Open stillwalker1234 opened 7 years ago
First of all apologies for the late response. You are absolutely right, the omniglot example only displays the training error and does no validation split (but it definitely should!). This project was first meant as a quick "hack", which explains why it is a bit rough on the edges. Contributions for adding a validation split are more that welcome!
Am I right to assume that example code does not compute validation error/accuracy, only train error/acc?
And btw, thanks for the effort :)