Examples from the lectures

From info216

This page currently shows the examples used in the Spring of 2023. It will be updated with examples from 2024 as the course progresses.

Lecture 1: Introduction to KGs

Turtle example:

@prefix ex: <http://example.org/> .
ex:Roger_Stone
    ex:name "Roger Stone" ;
    ex:occupation ex:lobbyist ;
    ex:significant_person ex:Donald_Trump .
ex:Donald_Trump
    ex:name "Donald Trump" .

Lecture 2: RDF

Blank nodes for anonymity, or when we have not decided on a URI:

from rdflib import Graph, Namespace, Literal, BNode, RDF, RDFS, DC, FOAF, XSD

EX = Namespace('http://example.org/')

g = Graph()
g.bind('ex', EX)  # this is why the line '@prefix ex: <http://example.org/> .'
                  # and the 'ex.' prefix are used when we print out Turtle later

robertMueller = BNode()
g.add((robertMueller, RDF.type, EX.Human))
g.add((robertMueller, FOAF.name, Literal('Robert Mueller', lang='en')))
g.add((robertMueller, EX.position_held, Literal('Director of the Federal Bureau of Investigation', lang='en')))

print(g.serialize(format='turtle'))

Blank nodes used to group related properties:

from rdflib import Graph, Namespace, Literal, BNode, RDF, RDFS, DC, FOAF, XSD

EX = Namespace('http://example.org/')

g = Graph()
g.bind('ex', EX)

# This is a task in Exercise 2

print(g.serialize(format='turtle'))

Literals:

from rdflib import Graph, Namespace, Literal, BNode, RDF, RDFS, DC, FOAF, XSD

EX = Namespace('http://example.org/')

g = Graph()
g.bind('ex', EX)

g.add((EX.Robert_Mueller, RDF.type, EX.Human))
g.add((EX.Robert_Mueller, FOAF.name, Literal('Robert Mueller', lang='en')))
g.add((EX.Robert_Mueller, FOAF.name, Literal('رابرت مولر', lang='fa')))
g.add((EX.Robert_Mueller, DC.description, Literal('sixth director of the FBI', datatype=XSD.string)))
g.add((EX.Robert_Mueller, EX.start_time, Literal(2001, datatype=XSD.integer)))

print(g.serialize(format='turtle'))

Alternative container (open):

from rdflib import Graph, Namespace, Literal, BNode, RDF, RDFS, DC, FOAF, XSD

EX = Namespace('http://example.org/')

g = Graph()
g.bind('ex', EX)

muellerReportArchives = BNode()
g.add((muellerReportArchives, RDF.type, RDF.Alt))

archive1 = 'https://archive.org/details/MuellerReportVolume1Searchable/' \
                    'Mueller%20Report%20Volume%201%20Searchable/'
archive2 = 'https://edition.cnn.com/2019/04/18/politics/full-mueller-report-pdf/index.html'
archive3 = 'https://www.politico.com/story/2019/04/18/mueller-report-pdf-download-text-file-1280891'

g.add((muellerReportArchives, RDFS.member, Literal(archive1, datatype=XSD.anyURI)))
g.add((muellerReportArchives, RDFS.member, Literal(archive2, datatype=XSD.anyURI)))
g.add((muellerReportArchives, RDFS.member, Literal(archive3, datatype=XSD.anyURI)))

g.add((EX.Mueller_Report, RDF.type, FOAF.Document))
g.add((EX.Mueller_Report, DC.contributor, EX.Robert_Mueller))
g.add((EX.Mueller_Report, SCHEMA.archivedAt, muellerReportArchives))

print(g.serialize(format='turtle'))

