Pycldf
python package to read and write CLDF datasets
Install / Use
/learn @cldf/PycldfREADME
pycldf
A python package to read and write CLDF datasets.
Install
Install pycldf from PyPI:
pip install pycldf
Command line usage
Installing the pycldf package will also install a command line interface cldf, which provides some sub-commands to manage CLDF datasets.
Dataset discovery
cldf subcommands support dataset discovery as specified in the standard.
So a typical workflow involving a remote dataset could look as follows.
Create a local directory to which to download the dataset (ideally including version info):
$ mkdir wacl-1.0.0
Accessing CLDF datasets on Zenodo requires installing cldfzenodo (via pip install cldfzenodo).
Validating a dataset from Zenodo will implicitly download it, so running
$ cldf validate https://zenodo.org/record/7322688#rdf:ID=wacl --download-dir wacl-1.0.0/
will download the dataset to wacl-1.0.0.
Subsequently we can access the data locally for better performance:
$ cldf stats wacl-1.0.0/#rdf:ID=wacl
<cldf:v1.0:StructureDataset at wacl-1.0.0/cldf>
value
------------------------ --------------------------------------------------------------------
dc:bibliographicCitation Her, One-Soon, Harald Hammarström and Marc Allassonnière-Tang. 2022.
dc:conformsTo http://cldf.clld.org/v1.0/terms.rdf#StructureDataset
dc:identifier https://wacl.clld.org
dc:license https://creativecommons.org/licenses/by/4.0/
dc:source sources.bib
dc:title World Atlas of Classifier Languages
dcat:accessURL https://github.com/cldf-datasets/wacl
rdf:ID wacl
rdf:type http://www.w3.org/ns/dcat#Distribution
Type Rows
-------------- -------------- ------
values.csv ValueTable 3338
parameters.csv ParameterTable 1
languages.csv LanguageTable 3338
codes.csv CodeTable 2
sources.bib Sources 2000
(Note that locating datasets on Zenodo requires installation of cldfzenodo.)
Summary statistics
$ cldf stats tests/data/wordlist_with_cognates/metadata.json
<cldf:v1.0:Wordlist at tests/data/wordlist_with_cognates>
value
------------- --------------------------------------------
dc:conformsTo http://cldf.clld.org/v1.0/terms.rdf#Wordlist
dc:source sources.bib
Type Rows
--------------- --------------- ------
languages.csv LanguageTable 2
parameters.csv ParameterTable 2
forms.csv FormTable 3
cognates.csv CognateTable 2
cognatesets.csv CognatesetTable 1
sources.bib Sources 1
Validation
Arguably the most important functionality of pycldf is validating CLDF datasets.
By default, data files are read in strict-mode, i.e. invalid rows will result in an exception being raised. To validate a data file, it can be read in validating-mode.
For example the following output is generated
$ cldf validate mydataset/forms.csv
WARNING forms.csv: duplicate primary key: (u'1',)
WARNING forms.csv:4:Source missing source key: Mei2005
when reading the file
ID,Language_ID,Parameter_ID,Value,Segments,Comment,Source
1,abcd1234,1277,word,,,Meier2005[3-7]
1,stan1295,1277,hand,,,Meier2005[3-7]
2,stan1295,1277,hand,,,Mei2005[3-7]
Extracting human readable metadata
The information in a CLDF metadata file can be converted to markdown (a human readable markup language) running
cldf markdown PATH/TO/metadata.json
A typical usage of this feature is to create a README.md for your dataset
(which, when uploaded to e.g. GitHub will be rendered nicely in the browser).
Downloading media listed in a dataset's MediaTable
Typically, CLDF datasets only reference media items. The MediaTable provides enough information, though, to download and save an item's content. This can be done running
cldf downloadmedia PATH/TO/metadata.json PATH/TO/DOWNLOAD/DIR
To minimize bandwidth usage, relevant items can be filtered by passing selection criteria in the form
COLUMN_NAME=SUBSTRING as optional arguments. E.g. downloading could be limited to audio files passing
Media_Type=audio/ (provided, Media_Type is the name of the column with propertyUrl
http://cldf.clld.org/v1.0/terms.rdf#mediaType)
Converting a CLDF dataset to an SQLite database
A very useful feature of CSVW in general and CLDF in particular is that it
provides enough metadata for a set of CSV files to load them into a relational
database - including relations between tables. This can be done running the
cldf createdb command:
$ cldf createdb -h
usage: cldf createdb [-h] [--infer-primary-keys] DATASET SQLITE_DB_PATH
Load a CLDF dataset into a SQLite DB
positional arguments:
DATASET Dataset specification (i.e. path to a CLDF metadata
file or to the data file)
SQLITE_DB_PATH Path to the SQLite db file
For a specification of the resulting database schema refer to the documentation in
src/pycldf/db.py.
