When you miss declaring a node in your causal graph, it's going to throw a KeyError: 'label' error. It could be more explicit to make debugging easier. I think it would be nice to inform what is the node hough used in the graph.

When profiling openalpr, the profile.coz file (named profile.txt in the attachment to keep github editor happy) show that a lot of time is spent in locations in /usr/include/c++/4.9/bits/stl_vector.h and /usr/include/opencv2/core/mat.hpp (matrix operations). This is not really helpful without knowing from where the STL vector or the matrix operation is used. Perhaps Coz should be adjusted to col

The documentation of some classes/methods is severely lacking. Here's a list of methods that needs more detailed documentation as has been pointed out by users:

• TabularCPD doesn't clearly specify what the arguments are expected to be. Ref: #1036
• TabularCPD methods need to describe in more detail what each of them does.
• Make sure that Return section is available in ea

As discussed on issue #70, it would be valuable to replicate the benchmark tests from the foundational research papers. For the meta-learners, we have the papers by Nie and Wager (2019) and Kunzel et al (2019) that contain such benchmark studies.

The simulation studies are:

•	Simulation setups A, B, C and D (Nie and Wager, p. 16)
•	Simulations 1—6

We need to add a yes and no label on the arrows when we plot a tree interpreter. Also we should be ideally using the population summary inference values, when calculating p-values and CIs at the leafs.

Description

Following the tutorial raises the following:

bn = bn.fit_cpds(train, method="BayesianEstimator", bayes_prior="K2")

Would be great to have a fully working jupyter notebook as an example.

Steps to Reproduce

/usr/local/lib/python3.7/site-packages/causalnex/network/network.py in fit_cpds(self, data, method, bayes_prior, equivalent_sample_size)
    344 
    345  

I ran a regression_forest for > 10 minutes and had no idea if it would complete in 15 min or an hour.

It would be great to have an argument "verbose" (default FALSE) which causes the function to
print the function's progress, to help the user estimate the remaining time before completion.

1. Add sample size estimation documentation.
2. Improve multiple correction documentation adding all formulas and assumptions.

feel free to add more

From the documentation:

To install deepiv, run this command in your terminal:
$ pip install deepiv

It looks like it's not installing through this method:

$ pip install deepiv
Collecting deepiv
  Could not find a version that satisfies the requirement deepiv (from versions: )
No matching distribution found for deepiv
``

On the website, create quick tutorials demonstrating each of the implemented estimators, descriptions of how they work, and why you might want to use them. Might be more digestible than the current docs (also better justify why to choose one over the other)

Reference to base on
https://lifelines.readthedocs.io/en/latest/jupyter_notebooks/Proportional%20hazard%20assumption.html
https://github.

So far, there is no guidance for new users on how to use the PC and FCI algorithm. Given that both algorithms come with a fair amount of theoretical baggage, it would be great if we could add a few tutorials discussing the theoretical background as well as basic usage of both algorithms.