Maybe use a different dataset with the levels of the oceans, and see if you can find the lowest point on earth (Challenger Deep) from the peek of the Everest ? :P
And now, do Adagrad, RMSProp, RMSProp + Nesterov and Adam, and maybe Newton, BFGS, L-BFGS and conjugate gradients, and then coordinate ascent :)
It would be a pretty good educational tool to teach people the different gradient descent methods, though it's probably too simple of a problem for these methods to be at all useful.
And now, do Adagrad, RMSProp, RMSProp + Nesterov and Adam, and maybe Newton, BFGS, L-BFGS and conjugate gradients, and then coordinate ascent :)
It would be a pretty good educational tool to teach people the different gradient descent methods, though it's probably too simple of a problem for these methods to be at all useful.