Improve documentation on bring your own entity/relation initializers

[mvd-lab]

Thank you for a very useful platform for building KGE models. I am working my own KG data set and I would like to initialize entity and relation embeddings with my own initial values for ConvKB and a few other models. I don't see any sample code for it. I see that the initializer functions can be set from one of several internally implemented (such uniform_ etc) but I don't see how I can make them use vectors pre-calculated. I appreciate your help.

[cthoyt]

Please have a look at https://pykeen.readthedocs.io/en/stable/reference/nn/init.html. We need to improve the high level documentation on this page and make sure its better linked from the tutorial

[mberr]

This should be the relevant part: https://pykeen.readthedocs.io/en/stable/api/pykeen.nn.init.PretrainedInitializer.html#pykeen.nn.init.PretrainedInitializer

[mvd-lab]

Yes! That worked. Thank you!

[shekarpour]

Hi
I have my own dataset containing three types of triples:

1. triples as relation between entities e.g., 234 Located_IN 1254
2. triples as properties of entities, e.g., 2356 First_Name Sarah
3. triples as label/type of nodes such as 2356 type Person or 234 type location

Triplefactory treats everything as entity for example all the properties and type values. How can I guide it to differentiate between entity ID and property values?

Thank you