Closed Zach-Attach closed 3 weeks ago
Hi @Zach-Attach, the embedding_dim
argument corresponds to the output dimensionality of the encoder network (often, the number of neurons in the last layer of the network) and is for when a user wants to pass custom encoders. For encoders that the package provides out-of-the-box, embedding_dim
is determined automatically and hence a None
value is accepted too.
Does this help?
Looks like it needs to be implemented in Line 118 in brain/builder.py
Closing. This was fixed.
Describe the bug
embedding_dim
doesn't appear to be used anywhere.To Reproduce See search of code here: https://github.com/search?q=repo%3Abuildingamind%2FNewbornEmbodiedTuringTest%20embedding_dim&type=code
Expected behavior embedding_dim should determine the number of dimensions in embedding vectors