Please share you experiences

[Joe]

Hi,

Would anyone in this group be willing to share their overall experience with Tetrad and what they most like and dislike about it? We are curious.

We are interested in what you think about the interface, the algorithms, the breakdown of responsibility between the different parts, the usability of it, what you use the most, what you use the least, etc.

We would surely appreciate it.

Thanks,

Joe

[John Weissert]

Good afternoon Joe, 

Just going to keep this simple and to the point regarding my experience with TETRAD

Like: 

User interface 

Variety of search algorithms 

Variety of data manipulation tools (although some additional discretization functionality would be great, like MDL etc)

Ease with which I can load and save data

Graph box and the ability to easily adapt search results

Updater box and classifier box are very useful and interesting!

Dislike: 

I see the estimator box as having the greatest room to grow.

- EM Bayes estimator does not populate an IM box (even though it should be a permissible parent)!

- Extra Bayes estimator that can provide better results than EM with missing data : Robins JM. (2000). Robust estimation in sequentially ignorable missing data and causal inference models. Proceedings of the American Statistical Association Section on Bayesian Statistical Science 1999, pp. 6-10.

- It would be very helpful to know how many rows of training data are used to calculate the conditional probabilities within the PM 

- It would be fantastic to be able to average estimations across different sets of data

Even better yet, make a sort of Frankenstein estimator box that can use the ML Bayes estimator for certain rows and/or variables, and EM for other rows. 

OR instead of a Frankenstein estimator box, develop a more flexible IM box that can manipulate estimations from multiple parent estimator box and combine them into one IM. 

I would also really like a way to easily combine the results of multiple searches into a single matrix or graphic interface 

What I use the most / least

Generally, I use data with mixed characteristics (ie. continuous + binary etc) so the search algorithms that assume continuous data I don't use. 

I generally don't use scoring search algorithms,  

I don't  use the graph manipulator box or regression box often.

Other than that Im a big fan and use most everything else at certain points.

--John