Sequence container (open):

from rdflib import Graph, Namespace, Literal, BNode, RDF, RDFS, DC, FOAF, XSD

EX = Namespace('http://example.org/')

g = Graph()
g.bind('ex', EX)

donaldTrumpSpouses = BNode()
g.add((donaldTrumpSpouses, RDF.type, RDF.Seq))
g.add((donaldTrumpSpouses, RDF._1, EX.IvanaTrump))
g.add((donaldTrumpSpouses, RDF._2, EX.MarlaMaples))
g.add((donaldTrumpSpouses, RDF._3, EX.MelaniaTrump))

g.add((EX.Donald_Trump, SCHEMA.spouse, donaldTrumpSpouses))

print(g.serialize(format='turtle'))

Collection (closed list):

from rdflib import Graph, Namespace, Literal, BNode, RDF, RDFS, DC, FOAF, XSD

EX = Namespace('http://example.org/')

g = Graph()
g.bind('ex', EX)

from rdflib.collection import Collection

g = Graph()
g.bind('ex', EX)

donaldTrumpSpouses = BNode()
Collection(g, donaldTrumpSpouses, [
    EX.IvanaTrump, EX.MarlaMaples, EX.MelaniaTrump
])
g.add((EX.Donald_Trump, SCHEMA.spouse, donaldTrumpSpouses))

print(g.serialize(format='turtle'))
g.serialize(destination='s02_Donald_Trump_spouses_list.ttl', format='turtle')

print(g.serialize(format='turtle'))

Lecture 3: SPARQL

The KG4News knowledge graph can be accessed online here (namespace kb) (read-only), or you can load the Turtle file into your own GraphDB repository. (Remember to save the file with .ttl extension. You can use http://i2s.uib.no/kg4news/ as base URI.)

Limit

SELECT ?p WHERE {
    ?s ?p ?o .
}
LIMIT 10

List distinct properties only (with limit)

SELECT DISTINCT ?p WHERE {
    ?s ?p ?o .
}
LIMIT 10

Limit with offset

SELECT DISTINCT ?p WHERE {
    ?s ?p ?o .
}
LIMIT 10 OFFSET 9

List rdf:types

PREFIX rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>

SELECT DISTINCT ?t WHERE {
    ?s rdf:type ?t .
}
LIMIT 50

URI for Tim Berners-Lee

PREFIX foaf: <http://xmlns.com/foaf/0.1/>

SELECT DISTINCT ?person WHERE {
    ?person foaf:name ?name .  
 
    FILTER(CONTAINS(?name, "Berners-Lee"))
}
LIMIT 10  # best to use limit if something goes wrong

Describe Tim Berners-Lee

DESCRIBE <http://i2s.uib.no/kg4news/author/1432678629>

DESCRIBE returns a new RDF graph, whereas SELECT returns a table of rows.

Papers that mention "Semantic Web" in the title

PREFIX ss: <http://semanticscholar.org/>

SELECT DISTINCT ?paper ?title WHERE {
    ?paper ss:title ?title .  
 
    FILTER(CONTAINS(STR(?title), "Semantic Web"))
}
LIMIT 50

Alternative filter that ignores capitalisation (lower/upper case)

    FILTER(REGEX(STR(?title), "Semantic Web", "i"))

Authors sorted by name

PREFIX foaf: <http://xmlns.com/foaf/0.1/>

SELECT DISTINCT * WHERE {
    ?author foaf:name ?name .  
}
ORDER BY ?name
LIMIT 10

Sorted by descending name instead

ORDER BY DESC(?name)

Count papers by author

PREFIX rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>
PREFIX dct: <http://purl.org/dc/terms/>
PREFIX ss: <http://semanticscholar.org/>

SELECT DISTINCT ?author (COUNT(?paper) AS ?count) WHERE {
    ?author rdf:type ss:Author .  
    ?paper  rdf:type ss:Paper ;
            dct:contributor ?author .
}
GROUP BY ?author
LIMIT 10

Only list the most prolific authors

PREFIX rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>
PREFIX dct: <http://purl.org/dc/terms/>
PREFIX ss: <http://semanticscholar.org/>

