Methodic weakness

[mo-ba]

keras-io/examples/vision/ipynb/siamese_network.ipynb

Line 520 in be3791e

"sample = next(iter(train_dataset))\n",

I think this Is a bad demonstration, since you only show that you learned the train_dataset.
I think it would probably be better to make a 60%/20%/20% split and run against unknown data.