See also our applications for complete products.
A command-line (i.e. type-into-a-prompt) application that allows you to experiment with
This consists of most of the functionality of Vizier, but without the graphical user interface support. It is based on new kd-tree technology for making the traditionally computationally demanding tasks of locally weighted regression and massive-scale cross validation tractable. It also uses new search algorithms for the blackbox search and memory-based experiment design.
Documentation on running GMBLMAIN.
A command-line (i.e. type-into-a-prompt) application identical to GMBLMAIN above except that it can also pop up graphics windows to show plots (line plots and contour plots) of fitted surfaces.
Documentation on running GGMBLMAIN.
DSETMAIN is an application that allows you to manipulate small or large machine learning datasets, and run decision tree learning algorithms on them. When you run the program on a dataset you arrive at an interactive command line application that lets you perform operations such as:
- Inspect details of numeric and symbolic attributes.
- "Pack" very large, slow-to-load, datasets into a very fast but system-independent file-format (called .fds) that loads 100x faster than regular ASCII datafiles and uses far less space.
- Change symbolic attributes to numeric, numeric to symbolic etc. by means of various transformations. This can be useful, for example, for preprocessing symbolic data into a form usable by Vizier, GGMBLMAIN or ADTREEMAIN.
- Delete unwanted attributes.
- Select subsets of the records, or split into random test and training sets.
- Run and experiment with Decision Tree learning algorithms.
The underlying datset data structure and data file formats are the underpinnings of most of the other Auton lab data mining and machine learning programs, and also machine learning exercises in Andrew's classes and tutorials. Eventually it will be integrated into Vizier, Q2 and GGMBLMAIN. Particular care has been put into compact representations of very large datasets that might otherwise overflow your machine's main memory.
Documentation on running DSETMAIN.
ADTREEMAIN is an application that allows you to play with and evaluate the ADTREE technology resulting from recent Auton Lab research. It includes algorithms for learning Bayesian Network structures very quickly from large datasets. It also includes a very fast Rule Learner and a computationally tractable exhaustive feature selection algorithm. It uses the underlying datset data representation and is fully compatible with the DSETMAIN program.
Q2MAIN: A simple interactive command line application that allows you to test and evaluate the Q2 algorithm for memory-based blackbox noisy optimization.
PGMAIN: Code to run and test the Partigame algorithm.
PAYNE: A new Reinforcement Learning algorithm for large discrete MDPs. Project work by Hunter Payne.
MOTCON: A new highly autonomous reinforcement learning/optimal control algorithm for optimal control of dynamic continuous-state-space systems, extending our earlier work on embedding A-star within reinforcement learning. Project work by Andy Arnt, Dan Halstead, Paul Hsiung and Nat Lanza.