SELECT DISTINCT ?author (COUNT(?paper) AS ?count) WHERE {
    ?author rdf:type ss:Author .  
    ?paper  rdf:type ss:Paper ;
            dct:contributor ?author .
}
GROUP BY ?author
HAVING (?count >= 10)  # similar to a filter expression
LIMIT 10               # include limit when you test

Order by descending paper count

SELECT ... {
    ...
}
GROUP BY ?person
HAVING (?count > 10)
ORDER BY DESC(?count)
LIMIT 10

Order by descending paper count and then by author name

PREFIX rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>
PREFIX dct: <http://purl.org/dc/terms/>
PREFIX foaf: <http://xmlns.com/foaf/0.1/>
PREFIX ss: <http://semanticscholar.org/>

SELECT DISTINCT ?person (SAMPLE(?name) AS ?name) (COUNT(?paper) AS ?count) WHERE {
    ?person rdf:type ss:Author ;
            foaf:name ?name .  
    ?paper  rdf:type ss:Paper ;
            ss:title ?title ;
            dct:contributor ?person .
}
GROUP BY ?person
HAVING (?count > 10)
ORDER BY DESC(?count)
LIMIT 10

Embedded Wikidata query

This is a toy example only. Embedded queries like these are better suited for situations where the same URIs are used in more than one triple store and you want to combine data. But Wikidata and KG4News do not use the same URIs. So instead, the example searches for similar labels, and this is something graph databases may not be optimised for. Moreover, Wikidata uses language-tagged strings whereas KG4News uses plain strings, so the labels cannot even be directly compared.

PREFIX foaf: <http://xmlns.com/foaf/0.1/>
PREFIX skos: <http://www.w3.org/2004/02/skos/core#>

SELECT DISTINCT ?enname ?person ?wdperson WHERE {
    BIND("T. Berners-Lee"@en AS ?enname)  # language-tagger label

    SERVICE <https://query.wikidata.org/bigdata/namespace/wdq/sparql> {
        # return a Wikidata identifier (URI) with this label as alternative
        SELECT ?wdperson ?enname WHERE {
             ?wdperson skos:altLabel ?enname .
        }
        LIMIT 10  # we use limit in case the label does not match
    }
            
    BIND(STR(?enname) AS ?name)  # the same label, but with language tag removed
    # return a KG4News identifier (URI) with this label as name
    ?person foaf:name ?name .  

}
LIMIT 10

Add one or more triples

From now on you need a Blazegraph that allows writing, for example the I2S sandbox. Remember to create a new namespace first and make sure you use it afterwards.

PREFIX rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>
PREFIX dct: <http://purl.org/dc/terms/>
PREFIX kg: <http://i2s.uib.no/kg4news/>
PREFIX ss: <http://semanticscholar.org/>

INSERT DATA {  # note the Turtle-like syntax
    kg:paper_123 rdf:type ss:Paper ;
            ss:title "Semantic Knowledge Graphs for the News: A Review"@en ;
            kg:year 2022 ;
            dct:contributor kg:auth_456, kg:auth_789 . 
}

Remove one or more triples

PREFIX kg: <http://i2s.uib.no/kg4news/>

DELETE DATA
{
    kg:paper_123 kg:year 2022  .
}

Pattern-based addition and or removal of triples

PREFIX rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>
PREFIX dct: <http://purl.org/dc/terms/>
PREFIX kg: <http://i2s.uib.no/kg4news/>
PREFIX ss: <http://semanticscholar.org/>

DELETE DATA {    
    ?paper dct:contributor kg:auth_456
}
INSERT DATA {    
    ?paper dct:contributor kg:auth_654
}
WHERE {  # the patterns are similar to SELECT patterns
    ?paper dct:contributor kg:auth_456
}

Lecture 7: SHACL

The examples are for use with the interactive SHACL Playground.

