ENH: Linked Datasets (RDF)

• This is very much a meta ticket.
• There are a number of bare links here.
• They are for documentation

(UPDATE: see westurner/pandasrdf#1)

Use Case

So I:

• retrieved some data
  • from somewhere
  • about a certain #topic
• perfomed analysis
  • with certain transformations and aggregations
  • with certain versions of certain tools
  • confirmed/rejected a [null] hypothesis

and I want to share my findings so that others can find, review, repeat, reproduce, and verify (confirm/reject) a given conclusion.

User Story

As a data analyst, I would like to share or publish Series, DataFrames, Panels, and Panel4Ds as structured, hierarchical, RDF linked data ("DataSet").

Status Quo: Pandas IO

How do I go from a [CSV] to a DataFrame to something shareable with a URL?

http://pandas.pydata.org/pandas-docs/dev/io.html

• http://pandas.pydata.org/pandas-docs/dev/dsintro.html
• https://github.com/pydata/pandas/blob/master/pandas/core/format.py
• https://github.com/pydata/pandas/blob/master/pandas/rpy/common.py

.

• Series (1D)
  • index
  • data
    • NumPy datatypes
• DataFrame (2D)
  • index
  • column(s)
    • NumPy datatypes
• Panel (3D)
• Panel4D (4D)

Read or parse a data format into a DataSet:

• pandas.read_*
  • read_clipboard
  • read_csv
  • read_excel
  • read_fwf
  • read_gbq
  • read_hdf
  • read_html
  • read_json
  • read_msgpack
  • read_pickle
  • read_sql
  • read_stata
  • read_table
• pandas.HDFStore
  • https://pandas.pydata.org/docs/dev/io.html#hdf5-pytables

Add metadata:

• Add RDF metadata (RDFa, JSONLD)

Save or serialize a DataSet into a data format:

• pandas.DataFrame.
  • to_csv
  • to_dict
  • to_excel
  • to_gbq
  • to_html
  • to_latex
  • to_panel
  • to_period
  • to_records
  • to_sparse
  • to_sql
  • to_stata
  • to_string
  • to_timestamp
  • to_wide
• to_ RDF
• to_ CSVW
• to_ HTML + RDFa
• to_ JSONLD
  • create a JSONLD @context

Share or publish a serialized DataSet with the internet:

• Email Attachment (Table in a PDF)
  • opendatahandbook.org
  • project-open-data.github.io
• FTP, SFTP, RSYNC, NFS
• HTML web upload form with metadata form fields
• CLI tool
• Version Control: Git, Hg, Svn
  • challenge: 'large' files ("binary blobs") in VCS systems
• HTTP API: Object Storage (~LDP)
  • GET/POST /container/filename.csv # [.json|.xml|.xls|.rdf|.html]
  • challenge: indexing metadata from a separate document / named graph
    • GET/POST to/container/filename.csv`
• Push to CKAN
  • http://docs.ckan.org/en/latest/api.html
  • http://docs.ckan.org/en/latest/linked-data-and-rdf.html
  • https://github.com/okfn/ckan
• Host DataSet metadata
  • python -m SimpleHTTPServer 8088
  • e.g. http://datasets.schema-labs.appspot.com/about Indexes http://schema.org/Dataset s

Implementation

What changes would be needed for Pandas core to support this workflow?

• .meta schema
• to_rdf for Series, DataFrames, Panels, and Panel4Ds
• read_rdf for Series, DataFrames, Panels, and Panel 4Ds
• ~@datastep process decorators
• ~DataSet
• ~DataCatalog of precomputed aggregations/views/slices.
• Units support (.meta?)

.meta schema

It's easy enough to serialize a dict and a table to naieve RDF.

For interoperability, it would be helpful to standardize with a common
set of terms/symbols/structures/schema for describing
the tabular, hierarchical data which pandas is designed to handle.

There is currently no standard method for storing columnar metadata
within Pandas (e.g. in .meta['columns'][colname]['schema'], or as a JSON-LD @context).

