A novel user-friendly Python/Jupyter interface for Inductive Logic programming(ILP) system for teaching relational machine learning and comparing different algorithms.
We have incoperated two different ILP algorithms in this packge
- Aleph
- Metagol
Aleph is an Inductive Logic Programming (ILP) system developed by Ashwin Srinivasan: http://www.cs.ox.ac.uk/activities/machlearn/Aleph/
In this package we use a porting of Aleph v.5 to SWI-Prolog prepared by Fabrizio Riguzzi: https://github.com/friguzzi/aleph
Metagol is an Meta-Interpretive Learning (MIL) system maintained by Andrew Cropper: https://github.com/metagol/metagol
The following fuction will help you to perfom hold-out test using aleph.
aleph_learn(file="BK.pl",
test_size=0.33,
positive_example="pos_example.f",
negative_example="neg_example.n",
shuffle=False,
settings=[])The following fuction will help you to perfom k-fold cross-validation using aleph.
aleph_cross_validation(file="BK.pl",
CV=2,
positive_example="pos_example.f",
negative_example="neg_example.n",
shuffle=False,
settings=[])The following fuction will help you to perfom k-fold hold-out test using metagol.
metagol_learn(file="BK.pl",
test_size=0.33,
positive_example="pos_example.f",
negative_example="neg_example.n",
shuffle=False,
settings=[])The following fuction will help you to perfom k-fold cross-validation using metagol.
metagol_cross_validation(file="BK.pl",
CV=2,
positive_example="pos_example.f",
negative_example="neg_example.n",
shuffle=False,
settings=[])| Parameter | Explanation |
|---|---|
| file |
Default : BK.pl Allows to input background information. You can include aleph parameters or metagol settings including meta-rules in this file, if so dont input the parameter settings |
| test_size |
Default : 0.33 should be between 0.0 and 1.0 and represent the proportion of the dataset to include in the test split in hold-out strategy |
| CV |
Default : 2 Determines the cross-validation splitting strategy |
| positive_example |
Default : pos_example.f Allows to input the positive example |
| negative_example |
Default : neg_example.n Allows to input the negative example |
| shuffle |
Default : False Whether or not to shuffle the data before splitting |
| settings |
Default : [] Allows to include aleph parameters or metagol settings including meta-rules as file format |
All the learning function will return following values ;
| Parameter | Explanation |
|---|---|
| hypothesis | Hyothesis Learned |
| accuracy |
Accuracy of learned hypothesis on test dataset or individual folds in the case of cross-validation |
| precision |
Precision of learned hypothesis on test dataset or individual folds in the case of cross-validation |
| sensitivity |
Sensitivity of learned hypothesis on test dataset or individual folds in the case of cross-validation |
| specificity |
Specificity of learned hypothesis on test dataset or individual folds in the case of cross-validation |
| fscore |
F-score of learned hypothesis on test dataset or individual folds in the case of cross-validation |
| time_learn | Learning Time |
The folders aleph_examples and metagol_examples contains Jypyter-Notebook files.
