srlearn.rdn.BoostedRDNRegressor¶

class srlearn.rdn.BoostedRDNRegressor(background=None, target='None', n_estimators=10, node_size=2, max_tree_depth=3, neg_pos_ratio=2, solver='BoostSRL')[source]¶

Relational Dependency Networks Regressor

Wrappers around BoostSRL for learning and inference of RDNs for regression task.

Similar to sklearn.ensemble.GradientBoostingRegressor, this builds a model by fitting a series of regression trees.

Examples

>>> from srlearn.rdn import BoostedRDNRegressor
>>> from srlearn import Background
>>> from srlearn import Database
>>> train = Database.from_files(
...     pos="../datasets/Boston/train/pos.pl",
...     neg="../datasets/Boston/train/neg.pl",
...     facts="../datasets/Boston/train/facts.pl",
...     lazy_load=False,
... )
>>> test = Database.from_files(
...     pos="../datasets/Boston/test/pos.pl",
...     neg="../datasets/Boston/test/neg.pl",
...     facts="../datasets/Boston/test/facts.pl",
...     lazy_load=False,
... )
>>> train.modes = ["crim(+id,#varsrim).",
...     "zn(+id,#varzn).",
...     "indus(+id,#varindus).",
...     "chas(+id,#varchas).",
...     "nox(+id,#varnox).",
...     "rm(+id,#varrm).",
...     "age(+id,#varage).",
...     "dis(+id,#vardis).",
...     "rad(+id,#varrad).",
...     "tax(+id,#vartax).",
...     "ptratio(+id,#varptrat).",
...     "b(+id,#varb).",
...     "lstat(+id,#varlstat).",
...     "medv(+id)."]
>>> bk = Background(modes=train.modes)
>>> reg = BoostedRDNRegressor(background=bk, target="medv", n_estimators=5)
>>> reg.fit(train)
BoostedRDNRegressor(background=setParam: numOfClauses=100.
setParam: numOfCycles=100.
usePrologVariables: true.
setParam: nodeSize=2.
setParam: maxTreeDepth=3.
mode: crim(+id,#varsrim).
mode: zn(+id,#varzn).
mode: indus(+id,#varindus).
mode: chas(+id,#varchas).
mode: nox(+id,#varnox).
mode: rm(+id,#varrm).
mode: age(+id,#varage).
mode: dis(+id,#vardis).
mode: rad(+id,#varrad).
mode: tax(+id,#vartax).
mode: ptratio(+id,#varptrat).
mode: b(+id,#varb).
mode: lstat(+id,#varlstat).
mode: medv(+id).
, n_estimators=5, neg_pos_ratio=2, solver='BoostSRL', target='medv')
>>> reg.predict(test)   # doctest: +SKIP
array([10.04313307 13.55804603 20.549378   18.14681934 23.9393469  10.01292162
     29.83298024 20.34668817 27.81642572 32.04067867  9.41342835 20.975001
     19.21966845])
__init__(background=None, target='None', n_estimators=10, node_size=2, max_tree_depth=3, neg_pos_ratio=2, solver='BoostSRL')[source]¶

Initialize a BoostedRDN

Parameters:
background : srlearn.background.Background (default: None)

Background knowledge with respect to the database

target : str (default: “None”)

Target predicate to learn

n_estimators : int, optional (default: 10)

Number of trees to fit

node_size : int, optional (default: 2)

Maximum number of literals in each node.

max_tree_depth : int, optional (default: 3)

Maximum number of nodes from root to leaf (height) in the tree.

neg_pos_ratio : int or float, optional (default: 2)

Ratio of negative to positive examples used during learning.
Attributes:
estimators_ : array, shape (n_estimators)

Return the boosted regression trees

feature_importances_ : array, shape (n_features)

Return the feature importances (based on how often each feature appears)
feature_importances_¶

Return the features contained in a tree.

Parameters:
tree_number: int

Index of the tree to read.
fit(database)[source]¶

Learn structure and parameters.

Fit the structure and parameters of a Relational Dependency Network using a database of positive examples, negative examples, facts, and any relevant background knowledge.

Parameters:
database : srlearn.database.Database

Database containing examples and facts.
Returns:
self : object

Returns self.

Notes

The underlying algorithm is based on the “Relational Functional Gradient Boosting” as described in [1] and [2]_.

This fit function is based on subprocess calling the BoostSRL jar files. This will require a Java runtime to also be available. See [R81fc0061240d-3].

[1]Sriraam Natarajan, Tushar Khot, Kristian Kersting, and Jude Shavlik, “Boosted Statistical Relational Learners: From Benchmarks to Data-Driven Medicine”. SpringerBriefs in Computer Science, ISBN: 978-3-319-13643-1, 2015

– [2] https://starling.utdallas.edu/software/boostsrl/wiki/regression/ .. [R81fc0061240d-3] https://starling.utdallas.edu/software/boostsrl/

from_json(file_name)¶

Load a learned model from json.

Parameters:
file_name : str (or pathlike)

Path to a saved json file.
predict(database)[source]¶

Use the learned model to predict values on new data.

Parameters:
database : srlearn.Database

Database containing examples and facts.
Returns:
pred : ndarray

regression value predicted for each example.
to_json(file_name) → None¶

Serialize a learned model to json.

Parameters:
file_name : str (or pathlike)

Path to a saved json file.