PyKEEN Benchmarking Experiment Model Files
Lists
Only you see this
Abstract
Model Weights This repository provides weights of the models from the benchmarking study conducted in "Bringing Light Into the Dark: A Large-scale Evaluation of Knowledge Graph Embedding Models Under a Unified Framework" which have been upgraded to compatible with PyKEEN 1.9. The weights are organized as zipfiles, which are named by the dataset-interaction function configuration. For each of these combinations, we chose the best according to validation Hits@10 to include into this repository. For each model, we have three files: <code>configuration.json</code> contains the (pipeline) configuration used to train the model. It can loaded as <pre class="sourceCode python"><code class="sourceCode python">import pathlib import json configuration = json.loads(pathlib.Path("configuration.json").read_text())</code></pre> Since the configuration is intended for the <em>pipeline</em>, we need some custom code to re-create the model without re-training it. <pre class="sourceCode python"><code class="sourceCode python">from pykeen.datasets import get_dataset from pykeen.models import ERModel, model_resolver configuration = configuration["pipeline"] # load the triples factory dataset = get_dataset( dataset=configuration["dataset"], dataset_kwargs=configuration.get("dataset_kwargs", None) ) model: ERModel = model_resolver.make( configuration["model"], configuration["model_kwargs"], triples_factory=dataset.training )</code></pre> Note, that this only creates the model instance, but does <em>not</em> load the weights, yet. <code>state_dict.pt</code> contains the weights, stored via <code>torch.save</code>. They can be loaded via <pre class="sourceCode python"><code class="sourceCode python">import torch state_dict = torch.load("state_dict.pt")</code></pre> We can load these weights into the model by using <code>Module.load_state_dict</code> <pre class="sourceCode python"><code class="sourceCode python">model.load_state_dict(state_dict, strict=False)</code></pre> Note that we set <code>strict=False</code>, since the exported weights do <em>not</em> contain regularizers' state, while the re-instantiated models may have regularizers. <code>results.json</code> contains the results obtained by the original runs. It can be read by <pre class="sourceCode python"><code class="sourceCode python">import pathlib import json configuration = json.loads(pathlib.Path("results.json").read_text())</code></pre> Note that some of the recently added metrics are not available in those results.
BibTeX
@inproceedings{ali2022pykeen,
author = {Mehdi Ali and Max Berrendorf and Charles Tapley Hoyt and Laurent Vermue and Mikhail Galkin},
title = {{PyKEEN Benchmarking Experiment Model Files}},
booktitle = {Zenodo},
year = {2022},
doi = {10.5281/zenodo.7019181},
}Loading related content...