Minimal example

First shape graph:

@prefix sh: <http://www.w3.org/ns/shacl#> .
@prefix dcterm: <http://purl.org/dc/terms/> .
@prefix foaf: <http://xmlns.com/foaf/0.1/> .
@prefix skos: <http://www.w3.org/2004/02/skos/core#> .
@prefix xsd: <http://www.w3.org/2001/XMLSchema#> .
@prefix kg: <http://i2s.uib.no/kg4news/> .
@prefix th: <http://i2s.uib.no/kg4news/theme/> .
@prefix ss: <http://semanticscholar.org/> .


kg:MainPaperShape
    a sh:NodeShape ;
    sh:targetClass kg:MainPaper ;
    sh:property kg:MainPaperYearShape  .

kg:MainPaperYearShape
        sh:path kg:year .

First data graph:

@prefix sh: <http://www.w3.org/ns/shacl#> .
@prefix dcterm: <http://purl.org/dc/terms/> .
@prefix foaf: <http://xmlns.com/foaf/0.1/> .
@prefix skos: <http://www.w3.org/2004/02/skos/core#> .
@prefix xsd: <http://www.w3.org/2001/XMLSchema#> .
@prefix kg: <http://i2s.uib.no/kg4news/> .
@prefix th: <http://i2s.uib.no/kg4news/theme/> .
@prefix ss: <http://semanticscholar.org/> .


kg:LOD_Paper
    a kg:MainPaper ;
    dcterm:title "Linked Data - The Story so Far" .

This should not give a validation error.

Alternative shape graph

Keep the prefixes from the first examples. You can also write the property constraint as an anyonymous node like this:

kg:MainPaperShape
    a sh:NodeShape ;
    sh:targetClass kg:MainPaper ;
    sh:property [
        sh:path kg:year 
    ] .

This is equivalent to the previous example (no validation error).

Cardinality constraints on properties

Add a minimum constraint to the kg:year property:

kg:MainPaperShape
    a sh:NodeShape ;
    sh:targetClass kg:MainPaper ;
    sh:property [
        sh:path kg:year ;
        sh:minCount 1
    ] .

Now you should get a validation error.

  • In the data graph, add two kg:year properties (e.g., 2008 and 2009) to kg_LOD_Paper to get rid of the error.
  • In the shapes graph, add a sh:maxCount 1 constraint to get another validation error.
  • In the data graph, remove one kg:year property value from kg:LOD_Paper to get rid of the error (2009 is the right year).

Datatype constraint on literal property values

Add the following property constraint to the previous example:

kg:MainPaperShape
    a sh:NodeShape ;
    sh:targetClass kg:MainPaper ;
    sh:property [
        sh:path kg:year ;
        sh:minCount 1 ;
        sh:minCount 1 ;
        sh:datatype xsd:integer
    ] .

This should not give a validation error.

  • In the shapes graph, change the datatype constraint to sh:datatype xsd:year. Now you should get an error.
  • In the data graph, change the integer 2009 to the typed value "2009"^^xsd:year to get rid of the error.

Class and node kind constraints

Add the following property constraints to the shape graph, either with a URI like this:

kg:MainPaperShape
    a sh:NodeShape ;
    sh:targetClass kg:MainPaper ;
    sh:property kg:MainPaperYearShape, MainPaperContributorShape  

...

kg:MainPaperContributorShape
        sh:path dcterm:contributor ;
        sh:minCount 1 ;
        sh:class kg:MainAuthor ;
        sh:nodeKind sh:IRI .

Or like this:

kg:MainPaperShape
    a sh:NodeShape ;
    sh:targetClass kg:MainPaper ;
    sh:property [
        sh:path kg:year ;
        sh:minCount 1 ;
        sh:maxCount 1 ;
        sh:datatype xsd:year 
    ], [
        sh:path dcterm:contributor ;
        sh:minCount 1 ;
        sh:class kg:MainAuthor ;
        sh:nodeKind sh:IRI 
    ] .