Ontology Resources

• http://www.w3.org/TR/rdf-schema/ (rdfs:)
• http://www.w3.org/TR/owl-overview/ (owl:)
• http://www.w3.org/TR/sparql11-query/#sparqlDefinition
• http://lov.okfn.org
• http://prefix.cc

CSV2RDF (csvw)

• http://www.w3.org/ns/csvw
• https://github.com/w3c/csvw
• https://w3c.github.io/csvw/

W3C PROV (prov:)

• http://www.w3.org/TR/prov-primer/#intuitive-overview-of-prov
• http://www.w3.org/TR/prov-o/
• http://www.w3.org/2011/prov/wiki/ProvImplementations
  • https://pypi.python.org/pypi/prov
  • http://prov.readthedocs.org/en/latest/usage.html

schema.org (schema:)

• http://schema.org
• http://www.w3.org/wiki/WebSchemas
• http://schema.rdfs.org/
• https://schema.org/docs/full.html :
  • schema:Dataset -- A body of structured information describing some topic(s) of interest.
    • [schema:Thing, schema:CreativeWork]
    • distribution -- A downloadable form of this dataset, at a specific location, in a specific format (DataDownload)
    • spatial, temporal
    • catalog -- A data catalog which contains a dataset (DataCatalog)
  • schema:DataCatalog -- collection of Datasets
    • [schema:Thing, schema:CreativeWork]
    • dataset -- A dataset contained in a catalog. (Dataset)
  • schema:DataDownload -- A dataset in downloadable form.
    • [schema:Thing, schema:CreativeWork]
    • contentSize
    • contentURL
    • uploadDate

W3C RDF Data Cube (qb:)

• http://www.w3.org/TR/vocab-data-cube/
• http://www.w3.org/2011/gld/wiki/Data_Cube_Vocabulary#The_history_of_Data_Cube.2C_SDMX-RDF_and_SCOVO
• http://www.w3.org/TR/vocab-data-cube/#vocab-reference :
  • qb:DataSet -- a collection of observations, possibly organized into various slices, conforming to some common dimensional structure
    • qb:Slice -- a subset of a DataSet defined by fixing a subset of the dimensional values.
  • qb:Observation -- a single observation in the cube, may have one or more associated measured values.
    • qb:dataset -- data set of which this observation is a part.
  • qb:ObservationGroup -- a, possibly arbitrary, group of observations.
    • qb:observation -- an observation contained within this slice of the data set.
  • qb:Slice -- a subset of a DataSet defined by fixing a subset of the dimensional values, component properties on the Slice.
  • [Components, Properties, Dimensions, Attributes, Measures]

to_rdf

http://pandas.pydata.org/pandas-docs/dev/io.html

Arguments:

• output fmt
• JSON-LD: compaction

.

• Series.meta
• Series.to_rdf()
• DataFrame.meta
• DataFrame.to_rdf()
• Panel.meta
• Panel.to_rdf()
• Panel4D.meta
• Panel4D.to_rdf()

read_rdf

http://pandas.pydata.org/pandas-docs/dev/remote_data.html

• Series.read_rdf()
• DataFrame.read_rdf()
• Panel.read_rdf()
• Panel4D.read_rdf()

Arguments to read_rdf would need to describe which dimensions of data to
read into 1D/2D/3D/4D form.

@datastep / PROV

• Objective: Additive journal of transformations
• Link to source script(s) URIs
• Decorator for annotating data transformations with metadata.
• Generate PROV metadata for data transformations

Ten Simple Rules for Reproducible Computational Research (3, 4, 5, 7, 8, 10)

DataCatalog

A collection of Datasets.

• DataCatalog = {that=df1, this=df1.group().apply(), also_this=df2]
  • 'this is an aggregation of that'
    • 'this' has a URI
    • 'that' has a URI
• What if there is no metadata for df2?

