A lot!
 
-1. GraphViz (and LightGraph?) plots from the DAG model.
+1. GraphViz (and LightGraph?) plots from the DAG (and MC) model.
 2. @DAG (vs. the current Dict formulation).
-3. Method `adjustment_sets(dag, paths)` - options for conditioning
-4. Method `impliedConditionalIndependencies()`
-7. Other mixed graphs based on Sadeghi work for ggm
-8. Documentation
-9. Tests
-10. ...
-
-
-More testing:
-
-1. Method `all_paths(dag, :x, :y)` - find all paths between nodes :f and :l.
-2. Method `backdoor_paths(dag, paths, :x)` - which are backdoor paths to :x?
-3. Method `open_paths(dag, paths)` - which paths are open?
-4. Method `show_dag_path(dag, path)` - show path directions using arrows
-5. Dagitty & ggm conversion routines
-6. ...
+3. Latent variable support.
+4. AG/MC graphs and m_separation.
+5. MCMC based pcor_test().
+5. ...
 

docs/src/introduction.md

@@ -9,3 +9,5 @@ All three above references are great introductions to the use of causal models t
 [StructuralCausalModels.jl](https://github.com/StatisticalRethinkingJulia/StructuralCausalModels.jl) is part of the [StatisticalRethinkingJulia](https://github.com/StatisticalRethinkingJulia) eco system. The package, once registered, can be installed using `] add StructuralCausalModels`.
 
 Please report issues (or file a pull request) on Github if you find a problem.
+
+Note: Version 0.1.0 of StructuralCausalModels.jl does not support latent variables.

docs/src/versions.md

@@ -4,3 +4,4 @@
 
 1. Version for initial commit to Julia's registry.
 2. Introduction of basic functionality.
+3. *No* support latent variables.

docs/src/walkthrough.md

@@ -75,7 +75,21 @@ Rows ╲ Cols │  :mechanics     :vectors     :algebra    :analysis  :statistic
 
 ```
 
-Additional DAG related functions are `adjacency_matrix()`, `edge_matrix()`, `to_ggm()`, `from_ggm()`, `to_dagitty()`, `from_dagitty()`, `set_dag_df!()` and `set_dag_cov_matrix!()`.
+Additional DAG related functions are `adjacency_matrix()`, `edge_matrix()` and `dag_vars()`.
+
+# Importing from and exporting to [dagitty.net](http://www.dagitty.net/dags.html#), dagitty and ggm (both are R packages)
+
+Importing is easiest using the functions `from_dagitty()` and `from_ggm()` as shown above.
+
+To export to dagitty.net, copy and paste the output from `to_dagitty()` into the `Model code` field on the dagitty.net web interface.
+
+For both R packages, copy the output from `to_dagitty()` or `to_ggm()` to R.
+
+# Adding a observations DataFrame or a covariance matrix 
+
+Use `set_dag_df!()` and `set_dag_cov_matrix!()` for this. Note that if a DataFrame is added a covariance matrix is computed.
+
+Although this initial version of StructuralCausalModels does not support latent variables yet, by using the keyword argument `force=true` no check is performed if all vertices/variables in the causal diagram are present in the DataFrame or covariance matrix.
 
 # Directed separation
 
@@ -174,6 +188,10 @@ display(t)
 
 # Adjustment sets
 
+D_separation provides a set of conditional independencies given the causal model. The conditioning set closes (blocks) all paths. It provides ways to test the chosen causal model given observational data.
+
+The function `adjustment_sets()` answers a related question, i.e. how to prevent confounding in multiple regression models assuming the chosen causal model is correct.
+
 Setup the WaffleDivorce example from StatisticalRethinking:
 ```julia
 using StructuralCausalModels