Vectors/home

From Nordic Language Processing Laboratory
(Difference between revisions)
Jump to: navigation, search
(Using NLPL Models In-Situ)
 
(8 intermediate revisions by 2 users not shown)
Line 1: Line 1:
 
= Background =
 
= Background =
  
The purpose of the NLPL repository of word vectors/embeddings (which can comprise both ‘classic’, count-based and ‘modern’, dense models) is
+
The purpose of the NLPL repository of word vectors/embeddings (which can comprise both ‘classic’, count-based and ‘modern’, dense models, including the contextualized ones) is
 
to make available a large and carefully curated collection of large-scale distributional semantic models for many languages.
 
to make available a large and carefully curated collection of large-scale distributional semantic models for many languages.
 
For general background, please see
 
For general background, please see
Line 8: Line 8:
 
For interactive exploration and download of the repository, there is an
 
For interactive exploration and download of the repository, there is an
 
[http://vectors.nlpl.eu/repository/ on-line explorer].
 
[http://vectors.nlpl.eu/repository/ on-line explorer].
The underlying data is stored in the NLPL project directory below
+
The underlying data is stored in the NLPL community directory below
<tt>/projects/nlpl/data/vectors/</tt> (on Abel)
+
<tt>/cluster/shared/nlpl/data/vectors/</tt> (on Saga)
and <tt>/proj/nlpl/data/vectors/</tt> (on Taito).
+
and <tt>/projappl/nlpl/data/vectors/</tt> (on Puhti).
 
The repository is versioned, in the sense of assigning release numbers to different
 
The repository is versioned, in the sense of assigning release numbers to different
 
stages of repository construction.
 
stages of repository construction.
Line 22: Line 22:
 
and re-packaging the models from the original release in a more standardized
 
and re-packaging the models from the original release in a more standardized
 
format (see below).
 
format (see below).
 +
In December 2019, version 2.0 was released, which added BERT and ELMo models, made metadata more consistent
 +
and ensured that binary format models are always provided
 +
(to increase the loading speed, in comparison to the models stored as plain text).
  
 
= Repository Contents =
 
= Repository Contents =
Line 28: Line 31:
 
for programmatic access in the repository ''catalogue'', which is represented
 
for programmatic access in the repository ''catalogue'', which is represented
 
as a JSON file in the top-level repository directory, with catalogue names
 
as a JSON file in the top-level repository directory, with catalogue names
corresponding to each repository version, e.g. <tt>/projects/nlpl/data/vectors/10.json</tt>
+
corresponding to each repository version, e.g. <tt>/cluster/shared/nlpl/data/vectors/20.json</tt>
(on Abel) for the initial repository release.
+
(on Saga) for the current repository release.
  
 
The catalogue contains three top-level sections, one each for <tt>corpora</tt>
 
The catalogue contains three top-level sections, one each for <tt>corpora</tt>
 
(data sources), <tt>algorithms</tt> (model creation tools), and
 
(data sources), <tt>algorithms</tt> (model creation tools), and
 
<tt>models</tt> (resulting sets of word vectors).
 
<tt>models</tt> (resulting sets of word vectors).
NLPL users with access to Abel and Taito can read the
+
NLPL users with access to Saga and Puhti can read the
 
catalogue file directly from the
 
catalogue file directly from the
project directory, for example when executing a series of experiments
+
NLPL community directory, for example when executing a series of experiments
 
that make use of different pre-trained sets of word vectors.
 
that make use of different pre-trained sets of word vectors.
 +
Further
 +
[http://wiki.nlpl.eu/index.php/Vectors/metadata documentation] of the catalogue metadata is available
 +
as a [http://wiki.nlpl.eu/index.php/Vectors/metadata separate page].
  
 
Each repository entry (i.e. set of word vectors, or ‘model’) is
 
Each repository entry (i.e. set of word vectors, or ‘model’) is
Line 47: Line 53:
 
the specific contents; a <tt>README</tt> file included
 
the specific contents; a <tt>README</tt> file included
 
with each model entry provides a life-time unique identifier, e.g.
 
with each model entry provides a life-time unique identifier, e.g.
<tt>http://vectors.nlpl.eu/repository/11/3.zip</tt> for
+
<tt>http://vectors.nlpl.eu/repository/20/3.zip</tt> for
model #3 in the 1.1 release of the repository.
+
model #3 in the 2.0 release of the repository.
  
 
= Using NLPL Models In-Situ =
 
= Using NLPL Models In-Situ =
  
 
To avoid data duplication, it is recommended to load models from the NLPL repository
 
To avoid data duplication, it is recommended to load models from the NLPL repository
directly from the NLPL project directory, when working on Abel or Taito.
+
directly from the NLPL community directory, when working on Saga or Puhti.
 
