phil

phil is a suite of algorithms for learning both the parameters and the structure of Hierarchical Probabilistic Logic Programs (HPLP) from data. The parameters are learned applying gradient descent (dphil) or Expectation Maximization (emphil). To perform structure leaning, phil initially generates a large HPLP and applies a regularized parameter learning on it. Then clauses with small values of probabilities are dropped.

Installation

This is a SWI-Prolog (http://www.swi-prolog.org/) pack.

It can be installed with pack_install/1

$ swipl
?- pack_install(phil).

The pack uses a foreign library and contains the library binaries for 32 and 64 bits Linux and 32 and 64 bits Windows. If you want to recompile the foreign library you can use

?- pack_rebuild(phil).

On 32 and 64 bits Linux this should work out of the box. On 32 and 64 bits Windows the library must be rebuilt by hand. First run pack_rebuild(phil). This typically fails but produces the file buildenv.sh in the root folder. You can modify this file looking at the example files buildenvmingw32.sh and buildenvmingw64.sh. Then you can run

$ source buildenv.sh
$ source configure
$ make install

Requirements

It requires packs auc and matrix:

$ swipl
?- pack_install(auc).
?- pack_install(matrix).

Example of use

Datasets are available in pack cplint_datasets (https://github.com/ArnaudFadja/phil_datasets) Install the phil_datasets with pack_install/1

$ swipl
?- pack_install(phil_datasets).

Then

$ cd <pack>/phil/prolog/
$ swipl
?- [bongard].
?- induce_hplp([train],P),test_hplp(P,[test],LL,AUCROC,ROC,AUCPR,PR).