UNQ_C8 Model Accuracy too low

[Jacobbboyce]

After passing all previous unit tests, my model has an accuracy of ~97.7%, which is too low for the last unit test. Most of my batches have a high accuracy, though sporadically batches have an accuracy of ~60%, any reason why my model didn’t train as intended?

[arvyzukai]

Hi @Jacobbboyce

First you should check local vs. global variable use. (len(Vocab) vs. vocab_size etc.)

Also the usual suspects - data_generator and classifier functions:

@Jacobbboyce

sporadically batches have an accuracy of ~60%

Cheers