Either way should give you a validation error.

  • In the data graph, add dcterm:contributor "T. Berners-Lee" to get rid of the cardinality error. The error goes away, but you get two new ones instead.
  • In the data graph, add dcterm:contributor [ a kg:MainAuthor ; foaf:name "T. Berners-Lee" ] to get rid of the error. The class error goes away, but the IRI error remains.
  • In the data graph, create an IRI for Tim-Berners Lee to resolve the error:
    ...
    dcterm:contributor kg:TBL .

kg:TBL
    a kg:MainAuthor;
    foaf:name "T. Berners-Lee" .


Full example

Final shape graph from the lecture:

@prefix sh: <http://www.w3.org/ns/shacl#> .
@prefix dcterm: <http://purl.org/dc/terms/> .
@prefix foaf: <http://xmlns.com/foaf/0.1/> .
@prefix skos: <http://www.w3.org/2004/02/skos/core#> .
@prefix xsd: <http://www.w3.org/2001/XMLSchema#> .
@prefix kg: <http://i2s.uib.no/kg4news/> .
@prefix th: <http://i2s.uib.no/kg4news/theme/> .
@prefix ss: <http://semanticscholar.org/> .


kg:MainPaperShape
    a sh:NodeShape ;
    sh:targetClass kg:MainPaper ;
    sh:property kg:MainPaperYearShape, kg:MainPaperContributorShape .

kg:MainPaperYearShape
        sh:path kg:year ;
        sh:minCount 1 ;
        sh:maxCount 1 ;
        sh:datatype xsd:year .

kg:MainPaperContributorShape
        sh:path dcterm:contributor ;
        sh:minCount 1 ;
        sh:class kg:MainAuthor ;
        sh:nodeKind sh:IRI .

Final data graph from the lecture:

@prefix sh: <http://www.w3.org/ns/shacl#> .
@prefix dcterm: <http://purl.org/dc/terms/> .
@prefix foaf: <http://xmlns.com/foaf/0.1/> .
@prefix skos: <http://www.w3.org/2004/02/skos/core#> .
@prefix xsd: <http://www.w3.org/2001/XMLSchema#> .
@prefix kg: <http://i2s.uib.no/kg4news/> .
@prefix th: <http://i2s.uib.no/kg4news/theme/> .
@prefix ss: <http://semanticscholar.org/> .


kg:LOD_Paper
    a kg:MainPaper ;
    dcterm:title "Linked Data - The Story so Far" ;
    kg:year "2006"^^xsd:year ;
    dcterm:contributor kg:TBL, kg:CB .

kg:TBL
    a kg:MainAuthor;
    foaf:name "T. Berners-Lee" .
 
kg:CB
     a kg:MainAuthor;
     foaf:name "C. Bizer" .

Lecture 7: RDFS

Create a new Blazegraph Namespace with inference and Use it.

rdfs:subClassOf

In the Update tab, insert the Final data graph from the SHACL example above. (Select Type: RDF data and Format: Turtle.)

Go to the Query tab, and SELECT all papers of type kg:MainPaper.

PREFIX rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>
PREFIX sh: <http://www.w3.org/ns/shacl#>
PREFIX dcterm: <http://purl.org/dc/terms/>
PREFIX foaf: <http://xmlns.com/foaf/0.1/>
PREFIX skos: <http://www.w3.org/2004/02/skos/core#>
PREFIX xsd: <http://www.w3.org/2001/XMLSchema#>
PREFIX kg: <http://i2s.uib.no/kg4news/>
PREFIX th: <http://i2s.uib.no/kg4news/theme/>
PREFIX ss: <http://semanticscholar.org/>

SELECT ?paper WHERE {
    ?paper rdf:type kg:MainPaper
}

Change the query to select all papers first of type ss:Paper and then of type foaf:Document.

In the Update tab:

kg:MainPaper rdfs:subClassOf ss:Paper .

