Training and data

#4
by kenokall - opened

Hey, how did you get 100% accuracy on the evaluation data? Did you train and evaluate on different invoice layouts or did they all look like the example that can be seen in your Jupyter Notebook? Because I think the model might be heavily biased towards that layout. Also could you share what the rough number of training and evaluation samples are?

Greetings,
Keno

Sign up or log in to comment