We are doing a major rewrite of tune. This is motivated mostly by adding prediction postprocessing. As a side effect, we’ve refactored the code that runs tune_grid().
I spent a few hours researching how we will integrate agua/h2o into the system. I think it is fairly straightforward and will result in less code inside of agua.
One approach is to add a helper function in agua that:
- translates parameter names
- converts the data to h2o
- calls
h2o::h2o.grid()
- makes predictions.
These are all in agua (although some required functions are unexported). tune could call this helper to make the predictions in a way that we can easily segue into postprocessing (if any).
By doing this, we can eliminate agua’s dependence on tune and add agua to tune’s suggests.
Here’s a gist with a prototype of the helper function.
Let me know what you think. I’m hoping to have tune ready for CRAN by September.
We are doing a major rewrite of tune. This is motivated mostly by adding prediction postprocessing. As a side effect, we’ve refactored the code that runs
tune_grid().I spent a few hours researching how we will integrate agua/h2o into the system. I think it is fairly straightforward and will result in less code inside of agua.
One approach is to add a helper function in agua that:
h2o::h2o.grid()These are all in agua (although some required functions are unexported). tune could call this helper to make the predictions in a way that we can easily segue into postprocessing (if any).
By doing this, we can eliminate agua’s dependence on tune and add agua to tune’s suggests.
Here’s a gist with a prototype of the helper function.
Let me know what you think. I’m hoping to have tune ready for CRAN by September.