Repository entries are uniformly packaged as <tt>.zip</tt> compressed archives, but
 
Repository entries are uniformly packaged as <tt>.zip</tt> compressed archives, but
 
the uniform naming scheming makes it possible to directly read one or more of the
 
the uniform naming scheming makes it possible to directly read one or more of the
Line 63: Line 69:
 
  import zipfile
 
  import zipfile
 
  import gensim
 
  import gensim
  repository = "/projects/nlpl/data/vectors/11"
+
  repository = "/cluster/shared/nlpl/data/vectors/20"
 
  with zipfile.ZipFile(repository + "/30.zip", "r") as archive:
 
  with zipfile.ZipFile(repository + "/30.zip", "r") as archive:
 
   stream = archive.open("model.txt")
 
   stream = archive.open("model.txt")
Line 78: Line 84:
 
= Future Work =
 
= Future Work =
  
# The life-time <tt>handle</tt> for each model should be included in the JSON catalogue (in addition to being listed in the <tt>README</tt> file).
+
# Prepare Version 2.0, now with <code>/cluster/shared/nlpl/data/vectors/</code> as the master copy (DONE).
# Each corpus should be listed as a separate entry; corpus combinations go into the array-valued <tt>corpus</tt> property on models.
+
# The life-time <tt>handle</tt> for each model should be included in the JSON catalogue (in addition to being listed in the <tt>README</tt> file) (DONE).
# Where applicable, there should be an array-valued <tt>documentation</tt> field (of string, typically URLs) on corpora and models.
+
# For classic models, redundantly add binary binary <code>model.bin</code> for faster loading (DONE).
# The <tt>maintainers</tt> property may be over-promising, as often third-party models are in practice unmaintained; maybe rename to <tt>creator</tt>?
+
# Each corpus should be listed as a separate entry; corpus combinations go into the array-valued <tt>corpus</tt> property on models (DONE).
 +
# Where applicable, there should be an array-valued <tt>documentation</tt> field (of string, typically URLs) on corpora and models (DONE).
 +
# The <tt>maintainers</tt> property may be over-promising, as often third-party models are in practice unmaintained; maybe rename to <tt>creator</tt> (DONE).
 +
# Document (and possibly re-design) the metadata scheme; maybe invent a fourth category for a <tt>process</tt> applied prior model training.

Latest revision as of 00:31, 3 January 2020

Contents

[edit] Background

The purpose of the NLPL repository of word vectors/embeddings (which can comprise both ‘classic’, count-based and ‘modern’, dense models, including the contextualized ones) is to make available a large and carefully curated collection of large-scale distributional semantic models for many languages. For general background, please see Fares et al. (2017).

For interactive exploration and download of the repository, there is an on-line explorer. The underlying data is stored in the NLPL community directory below /cluster/shared/nlpl/data/vectors/ (on Saga) and /projappl/nlpl/data/vectors/ (on Puhti). The repository is versioned, in the sense of assigning release numbers to different stages of repository construction. Each repository entry, thus, is assigned a unique and persistent identifier; once published, a repository entry will never change (to aid replicability). The initial release (providing some two dozen models) was published in May 2017 as version 1.0. In March 2018, version 1.1 supersedes this initial release, adding a large number of models and languages (including those from the UD parsing task) and re-packaging the models from the original release in a more standardized format (see below). In December 2019, version 2.0 was released, which added BERT and ELMo models, made metadata more consistent and ensured that binary format models are always provided (to increase the loading speed, in comparison to the models stored as plain text).

[edit] Repository Contents

The on-line browser dynamically presents parts of the information encoded for programmatic access in the repository catalogue, which is represented as a JSON file in the top-level repository directory, with catalogue names corresponding to each repository version, e.g. /cluster/shared/nlpl/data/vectors/20.json (on Saga) for the current repository release.

The catalogue contains three top-level sections, one each for corpora (data sources), algorithms (model creation tools), and models (resulting sets of word vectors). NLPL users with access to Saga and Puhti can read the catalogue file directly from the NLPL community directory, for example when executing a series of experiments that make use of different pre-trained sets of word vectors. Further documentation of the catalogue metadata is available as a separate page.

Each repository entry (i.e. set of word vectors, or ‘model’) is packaged in the form of a .zip archive, with uniform conventions for file naming inside the file, using the model.txt and model.bin entries for the actual vectors. Each archive includes the relevant excerpts from the catalogue as a file meta.json to help identify the specific contents; a README file included with each model entry provides a life-time unique identifier, e.g. http://vectors.nlpl.eu/repository/20/3.zip for model #3 in the 2.0 release of the repository.

[edit] Using NLPL Models In-Situ

To avoid data duplication, it is recommended to load models from the NLPL repository directly from the NLPL community directory, when working on Saga or Puhti. Repository entries are uniformly packaged as .zip compressed archives, but the uniform naming scheming makes it possible to directly read one or more of the model files from the archive.

In Python, for example, something along the following lines should work to iterate over all of the entries in the model

import zipfile
import gensim
repository = "/cluster/shared/nlpl/data/vectors/20"
with zipfile.ZipFile(repository + "/30.zip", "r") as archive:
  stream = archive.open("model.txt")
  for line in stream:
    ...

Alternatively, if working in a framework like gensim

  model = gensim.models.KeyedVectors.load_word2vec_format(stream, binary=False, unicode_errors='replace')

Binary fastText models (stored as parameters.bin files) should be first extracted from the .zip archive, and then loaded with

  model = gensim.models.fasttext.load_facebook_vectors("parameters.bin")

[edit] Future Work

  1. Prepare Version 2.0, now with /cluster/shared/nlpl/data/vectors/ as the master copy (DONE).
  2. The life-time handle for each model should be included in the JSON catalogue (in addition to being listed in the README file) (DONE).
  3. For classic models, redundantly add binary binary model.bin for faster loading (DONE).
  4. Each corpus should be listed as a separate entry; corpus combinations go into the array-valued corpus property on models (DONE).
  5. Where applicable, there should be an array-valued documentation field (of string, typically URLs) on corpora and models (DONE).
  6. The maintainers property may be over-promising, as often third-party models are in practice unmaintained; maybe rename to creator (DONE).
  7. Document (and possibly re-design) the metadata scheme; maybe invent a fourth category for a process applied prior model training.
Personal tools
Namespaces

Variants
Actions
Navigation
Tools