Causality Causality Workbench                                                             Challenges in Machine Learning Causality

Virtual Laboratory

Welcome to the Virtual Lab. The table below lists the models available for experimentation. More detailed information bout a given model is obtained by clicking on its name. The goal is to make predictions of some unknown target values on a test set. Because test data include "manipulations", unraveling the causal structure of the model may be necessary to make good predictions. To do so, you can perform "experiments", i.e. request data for given variables, eventually while manipulating others. See the Info page for instructions on how to prepare queries.


Model Status Time dept. Number of variables Price per unit Training set Test set Budget
Target Observable Actionable Unobservable Sample Target Observation Manipulation Size Price Size Price

Note: The budget allows you to buy default training data and one test set. Buying the training set is optional. We show the price for the labeled training set. If you do not specify, you get it unlabeled, which is cheaper. The target values can be purchased separately with a SURVEY query. But you may be better off not buying all the labels of the training data and rather use some virtual cash to buy manipulations. Some savings to buy manipulations can also be made by observing a subset of the variables in test data.

The table below lists models we are currently working on.

Model Status Time dept. Number of variables Price per unit Training set Test set Budget
Target Observable Actionable Unobservable Sample Target Observation Manipulation Size Price Size Price
REGED
Announced
No
1
1000
1000
0




500

20000


MARTI
Announced
No
1
1025
1025
0




500

20000


PROMO
Announced
Yes
100
1100
1000
0




1095

365


TIED
Announced
No
1
1000
1000
0




750

3000


SIGNET
Announced Yes
1
43
43
0




6000

20000


MIDS
Announced Yes
0
9
9
0




10000

10000


SEFTI
Announced
Yes
1
602
602
0




4000

4000