Mini-batch training incorrectly implemented? about tensorflow-som

[cgorman]

Hi Felix, I can see why this is confusing, the SOM code itself makes few assumptions about the structure of the data it receives. Instead, the batching is handled by the caller. For example, these lines in the example script use the tensorflow datasets interface to handle the batching. The SOM code just gets the iterator and pulls however many tensors that the iterator specifies. The loop in the training function needs to know the batch size to make sure it doesn't loop too many times and exhaust the data stream.

Hope that makes sense, let me know if you have any further questions!

[felix-hilden]

I see, thanks for the clarification!