View source for Examples from the lectures


''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:
<syntaxhighlight>
@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" .
</syntaxhighlight>

<!--
==Lecture 2: RDF==
Blank nodes for anonymity, or when we have not decided on a URI:
<syntaxhighlight lang="Python">
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'))
</syntaxhighlight>

Blank nodes used to group related properties:
<syntaxhighlight>
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'))
</syntaxhighlight>

Literals:
<syntaxhighlight>
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'))
</syntaxhighlight>

Alternative container (open):
<syntaxhighlight>
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'))
</syntaxhighlight>

Sequence container (open):
<syntaxhighlight>
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'))
</syntaxhighlight>

Collection (closed list):
<syntaxhighlight>
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'))
</syntaxhighlight>

==Lecture 3: SPARQL==

The KG4News knowledge graph can be accessed online [http://bg.newsangler.uib.no 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===
<syntaxhighlight lang="SPARQL">
SELECT ?p WHERE {
    ?s ?p ?o .
}
LIMIT 10</syntaxhighlight>

===List distinct properties only (with limit)===
<syntaxhighlight lang="SPARQL">
SELECT DISTINCT ?p WHERE {
    ?s ?p ?o .
}
LIMIT 10
</syntaxhighlight>

===Limit with offset===
<syntaxhighlight lang="SPARQL">
SELECT DISTINCT ?p WHERE {
    ?s ?p ?o .
}
LIMIT 10 OFFSET 9
</syntaxhighlight>

===List rdf:types===
<syntaxhighlight lang="SPARQL">
PREFIX rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>

SELECT DISTINCT ?t WHERE {
    ?s rdf:type ?t .
}
LIMIT 50
</syntaxhighlight>

===URI for Tim Berners-Lee===
<syntaxhighlight lang="SPARQL">
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
</syntaxhighlight>

===Describe Tim Berners-Lee===
<syntaxhighlight lang="SPARQL">
DESCRIBE <http://i2s.uib.no/kg4news/author/1432678629>
</syntaxhighlight>
DESCRIBE returns a new RDF graph, whereas SELECT returns a table of rows.

===Papers that mention "Semantic Web" in the title===
<syntaxhighlight lang="SPARQL">
PREFIX ss: <http://semanticscholar.org/>

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

===Alternative filter that ignores capitalisation (lower/upper case)===
<syntaxhighlight lang="SPARQL">
    FILTER(REGEX(STR(?title), "Semantic Web", "i"))
</syntaxhighlight>

===Authors sorted by name===
<syntaxhighlight lang="SPARQL">
PREFIX foaf: <http://xmlns.com/foaf/0.1/>

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

===Sorted by descending name instead===
<syntaxhighlight lang="SPARQL">
ORDER BY DESC(?name)
</syntaxhighlight>

===Count papers by author===
<syntaxhighlight lang="SPARQL">
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
</syntaxhighlight>

===Only list the most prolific authors===
<syntaxhighlight lang="SPARQL">
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
</syntaxhighlight>

===Order by descending paper count===
<syntaxhighlight lang="SPARQL">
SELECT ... {
    ...
}
GROUP BY ?person
HAVING (?count > 10)
ORDER BY DESC(?count)
LIMIT 10
</syntaxhighlight>

===Order by descending paper count and then by author name===
<syntaxhighlight lang="SPARQL">
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
</syntaxhighlight>

===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.
<syntaxhighlight lang="SPARQL">
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
</syntaxhighlight>

===Add one or more triples===
From now on you need a Blazegraph that allows writing, for example the [http://sandbox.i2s.uib.no I2S sandbox]. Remember to ''create a new namespace'' first and make sure you ''use'' it afterwards.
<syntaxhighlight lang="SPARQL">
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 . 
}
</syntaxhighlight>

===Remove one or more triples===
<syntaxhighlight lang="SPARQL">
PREFIX kg: <http://i2s.uib.no/kg4news/>

DELETE DATA
{
    kg:paper_123 kg:year 2022  .
}
</syntaxhighlight>

===Pattern-based addition and or removal of triples===
<syntaxhighlight lang="SPARQL">
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
}
</syntaxhighlight>

==Lecture 8: SHACL==
The examples are for use with the [https://shacl.org/playground/ interactive SHACL Playground].

