tf.opt is a tool for solving constrained optimization problems with objective functions that cannot be directly optimized in a tractable manner, but can be learned via neural networks.
The source is being released incrementally, the process is still incomplete.
This is not an officially supported Google product.
Below are links to related data from papers that use tf.opt.
- A set of MIP instances generated by tf.opt related to the paper "Constrained Discrete Black-Box Optimization using Mixed-Integer Programming". They are snapshots of the MIP acquisition problems solved in the method described in the paper. See the paper and the README file in the package for more details.