Back in the Query tab, rerun the query to select all papers first of type ss:Paper and then of type foaf:Document.

Transitive rdfs:subClassOf

In the Update tab, add the following RDFS rule:

ss:Paper rdfs:subClassOf foaf:Document .

Back in the Query tab, rerun the query to select all papers first of type ss:Paper and then of type foaf:Document.

rdfs:range

In the Update tab, add another paper, which is cited (referenced) by the LOD paper:

kg:SW_Paper 
    dcterm:title "The Semantic Web" ;
    dcterm:contributor kg:TBL, kg:JH, kg:OL .

kg:JH
    foaf:name "J. Hendler" .

kg:OL
    foaf:name "O. Lassila" .

kg:LOD_Paper kg:cites kg:SW_Paper  # the new SW paper is cited by the LOD paper

Note that we have not stated the RDF types of the new paper and new authors.

In the Query tab, rerun the SPARQL queries to check that kg:SW_Paper has no rdf:type.

In the Update tab add this RDFS rule:

kg:cites rdfs:range ss:Paper .  # any paper that is cited by a kg:MainPaper is an ss:Paper

Back in the Query tab, rerun the SPARQL queries to check that the new SW paper is now an ss:Paper and foaf:Document, but not a kg:MainPaper.

Lecture 10: OWL-DL

Here is the OWL file with the minimal Protégé / HermiT example from S10. Save it with the .owl suffix, for example s10-example.owl, and File -> Open it in Protégé.

<?xml version="1.0"?>
<rdf:RDF xmlns="http://www.semanticweb.org/sinoa/ontologies/2023/2/untitled-ontology-3#"
     xml:base="http://www.semanticweb.org/sinoa/ontologies/2023/2/untitled-ontology-3"
     xmlns:owl="http://www.w3.org/2002/07/owl#"
     xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"
     xmlns:xml="http://www.w3.org/XML/1998/namespace"
     xmlns:xsd="http://www.w3.org/2001/XMLSchema#"
     xmlns:rdfs="http://www.w3.org/2000/01/rdf-schema#"
     xmlns:untitled-ontology-3="http://www.semanticweb.org/sinoa/ontologies/2023/2/untitled-ontology-3#">
    <owl:Ontology rdf:about="http://www.semanticweb.org/sinoa/ontologies/2023/2/untitled-ontology-3"/>

    <!-- 
    ///////////////////////////////////////////////////////////////////////////////////////
    //
    // Object Properties
    //
    ///////////////////////////////////////////////////////////////////////////////////////
     -->

    <!-- http://www.semanticweb.org/sinoa/ontologies/2023/2/untitled-ontology-3#hasSource -->
    <owl:ObjectProperty rdf:about="http://www.semanticweb.org/sinoa/ontologies/2023/2/untitled-ontology-3#hasSource"/>

    <!-- 
    ///////////////////////////////////////////////////////////////////////////////////////
    //
    // Classes
    //
    ///////////////////////////////////////////////////////////////////////////////////////
     -->

    <!-- http://www.semanticweb.org/sinoa/ontologies/2023/2/untitled-ontology-3#Content -->
    <owl:Class rdf:about="http://www.semanticweb.org/sinoa/ontologies/2023/2/untitled-ontology-3#Content"/>

    <!-- http://www.semanticweb.org/sinoa/ontologies/2023/2/untitled-ontology-3#Source -->
    <owl:Class rdf:about="http://www.semanticweb.org/sinoa/ontologies/2023/2/untitled-ontology-3#Source">
        <owl:equivalentClass>
            <owl:Restriction>
                <owl:onProperty>
                    <rdf:Description>
                        <owl:inverseOf rdf:resource="http://www.semanticweb.org/sinoa/ontologies/2023/2/untitled-ontology-3#hasSource"/>
                    </rdf:Description>
                </owl:onProperty>
                <owl:someValuesFrom rdf:resource="http://www.semanticweb.org/sinoa/ontologies/2023/2/untitled-ontology-3#Content"/>
            </owl:Restriction>
        </owl:equivalentClass>
    </owl:Class>

