DLS Course 4 week 2: how is the model accuracy calculated for the alpaca model?

Dear mentor,

I have successfully completed and submitted the assignment on transfer learning of course 4, week2, of the DLS course, but I have a question on how the model accuracy is calculated for the alpaca model.

When I print out the results of predicting alpaca vs. not alpaca with model2, I observe that all predictions are <0.5, and in most cases even negative, see screenshot below.

How is it possible that the model accuracy is around 80%, and not “zero”?