Handling large media files
Often, platforms like GitHub impose limits on the size of individual files in a repository. Thus,
in order to facilitate curation of datasets with large media files on such platforms, pycldf
provides a pragmatic solution as follows:
Running
cldf splitmedia <dataset-locator>
on a dataset will split all media files with sizes bigger than a configurable threshold into
multiple files, just like UNIX' split command would.
A file named audio.wav will be split into files audio.wav.aa, audio.wav.ab and so on.
[!CAUTION] With large files split (and removed) the dataset will not validate anymore.
In order to restore the files, the corresponding command
cldf catmedia <dataset-locator>
can be used.
Thus, in a typical workflow each commit to the repository would be wrapped in a cldf splitmedia
and a cldf catmedia call (possibly automated via git hooks).
Python API
For a detailed documentation of the Python API, refer to the docs on ReadTheDocs.
Reading CLDF
As an example, we'll read data from WALS Online, v2020:
>>> from pycldf import Dataset
>>> wals2020 = Dataset.from_metadata('https://raw.githubusercontent.com/cldf-datasets/wals/v2020/cldf/StructureDataset-metadata.json')
For exploratory purposes, accessing a remote dataset over HTTP is fine. But for real analysis, you'd want to download
the datasets first and then access them locally, passing a local file path to Dataset.from_metadata.
Let's look at what we got:
>>> print(wals2020)
<cldf:v1.0:StructureDataset at https://raw.githubusercontent.com/cldf-datasets/wals/v2020/cldf/StructureDataset-metadata.json>
>>> for c in wals2020.components:
... print(c)
...
ValueTable
ParameterTable
CodeTable
LanguageTable
ExampleTable
As expected, we got a StructureDataset, and in
addition to the required ValueTable, we also have a couple more components.
We can investigate the values using pycldf's ORM functionality, i.e. mapping rows in the CLDF
data files to convenient python objects. (Take note of the limitations describe in orm.py, though.)
>>> for value in wals2020.objects('ValueTable'):
... break
...
>>> value
<pycldf.orm.Value id="81A-aab">
>>> value.language
<pycldf.orm.Language id="aab">
>>> value.language.cldf
Namespace(glottocode=None, id='aab', iso639P3code=None, latitude=Decimal('-3.45'), longitude=Decimal('142.95'), macroarea=None, name='Arapesh (Abu)')
>>> value.parameter
<pycldf.orm.Parameter id="81A">
>>> value.parameter.cldf
Namespace(description=None, id='81A', name='Order of Subject, Object and Verb')
>>> value.references
(<Reference Nekitel-1985[94]>,)
>>> value.references[0]
<Reference Nekitel-1985[94]>
>>> print(value.references[0].source.bibtex())
@misc{Nekitel-1985,
olac_field = {syntax; general_linguistics; typology},
school = {Australian National University},
title = {Sociolinguistic Aspects of Abu', a Papuan Language of the Sepik Area, Papua New Guinea},
wals_code = {aab},
year = {1985},
author = {Nekitel, Otto I. M. S.}
}
If performance is important, you can just read rows of data as python dicts, in which case the references between
tables must be resolved "by hand":
>>> params = {r['id']: r for r in wals2020.iter_rows('ParameterTable', 'id', 'name')}
>>> for v in wals2020.iter_rows('ValueTable', 'parameterReference'):
... print(params[v['parameterReference']]['name'])
... break
...
Order of Subject, Object and Verb
Note that we passed names of CLDF terms to Dataset.iter_rows (e.g. id) specifying which columns we want to access
by CLDF term - rather than by the column names they are mapped to in the dataset.
Writing CLDF
Warning: Writing CLDF with pycldf does not automatic