    <!-- 
    ///////////////////////////////////////////////////////////////////////////////////////
    //
    // Individuals
    //
    ///////////////////////////////////////////////////////////////////////////////////////
     -->

    <!-- http://www.semanticweb.org/sinoa/ontologies/2023/2/untitled-ontology-3#NYTimes -->
    <owl:NamedIndividual rdf:about="http://www.semanticweb.org/sinoa/ontologies/2023/2/untitled-ontology-3#NYTimes"/>

    <!-- http://www.semanticweb.org/sinoa/ontologies/2023/2/untitled-ontology-3#article_about_Trump_rally -->
    <owl:NamedIndividual rdf:about="http://www.semanticweb.org/sinoa/ontologies/2023/2/untitled-ontology-3#article_about_Trump_rally">
        <rdf:type rdf:resource="http://www.semanticweb.org/sinoa/ontologies/2023/2/untitled-ontology-3#Content"/>
        <hasSource rdf:resource="http://www.semanticweb.org/sinoa/ontologies/2023/2/untitled-ontology-3#NYTimes"/>
    </owl:NamedIndividual>
</rdf:RDF>

<!-- Generated by the OWL API (version 4.5.9.2019-02-01T07:24:44Z) https://github.com/owlcs/owlapi -->

Lecture 11: Graph embeddings

Here is the example that used graph embeddings from ConceptNet. The numbernatch-en-19.08.txt.gz file can be downloaded from GitHub.

Precomputed pickle files are available here (requires UiB login).

import pickle

import pandas as pd
import numpy as np
from sklearn.neighbors import NearestNeighbors


NUMBERBATCH_FILE= './conceptnet/numberbatch-en-19.08.txt'
EMBEDDINGS_FILE = f'./conceptnet/numberbatch-embeddings.pickle'
N_NEIGHBOURS = 5
NUMBERBATCH_NEIGHBOURS_FILE = f'./conceptnet/numberbatch-neighbours-{N_NEIGHBOURS}.pickle'


# load numberbatch
reload = False  # set False first time you run, or if you already have the pickle file
if not reload:
    embeddings = {}
    with open(NUMBERBATCH_FILE) as file:
        n_terms, n_dims = map(int, file.readline().strip().split(' '))
        for line in file: 
            splits = line.strip().split(' ')
            embeddings[splits[0]] = np.array(list(map(float, splits[1:])))
    with open(EMBEDDINGS_FILE, 'wb') as file:
        pickle.dump(embeddings, file)
else:
    with open(EMBEDDINGS_FILE, 'rb') as file:
        embeddings = pickle.load(file)

# find nearest neighbours
reload = False  # set False first time you run, or if you already have the pickle file
if not reload:
    knn = NearestNeighbors(n_neighbors=N_NEIGHBOURS, algorithm='ball_tree')
    np_embeddings = np.array(list(embeddings.values()))
    knn.fit(np_embeddings)
    with open(NUMBERBATCH_NEIGHBOURS_FILE, 'wb') as file:
        pickle.dump(knn, file)
else:
    with open(NUMBERBATCH_NEIGHBOURS_FILE, 'rb') as file:
        knn = pickle.load(file)
     

# test
distances, indexes = knn.kneighbors([embeddings['bergen']])
for dist, idx in zip(distances[0], indexes[0]):
    print(f'{dist}:\t{list(embeddings.keys())[idx]}')


def vector_neighbours(vector):
    distances, indexes = knn.kneighbors([vector])
    for dist, idx in zip(distances[0], indexes[0]):
        print(f'{dist}:\t{list(embeddings.keys())[idx]}')


vector_neighbours(embeddings['oslo'] - embeddings['norway'] + embeddings['france'])