Units support

• Series.meta
• DataFrame.column.meta
• NumPy [, PyTables]
• http://docs.scipy.org/doc/numpy/reference/arrays.dtypes.html
• https://pint.readthedocs.org/en/latest/
• http://pythonhosted.org/quantities/

RDF Datatypes

• http://en.wikipedia.org/wiki/ISO_8601
• http://www.w3.org/TR/xmlschema-2/#decimal
• http://schema.org/Date
• http://schema.org/DateTime
• http://schema.org/Float
• http://schema.org/Quantity
• https://github.com/RDFLib/rdflib
  • from rdflib.namespace import XSD, RDF, RDFS
  • from rdflib import URIRef, Literal
  • https://github.com/RDFLib/rdflib-sqlalchemy (SQLAlchemy)

JSON-LD (RDF in JSON)

• https://github.com/digitalbazaar/pyld (JSON-LD)
• https://github.com/RDFLib/rdflib-jsonld (JSON-LD)

Linked Data Primer

Linked Data Abstractions

• Graphs are represented as triples of (s,p,o)
• Subject, Predicate, Object
• Queries are patterns with ?references
  • graph.triples((None, None, None))
  • SELECT ?s, ?p, ?o WHERE { ?s ?p ?o };
• subjects are linked to objects by predicates
  • subjects and predicate are identified by URI 'keys'

URIs and URLs

• a URI is like a URL
• usually, we expect URLs to be 'dereferencable` HTTP URIs
  • HTTP GET http://en.wikipedia.org/
• a URI may start with a different URI prefix
  • urn:
  • uuid:

SQL and Linked Data

• there exist standard mappings for whole SQL tablesets
  • rdb2rdf
  • similar to application scaffolding
  • ACL support adds complexity
• virtuoso supports SQL and RDF and SPARQL
  • standard mappings
  • virtuoso powers http://dbpedia.org/
    • dbpedia.org has a high degree of centrality
      • http://lod-cloud.net/
• rdflib-sqlalchemy maps RDF onto SQL tables
  • fairly inefficiently, when compared to native triplestores

Named Graphs

• Quads: (g, s, p, o)
• g: sometimes called the 'context' of a triple
• Metadata about GRAPH ?g
• Multiple named graphs in one file: TriX, TriG

Linked Data Formats

• NTriples
• RDF/XML
  • TriX
• Turtle, N3
  • TriG
• JSON-LD

Choosing Schema

• XSD, RDF, RDFS, DCTERMS
• Which schema is most popular?
• Which schema is a best fit for the data?
• Which schema will search engines index for us?
• What do the queries look like?
• Years Later... What is OWL?
• Why would we start with RDFS now?

Linked Data Process, Provenance, and Schema

DataSets have [implicit] URIs:

http://example.com/datasets/#<key> 

Shared or published DataSets have URLs:

http://ckan.example.org/datasets/<key> 

DataSets are about certain things:

e.g. URIs for #Tags, Categories, Taxonomy, Ontology 

DataSets are derived from somewhere, somehow:

• where and how was it downloaded? (digital sense)
• how was it collected? (process control sense)

Datasets have structure:

• Tabular, Hierarchical
• 1D, 2D, 3D, 4D
• Graph-based
  • Chains
  • Flows
• Schema

5 ★ Open Data
http://5stardata.info/
http://www.w3.org/TR/ld-glossary/#x5-star-linked-open-data

☆ Publish data on the Web in any format (e.g., PDF, JPEG) accompanied by an explicit Open License (expression of rights).
☆☆ Publish structured data on the Web in a machine-readable format (e.g., XML).
☆☆☆ Publish structured data on the Web in a documented, non-proprietary data format (e.g., CSV, KML).
☆☆☆☆ Publish structured data on the Web as RDF (eg Turtle, RDFa, JSON-LD, SPARQL)
☆☆☆☆☆ In your RDF, have the identifiers be links (URLs) to useful data sources.

https://en.wikipedia.org/wiki/Linked_Data