This is an attempt to resuscitate the Alchemy2 project.
Alchemy 2.0 includes the following algorithms from the original Alchemy system:
- Discriminative weight learning (Voted Perceptron, Conjugate Gradient, and Newton's Method)
- Generative weight learning
- Structure learning
- Propositional MAP/MPE inference (including memory efficient)
- Propositional and lazy Probabilistic inference algorithms: MC-SAT, Gibbs Sampling and Simulated Tempering
- Lifted Belief propagation
- Support for native and linked-in functions
- Block inference and learning over variables with mutually exclusive and exhaustive values
- EM (to handle ground atoms with unknown truth values during learning)
- Specification of indivisible formulas (i.e. formulas that should not be broken up into separate clauses)
- Support of continuous features and domains
- Online inference
- Decision Theory
The key new feature of Alchemy 2.0 is lifted inference algorithms (both exact and sampling-based). Specifically, it includes the following inference algorithms:
- Probabilistic theorem proving (lifted weighted model counting)
- Lifted importance sampling
- Lifted Gibbs sampling
By using Alchemy, you agree to accept the license agreement in license.txt
src/ contains source code and a makefile. doc/ contains a change log, and a manual in PDF, PostScript and html formats. exdata/ contains a simple example of Alchemy input files. bin/ is used to contain compiled executables.
Please refer to the change log at http://alchemy.cs.washington.edu/ for the latest changes to Alchemy.