pykeen 1.0.2

A package for training and evaluating multimodal knowledge graph embeddings

PyKEEN

PyKEEN (Python KnowlEdge EmbeddiNgs) is a Python package designed to train and evaluate knowledge graph embedding models (incorporating multi-modal information). It is part of the KEEN Universe.

Installation • Quickstart • Datasets • Models • Support

Installation

The development version of PyKEEN can be downloaded and installed from PyPI on Python 3.7+ with:

$ pip install pykeen

The development version of PyKEEN can be downloaded and installed from GitHub on Python 3.7+ with:

$ git clone https://github.com/pykeen/pykeeen.git pykeen
$ cd pykeen
$ pip install -e .
$ # Install pre-commit
$ pip install pre-commit
$ pre-commit install

PyKEEN has several extras for installation that are defined in the [options.extras_require] section of the setup.cfg. They can be included with installation using the bracket notation like in pip install pykeen[docs] or pip install -e .[docs]. Several can be listed, comma-delimited like in pip install pykeen[docs,plotting].

Name	Description
plotting	Plotting with seaborn and generation of word clouds
mlflow	Tracking of results with mlflow
docs	Building of the documentation
templating	Building of templated documentation, like the README

Contributing

Contributions, whether filing an issue, making a pull request, or forking, are appreciated. See CONTRIBUTING.md for more information on getting involved.

Quickstart

This example shows how to train a model on a data set and test on another data set.

The fastest way to get up and running is to use the pipeline function. It provides a high-level entry into the extensible functionality of this package. The following example shows how to train and evaluate the TransE model on the Nations dataset. By default, the training loop uses the stochastic local closed world assumption (sLCWA) training approach and evaluates with rank-based evaluation.

from pykeen.pipeline import pipeline
result = pipeline(
    model='TransE',
    dataset='nations',
)

The results are returned in a dataclass that has attributes for the trained model, the training loop, and the evaluation.

PyKEEN is extensible such that:

• Each model has the same API, so anything from pykeen.models can be dropped in
• Each training loop has the same API, so pykeen.training.LCWATrainingLoop can be dropped in
• Triples factories can be generated by the user with from pykeen.triples.TriplesFactory

Implementation

Below are the models, data sets, training modes, evaluators, and metrics implemented in pykeen.

Datasets (13)

Name	Reference	Description
fb15k	pykeen.datasets.FB15k	The FB15k data set.
fb15k237	pykeen.datasets.FB15k237	The FB15k-237 data set.
hetionet	pykeen.datasets.Hetionet	The Hetionet dataset is a large biological network.
kinships	pykeen.datasets.Kinships	The Kinships data set.
nations	pykeen.datasets.Nations	The Nations data set.
openbiolink	pykeen.datasets.OpenBioLink	The OpenBioLink dataset.
openbiolinkf1	pykeen.datasets.OpenBioLinkF1	The PyKEEN First Filtered OpenBioLink 2020 Dataset.
openbiolinkf2	pykeen.datasets.OpenBioLinkF2	The PyKEEN Second Filtered OpenBioLink 2020 Dataset.
openbiolinklq	pykeen.datasets.OpenBioLinkLQ	The low-quality variant of the OpenBioLink dataset.
umls	pykeen.datasets.UMLS	The UMLS data set.
wn18	pykeen.datasets.WN18	The WN18 data set.
wn18rr	pykeen.datasets.WN18RR	The WN18-RR data set.
yago310	pykeen.datasets.YAGO310	The YAGO3-10 data set is a subset of YAGO3 that only contains entities with at least 10 relations.

Models (23)

Name	Reference	Citation
ComplEx	pykeen.models.ComplEx	Trouillon et al., 2016
ComplExLiteral	pykeen.models.ComplExLiteral	Agustinus et al., 2018
ConvE	pykeen.models.ConvE	Dettmers et al., 2018
ConvKB	pykeen.models.ConvKB	Nguyen et al., 2018
DistMult	pykeen.models.DistMult	Yang et al., 2014
DistMultLiteral	pykeen.models.DistMultLiteral	Agustinus et al., 2018
ERMLP	pykeen.models.ERMLP	Dong et al., 2014
ERMLPE	pykeen.models.ERMLPE	Sharifzadeh et al., 2019
HolE	pykeen.models.HolE	Nickel et al., 2016
KG2E	pykeen.models.KG2E	He et al., 2015
NTN	pykeen.models.NTN	Socher et al., 2013
ProjE	pykeen.models.ProjE	Shi et al., 2017
RESCAL	pykeen.models.RESCAL	Nickel et al., 2011
RGCN	pykeen.models.RGCN	Schlichtkrull et al., 2018
RotatE	pykeen.models.RotatE	Sun et al., 2019
SimplE	pykeen.models.SimplE	Kazemi et al., 2018
StructuredEmbedding	pykeen.models.StructuredEmbedding	Bordes et al., 2011
TransD	pykeen.models.TransD	Ji et al., 2015
TransE	pykeen.models.TransE	Bordes et al., 2013
TransH	pykeen.models.TransH	Wang et al., 2014
TransR	pykeen.models.TransR	Lin et al., 2015
TuckER	pykeen.models.TuckER	Balazevic et al., 2019
UnstructuredModel	pykeen.models.UnstructuredModel	Bordes et al., 2014

Losses (7)

Name	Reference	Description
bce	pykeen.losses.BCELoss	A wrapper around the PyTorch binary cross entropy loss.
bceaftersigmoid	pykeen.losses.BCEAfterSigmoidLoss	A loss function which uses the numerically unstable version of explicit Sigmoid + BCE.
crossentropy	pykeen.losses.CrossEntropyLoss	Evaluate cross entropy after softmax output.
marginranking	pykeen.losses.MarginRankingLoss	A wrapper around the PyTorch margin ranking loss.
mse	pykeen.losses.MSELoss	A wrapper around the PyTorch mean square error loss.
nssa	pykeen.losses.NSSALoss	An implementation of the self-adversarial negative sampling loss function proposed by [sun2019]_.
softplus	pykeen.losses.SoftplusLoss	A loss function for the softplus.

