org.apache.solr.ltr.model

Class NeuralNetworkModel

java.lang.Object

org.apache.solr.ltr.model.LTRScoringModel

org.apache.solr.ltr.model.NeuralNetworkModel

All Implemented Interfaces:

Accountable

public class NeuralNetworkModel
extends LTRScoringModel

A scoring model that computes document scores using a neural network.

Supported activation functions are: identity, relu, sigmoid, tanh, leakyrelu and contributions to support additional activation functions are welcome.

Example configuration:

{
    "class" : "org.apache.solr.ltr.model.NeuralNetworkModel",
    "name" : "rankNetModel",
    "features" : [
        { "name" : "documentRecency" },
        { "name" : "isBook" },
        { "name" : "originalScore" }
    ],
    "params" : {
        "layers" : [
            {
                "matrix" : [ [ 1.0, 2.0, 3.0 ],
                             [ 4.0, 5.0, 6.0 ],
                             [ 7.0, 8.0, 9.0 ],
                             [ 10.0, 11.0, 12.0 ] ],
                "bias" : [ 13.0, 14.0, 15.0, 16.0 ],
                "activation" : "sigmoid"
            },
            {
                "matrix" : [ [ 17.0, 18.0, 19.0, 20.0 ],
                             [ 21.0, 22.0, 23.0, 24.0 ] ],
                "bias" : [ 25.0, 26.0 ],
                "activation" : "relu"
            },
            {
                "matrix" : [ [ 27.0, 28.0 ],
                             [ 29.0, 30.0 ] ],
                "bias" : [ 31.0, 32.0 ],
                "activation" : "leakyrelu"
            },
            {
                "matrix" : [ [ 33.0, 34.0 ],
                             [ 35.0, 36.0 ] ],
                "bias" : [ 37.0, 38.0 ],
                "activation" : "tanh"
            },
            {
                "matrix" : [ [ 39.0, 40.0 ] ],
                "bias" : [ 41.0 ],
                "activation" : "identity"
            }
        ]
    }
}

Training libraries:

• Keras is a high-level neural networks API, written in Python. A Keras and Solr implementation of RankNet can be found at https://github.com/airalcorn2/RankNet.

Background reading:

• C. Burges, T. Shaked, E. Renshaw, A. Lazier, M. Deeds, N. Hamilton, and G. Hullender. Learning to Rank Using Gradient Descent. Proceedings of the 22nd International Conference on Machine Learning (ICML), ACM, 2005.

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
protected static interface	NeuralNetworkModel.Activation
class	NeuralNetworkModel.DefaultLayer
static interface	NeuralNetworkModel.Layer

Field Summary

Fields inherited from class org.apache.solr.ltr.model.LTRScoringModel

features, name, norms

Constructor Summary

Constructors

Constructor and Description
NeuralNetworkModel(String name, List<Feature> features, List<Normalizer> norms, String featureStoreName, List<Feature> allFeatures, Map<String,Object> params)

Method Summary

Modifier and Type	Method and Description
protected NeuralNetworkModel.Layer	createLayer(Object o)
Explanation	explain(LeafReaderContext context, int doc, float finalScore, List<Explanation> featureExplanations) Similar to the score() function, except it returns an explanation of how the features were used to calculate the score.
float	score(float[] inputFeatures) Given a list of normalized values for all features a scoring algorithm cares about, calculate and return a score.
void	setLayers(Object layers)
protected void	validate() Validate that settings make sense and throws ModelException if they do not make sense.

Methods inherited from class org.apache.solr.ltr.model.LTRScoringModel

equals, getAllFeatures, getFeatures, getFeatureStoreName, getInstance, getName, getNormalizerExplanation, getNorms, getParams, hashCode, normalizeFeaturesInPlace, ramBytesUsed, toString

Methods inherited from class java.lang.Object

clone, finalize, getClass, notify, notifyAll, wait, wait, wait

Methods inherited from interface org.apache.lucene.util.Accountable

getChildResources

Constructor Detail

NeuralNetworkModel

public NeuralNetworkModel(String name,
                          List<Feature> features,
                          List<Normalizer> norms,
                          String featureStoreName,
                          List<Feature> allFeatures,
                          Map<String,Object> params)

Method Detail

createLayer

protected NeuralNetworkModel.Layer createLayer(Object o)

setLayers

public void setLayers(Object layers)

validate

protected void validate()
                 throws ModelException

Description copied from class: LTRScoringModel

Validate that settings make sense and throws ModelException if they do not make sense.

Overrides:

validate in class LTRScoringModel

Throws:

ModelException

score

public float score(float[] inputFeatures)

Description copied from class: LTRScoringModel

Given a list of normalized values for all features a scoring algorithm cares about, calculate and return a score.

Specified by:

score in class LTRScoringModel

Parameters:

inputFeatures - List of normalized feature values. Each feature is identified by its id, which is the index in the array

Returns:

The final score for a document

explain

public Explanation explain(LeafReaderContext context,
                           int doc,
                           float finalScore,
                           List<Explanation> featureExplanations)

Description copied from class: LTRScoringModel

Similar to the score() function, except it returns an explanation of how the features were used to calculate the score.

Specified by:

explain in class LTRScoringModel

Parameters:

context - Context the document is in

doc - Document to explain

finalScore - Original score

featureExplanations - Explanations for each feature calculation

Returns:

Explanation for the scoring of a document