search

Anmol6 / DNGO-BO

Bayesian optimization with DNGO (Deep Networks for Global Optimization)
11 stars 4 forks source link

Use an optimizer instead of working on a grid #1

Closed mgelbart closed 7 years ago

mgelbart commented 8 years ago

For example http://docs.scipy.org/doc/scipy/reference/generated/scipy.optimize.fmin_l_bfgs_b.html with random restarts

mgelbart commented 8 years ago

For best results you'll want to use a gradient-based optimizer. TensorFlow should be able to get these gradients for you. In particular, we want the gradient of the acquisition function with respect to the "candidate input space" x. NOT the derivative w.r.t. existing data points.

Remember to maximize not minimize.

Anmol6 commented 7 years ago

Yes, using gradient-descent on Tensorflow