Regularizers (5)

Name	Reference	Description
combined	pykeen.regularizers.CombinedRegularizer	A convex combination of regularizers.
lp	pykeen.regularizers.LpRegularizer	A simple L_p norm based regularizer.
no	pykeen.regularizers.NoRegularizer	A regularizer which does not perform any regularization.
powersum	pykeen.regularizers.PowerSumRegularizer	A simple x^p based regularizer.
transh	pykeen.regularizers.TransHRegularizer	A regularizer for the soft constraints in TransH.

Optimizers (6)

Name	Reference	Description
adadelta	torch.optim.Adadelta	Implements Adadelta algorithm.
adagrad	torch.optim.Adagrad	Implements Adagrad algorithm.
adam	torch.optim.Adam	Implements Adam algorithm.
adamax	torch.optim.Adamax	Implements Adamax algorithm (a variant of Adam based on infinity norm).
adamw	torch.optim.AdamW	Implements AdamW algorithm.
sgd	torch.optim.SGD	Implements stochastic gradient descent (optionally with momentum).

Training Loops (2)

Name	Reference	Description
lcwa	pykeen.training.LCWATrainingLoop	A training loop that uses the local closed world assumption training approach.
slcwa	pykeen.training.SLCWATrainingLoop	A training loop that uses the stochastic local closed world assumption training approach.

Negative Samplers (2)

Name	Reference	Description
basic	pykeen.sampling.BasicNegativeSampler	A basic negative sampler.
bernoulli	pykeen.sampling.BernoulliNegativeSampler	An implementation of the bernoulli negative sampling approach proposed by [wang2014]_.

Stoppers (2)

Name	Reference	Description
early	pykeen.stoppers.EarlyStopper	A harness for early stopping.
nop	pykeen.stoppers.NopStopper	A stopper that does nothing.

Evaluators (2)

Name	Reference	Description
rankbased	pykeen.evaluation.RankBasedEvaluator	A rank-based evaluator for KGE models.
sklearn	pykeen.evaluation.SklearnEvaluator	An evaluator that uses a Scikit-learn metric.

Metrics (6)

Metric	Description	Evaluator	Reference
Adjusted Mean Rank	The mean over all chance-adjusted ranks: mean_i (2r_i / (num_entities+1)). Lower is better.	rankbased	pykeen.evaluation.RankBasedMetricResults
Average Precision Score	The area under the precision-recall curve, between [0.0, 1.0]. Higher is better.	sklearn	pykeen.evaluation.SklearnMetricResults
Hits At K	The hits at k for different values of k, i.e. the relative frequency of ranks not larger than k. Higher is better.	rankbased	pykeen.evaluation.RankBasedMetricResults
Mean Rank	The mean over all ranks: mean_i r_i. Lower is better.	rankbased	pykeen.evaluation.RankBasedMetricResults
Mean Reciprocal Rank	The mean over all reciprocal ranks: mean_i (1/r_i). Higher is better.	rankbased	pykeen.evaluation.RankBasedMetricResults
Roc Auc Score	The area under the ROC curve between [0.0, 1.0]. Higher is better.	sklearn	pykeen.evaluation.SklearnMetricResults

Hyper-parameter Optimization

Samplers (2)

Name	Reference	Description
random	optuna.samplers.RandomSampler	Sampler using random sampling.
tpe	optuna.samplers.TPESampler	Sampler using TPE (Tree-structured Parzen Estimator) algorithm.

Experimentation

Reproduction

PyKEEN includes a set of curated experimental settings for reproducing past landmark experiments. They can be accessed and run like:

pykeen experiments reproduce tucker balazevic2019 fb15k

Where the three arguments are the model name, the reference, and the data set. The output directory can be optionally set with -d.

Ablation

PyKEEN includes the ability to specify ablation studies using the hyper-parameter optimization module. They can be run like:

pykeen experiments ablation ~/path/to/config.json

Acknowledgements

Supporters

This project has been supported by several organizations (in alphabetical order):

• Bayer
• Enveda Therapeutics
• Fraunhofer Institute for Algorithms and Scientific Computing
• Fraunhofer Institute for Intelligent Analysis and Information Systems
• Fraunhofer Center for Machine Learning
• Ludwig-Maximilians-Universität München
• Munich Center for Machine Learning (MCML)
• Siemens
• Smart Data Analytics Research Group (University of Bonn & Fraunhofer IAIS)
• Technical University of Denmark - DTU Compute - Section for Cognitive Systems
• Technical University of Denmark - DTU Compute - Section for Statistics and Data Analysis
• University of Bonn

Logo

The PyKEEN logo was designed by Carina Steinborn.