search

suiji / Arborist

Scalable decision tree training and inference.
Other
82 stars 14 forks source link

[New Feature]: Increase "verbosity" option for tracking progress. #13

Closed coforfe closed 6 years ago

coforfe commented 8 years ago

It would be interesting to include in RBorist a functionality that package ranger has already implemented.

ranger package has a nice feature that although it is purely informative, it helps monitor the progress during the model execution.

The output of a ranger model with this verbosity activated is like this:

> datIn  <- twoClassSim(n = 50000)
> modFit <- ranger( Class ~., data= datIn, num.trees = 1000, mtry = 4, verbose = TRUE)
Growing trees.. Progress: 19%. Estimated remaining time: 2 minutes, 12 seconds.
Growing trees.. Progress: 35%. Estimated remaining time: 1 minute, 56 seconds.
Growing trees.. Progress: 53%. Estimated remaining time: 1 minute, 24 seconds.
Growing trees.. Progress: 71%. Estimated remaining time: 50 seconds.
Growing trees.. Progress: 89%. Estimated remaining time: 19 seconds.

Thanks, Carlos.

suiji commented 7 years ago

The development branch now has the hooks to report diagnostics back to the user. Although these were introduced to report the state of coprocessor involvement, the underlying mechanism should be versatile enough to support verbose observations. IOW, we should see something in the next release.

suiji commented 6 years ago

Initial support for verbose has been introduced. Progress is announced as each major phase is completed. The option will become a bit more informative when training is performed in batches, rather than en banc.