> That is, there is no "testing" vs "training" dataset distinction at this point; there is just cross-validation of the training set.
If they just create a test set from the training set, and omit that from the training, what's the difference? The main thing is that you don't want to include the test set in the training step, and I assume they're doing that.
According to what they wrote, there is no separate testing set, so they estimate accuracy based on the training set. They use cross-validation to reduce their overfitting bias. (I could be wrong since I haven't actually used their service, but this is the take-home message of the wording of their explanation. If they intended to convey otherwise, they used the wrong language.)
Comments
> That is, there is no "testing" vs "training" dataset distinction at this point; there is just cross-validation of the training set.
If they just create a test set from the training set, and omit that from the training, what's the difference? The main thing is that you don't want to include the test set in the training step, and I assume they're doing that.
According to what they wrote, there is no separate testing set, so they estimate accuracy based on the training set. They use cross-validation to reduce their overfitting bias. (I could be wrong since I haven't actually used their service, but this is the take-home message of the wording of their explanation. If they intended to convey otherwise, they used the wrong language.)