Lab: Using Graph Embeddings - Revision history

Sinoa: /* Topics */

2025-04-19T12:09:28Z

Topics

← Older revision		Revision as of 12:09, 19 April 2025
Line 1:		Line 1:

	==Topics==		==Topics==
	Using knowledge graph embeddings with TorchKGE.		Using knowledge graph embeddings with TorchKGE. (PyKEEN is an alternative that may have become more developed than TorchKGE.)

	==Useful readings==		==Useful readings==

Bamos3003 at 12:03, 2 May 2024

2024-05-02T12:03:29Z

← Older revision		Revision as of 12:03, 2 May 2024
Line 13:		Line 13:
	'''Task: knowledge graph''':		'''Task: knowledge graph''':
	* Use a [https://torchkge.readthedocs.io/en/latest/reference/utils.html#pre-trained-models dataset loader] to load a KG you want to work with. Freebase FB15k237 is a good choice. (You will need a pre-trained model for your KG later, to choose one of FB15k, FB15k237, WDV5, WN18RR, or Yago3-10. This lab has mostly been tested on FB15k.)		* Use a [https://torchkge.readthedocs.io/en/latest/reference/utils.html#pre-trained-models dataset loader] to load a KG you want to work with. Freebase FB15k237 is a good choice. (You will need a pre-trained model for your KG later, to choose one of FB15k, FB15k237, WDV5, WN18RR, or Yago3-10. This lab has mostly been tested on FB15k.)
	* Use the methods provided by the [https://torchkge.readthedocs.io/en/latest/reference/data.html#knowledge-graph ~~KnolwedgeGraph~~ class] to inspect the KG.		* Use the methods provided by the [https://torchkge.readthedocs.io/en/latest/reference/data.html#knowledge-graph KnowledgeGraph class] to inspect the KG.
	** Print out the numbers of entities, relations, and facts in the training, validation, and testing sets.		** Print out the numbers of entities, relations, and facts in the training, validation, and testing sets.
	** Print the identifiers for the first 10 entities and relations (''tip:'' ent2ix and rel2ix).		** Print the identifiers for the first 10 entities and relations (''tip:'' ent2ix and rel2ix).

Rbo027: /* Code to get started */

2023-04-18T19:15:14Z

Code to get started

← Older revision		Revision as of 19:15, 18 April 2023
Line 40:		Line 40:

	==Code to get started==		==Code to get started==
	With graph embeddings, we ideally want to work with ipynb files. The code ~~bellow~~ is prepared in the following link: https://colab.research.google.com/drive/1gS2D1XYSviAmhkS8moJIpY0N8ltJFM3C		With graph embeddings, we ideally want to work with ipynb files. The code below is prepared in the following link: https://colab.research.google.com/drive/1gS2D1XYSviAmhkS8moJIpY0N8ltJFM3C

	<syntaxhighlight>		<syntaxhighlight>

Rbo027: /* Tasks */

2023-04-18T19:08:17Z

Tasks

← Older revision		Revision as of 19:08, 18 April 2023
Line 28:		Line 28:
	'''Task: model''':		'''Task: model''':
	* Load a [https://torchkge.readthedocs.io/en/latest/reference/utils.html#pre-trained-models pre-trained TransE model] that matches your KnowledgeGraph.		* Load a [https://torchkge.readthedocs.io/en/latest/reference/utils.html#pre-trained-models pre-trained TransE model] that matches your KnowledgeGraph.
	** Print out the numbers of entities, relations, and the dimensions of the entity and relation vectors. Do they match your KnowledgeGraph.		** Print out the numbers of entities, relations, and [https://torchkge.readthedocs.io/en/latest/reference/models.html#transe the dimensions] of the entity and relation vectors. Do they match your KnowledgeGraph.
	* Get the vectors for your test entities and relations (for example, 'J. K. Rowling' and 'influenced by').		* Get the vectors for your test entities and relations (for example, 'J. K. Rowling' and 'influenced by').
	* Find vectors for a few more entities (both unrelated and related ones, e.g., 'J. R. R. Tolkien', 'C. S. Lewis', ...). Use the [https://torchkge.readthedocs.io/en/latest/reference/models.html#translationalmodels model.dissimilarity()-method] to estimate how semantically close your entities are. Do the distances make sense?		* Find vectors for a few more entities (both unrelated and related ones, e.g., 'J. R. R. Tolkien', 'C. S. Lewis', ...). Use the [https://torchkge.readthedocs.io/en/latest/reference/models.html#translationalmodels model.dissimilarity()-method] to estimate how semantically close your entities are. Do the distances make sense?

Rbo027: /* Useful readings */

2023-04-18T18:48:02Z

Useful readings

← Older revision		Revision as of 18:48, 18 April 2023
Line 8:		Line 8:
	** ''KnowledgeGraph'' - contains the knowledge graph (KG)		** ''KnowledgeGraph'' - contains the knowledge graph (KG)
	** ''Model'' - contains the embeddings (entity and relation vectors) for some KG		** ''Model'' - contains the embeddings (entity and relation vectors) for some KG
			* [https://pytorch.org/docs/stable/tensors.html PyTorch Tensor Documentation]

	==Tasks==		==Tasks==

Rbo027: /* Tasks */ added a tip to test entities and relations

2023-04-18T14:53:43Z

Tasks: added a tip to test entities and relations

← Older revision		Revision as of 14:53, 18 April 2023
Line 23:		Line 23:

	'''Task: test entities and relations''':		'''Task: test entities and relations''':
	* Get the KG indexes for a few entities and relations. If you use the Freebase or Wikidata graphs, you can try 'J. K. Rowling' and 'WALL·E' as entities (''note'' that the dot in 'WALL·E' is not a hyphen or usual period.) For relations you can try 'influenced by' and 'genre'.		* Get the KG indexes for a few entities and relations. If you use the Freebase or Wikidata graphs, you can try 'J. K. Rowling' and 'WALL·E' as entities (''note'' that the dot in 'WALL·E' is not a hyphen or usual period.) For relations you can try 'influenced by' and 'genre'. (''tip'': to check names of entites and relations, open the train.txt file you cloned)

	'''Task: model''':		'''Task: model''':

Rbo027: /* Code to get started */

2023-04-18T13:36:16Z

Code to get started

← Older revision		Revision as of 13:36, 18 April 2023
Line 39:		Line 39:

	==Code to get started==		==Code to get started==
	With graph embeddings, we ideally want to work with ipynb files. The code bellow is prepared in the following link: https://colab.research.google.com/drive/1gS2D1XYSviAmhkS8moJIpY0N8ltJFM3C~~?authuser=1#scrollTo=oZcIgRVKe__2~~		With graph embeddings, we ideally want to work with ipynb files. The code bellow is prepared in the following link: https://colab.research.google.com/drive/1gS2D1XYSviAmhkS8moJIpY0N8ltJFM3C

	<syntaxhighlight>		<syntaxhighlight>

Rbo027: /* Code to get started */

2023-04-18T13:35:28Z

Code to get started

← Older revision		Revision as of 13:35, 18 April 2023
Line 39:		Line 39:

	==Code to get started==		==Code to get started==
	With graph embeddings, we ideally want to work with ipynb files~~. Therefore, Google Colab can be a useful tool~~. The code bellow is prepared in the following link: https://colab.research.google.com/drive/1gS2D1XYSviAmhkS8moJIpY0N8ltJFM3C?authuser=1#scrollTo=oZcIgRVKe__2		With graph embeddings, we ideally want to work with ipynb files. The code bellow is prepared in the following link: https://colab.research.google.com/drive/1gS2D1XYSviAmhkS8moJIpY0N8ltJFM3C?authuser=1#scrollTo=oZcIgRVKe__2

	<syntaxhighlight>		<syntaxhighlight>

Rbo027: /* Code to get started */

2023-04-18T13:33:42Z

Code to get started

← Older revision		Revision as of 13:33, 18 April 2023
Line 39:		Line 39:

	==Code to get started==		==Code to get started==
			With graph embeddings, we ideally want to work with ipynb files. Therefore, Google Colab can be a useful tool. The code bellow is prepared in the following link: https://colab.research.google.com/drive/1gS2D1XYSviAmhkS8moJIpY0N8ltJFM3C?authuser=1#scrollTo=oZcIgRVKe__2

	<syntaxhighlight>		<syntaxhighlight>
	!pip install torchkge		!pip install torchkge

Sinoa: /* Tasks */

2023-04-18T06:34:44Z

Tasks

← Older revision		Revision as of 06:34, 18 April 2023
Line 31:		Line 31:
	* Find vectors for a few more entities (both unrelated and related ones, e.g., 'J. R. R. Tolkien', 'C. S. Lewis', ...). Use the [https://torchkge.readthedocs.io/en/latest/reference/models.html#translationalmodels model.dissimilarity()-method] to estimate how semantically close your entities are. Do the distances make sense?		* Find vectors for a few more entities (both unrelated and related ones, e.g., 'J. R. R. Tolkien', 'C. S. Lewis', ...). Use the [https://torchkge.readthedocs.io/en/latest/reference/models.html#translationalmodels model.dissimilarity()-method] to estimate how semantically close your entities are. Do the distances make sense?

	'''K-nearest neighbours''':		'''Task: K-nearest neighbours''':
	* Find the indexes of the 10 entity vectors that are nearest neighbours to your entity of choice. You can use [https://scikit-learn.org/stable/modules/generated/sklearn.neighbors.NearestNeighbors.html sciKit-learn's sklearn.neighbors.NearestNeighbors.kneighbors()-method] for this.		* Find the indexes of the 10 entity vectors that are nearest neighbours to your entity of choice. You can use [https://scikit-learn.org/stable/modules/generated/sklearn.neighbors.NearestNeighbors.html sciKit-learn's sklearn.neighbors.NearestNeighbors.kneighbors()-method] for this.
	* Map the indexes of the 10-nearest neighbouring entities back into human-understandable labels. Does this make sense? Try the same thing with another entity (e.g., 'WALL·E').		* Map the indexes of the 10-nearest neighbouring entities back into human-understandable labels. Does this make sense? Try the same thing with another entity (e.g., 'WALL·E').

	'''~~Translation~~''':		'''Task: translation''':
	* Add together the vectors for an entity and a relation that that gives meaning for the entity (e.g., 'J. K. Rowling' - 'influenced by', 'WALL·E' - 'genre'). Find the 10-nearest neighbouring entities for the vector sum. Does this make sense? Try more entities and relations. Try to find examples that work and that do not work well.		* Add together the vectors for an entity and a relation that that gives meaning for the entity (e.g., 'J. K. Rowling' - 'influenced by', 'WALL·E' - 'genre'). Find the 10-nearest neighbouring entities for the vector sum. Does this make sense? Try more entities and relations. Try to find examples that work and that do not work well.