===Minimal example===
First shape graph:
<syntaxhighlight lang="ttl">
@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 .
</syntaxhighlight>

First data graph:
<syntaxhighlight lang="ttl">
@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" .
</syntaxhighlight>
''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:
<syntaxhighlight lang="ttl">
kg:MainPaperShape
    a sh:NodeShape ;
    sh:targetClass kg:MainPaper ;
    sh:property [
        sh:path kg:year 
    ] .
</syntaxhighlight>
''This is equivalent to the previous example (no validation error).''

===Cardinality constraints on properties===
Add a minimum constraint to the ''kg:year'' property:
<syntaxhighlight lang="ttl">
kg:MainPaperShape
    a sh:NodeShape ;
    sh:targetClass kg:MainPaper ;
    sh:property [
        sh:path kg:year ;
        sh:minCount 1
    ] .
</syntaxhighlight>
''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:
<syntaxhighlight lang="ttl">
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
    ] .
</syntaxhighlight>
''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:
<syntaxhighlight lang="ttl">
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 .
</syntaxhighlight>

Or like this:
<syntaxhighlight lang="ttl">
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 
    ] .
</syntaxhighlight>
''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:
<syntaxhighlight lang="ttl">
    ...
    dcterm:contributor kg:TBL .

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

</syntaxhighlight>


===Full example===
Final shape graph from the lecture:
<syntaxhighlight lang="ttl">
@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 .
</syntaxhighlight>

Final data graph from the lecture:
<syntaxhighlight lang="ttl">
@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" .
</syntaxhighlight>

==Lecture 8: RDFS==
Create two new GraphDB Repositories, one with RDFS inference and one with No inference. Try the SPARQL statements on both, so you can compare them.

To test ''rdfs:subClassOf'':
<syntaxhighlight lang="ttl">
PREFIX kg: <http://i2s.uib.no/kg4news/>
PREFIX rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>
PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>

INSERT DATA { 
	kg:LOD_paper rdf:type kg:MainPaper .
    kg:MainPaper rdfs:subClassOf kg:Paper .
}
</syntaxhighlight>

<syntaxhighlight lang="ttl">
PREFIX kg: <http://i2s.uib.no/kg4news/>
PREFIX rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>
PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>

ASK {
	kg:LOD_paper rdf:type kg:Paper .
}
</syntaxhighlight>


To test ''rdfs:domain'':
<syntaxhighlight lang="ttl">
PREFIX kg: <http://i2s.uib.no/kg4news/>
PREFIX rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>
PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>

INSERT DATA { 
	kg:TBL kg:author kg:LOD_paper .
    kg:author rdfs:domain kg:Author .
}
</syntaxhighlight>

<syntaxhighlight lang="ttl">
PREFIX kg: <http://i2s.uib.no/kg4news/>
PREFIX rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>
PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>

ASK {
	kg:TBL rdf:type kg:Author .
}
</syntaxhighlight>


To see all the triples in the KG:
<syntaxhighlight lang="ttl">
SELECT * WHERE {
  ?s ?p ?o
}
</syntaxhighlight>

==Lecture 11: Graph embeddings==
Here is the example that used graph embeddings from ConceptNet. The ''numbernatch-en-19.08.txt.gz'' file can be [https://github.com/commonsense/conceptnet-numberbatch downloaded from GitHub].

Precomputed pickle files are [https://universityofbergen-my.sharepoint.com/:f:/g/personal/andreas_opdahl_uib_no/ElJv6Kl6zfJNmQrXMiz5eDoBogFpktjx8xoWgxOgAI70tw?e=99cClh available here (requires UiB login)].
<syntaxhighlight lang="python">
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'])
</syntaxhighlight>
-->