TunerGridSearch

mlr_tuners_grid_search

Subclass for grid search tuning.
The grid is constructed as a Cartesian product over discretized values per parameter,
see <code><a rd-options="paradox:generate_design_grid" href="/link/paradox%3A%3Agenerate_design_grid()?package=mlr3tuning&version=0.1.0&to=paradox%3Agenerate_design_grid" data-mini-rdoc="paradox:generate_design_grid::paradox::generate_design_grid()">paradox::generate_design_grid()</a></code>.
The points of the grid are evaluated in a random order.
In order to support general termination criteria and parallelization,
we evaluate points in a batch-fashion of size <code>batch_size</code>.
Larger batches mean we can parallelize more, smaller batches imply a more fine-grained checking
of termination criteria.

datasets

Implements methods for hyperparameter tuning with
'mlr3', e.g. Grid Search, Random Search, or Simulated Annealing.
Various termination criteria can be set and combined.  The class
'AutoTuner' provides a convenient way to perform nested resampling in
combination with 'mlr3'.

Michel Lang

mlr3tuning

Tuning for 'mlr3'

Jakob Richter

Bernd Bischl

Daniel Schalk

TunerGridSearch function

<a rd-options="R6:R6Class" href="/link/R6%3A%3AR6Class?package=mlr3tuning&version=0.1.0&to=R6%3AR6Class" data-mini-rdoc="R6:R6Class::R6::R6Class">R6::R6Class</a> object inheriting from <a rd-options="" href="/link/Tuner?package=mlr3tuning&version=0.1.0" data-mini-rdoc="mlr3tuning::Tuner">Tuner</a>.

Format

<pre>TunerGridSearch$new()
tnr("grid_search")
</pre>

Construction

<ul>
<li><code>resolution</code> :: <code>integer(1)</code>
Resolution of the grid, see <code><a rd-options="paradox:generate_design_grid" href="/link/paradox%3A%3Agenerate_design_grid()?package=mlr3tuning&version=0.1.0&to=paradox%3Agenerate_design_grid" data-mini-rdoc="paradox:generate_design_grid::paradox::generate_design_grid()">paradox::generate_design_grid()</a></code>.</li>
<li><code>param_resolutions</code> :: named <code>integer()</code> 
Resolution per parameter, named by parameter ID, see <code><a rd-options="paradox:generate_design_grid" href="/link/paradox%3A%3Agenerate_design_grid()?package=mlr3tuning&version=0.1.0&to=paradox%3Agenerate_design_grid" data-mini-rdoc="paradox:generate_design_grid::paradox::generate_design_grid()">paradox::generate_design_grid()</a></code>.</li>
<li><code>batch_size</code> :: <code>integer(1)</code>
Maximum number of configurations to try in a batch.</li>
</ul>

Parameters

Subclass for grid search tuning.
The grid is constructed as a Cartesian product over discretized values per parameter,
see <code><a rd-options='paradox:generate_design_grid' href='paradox::generate_design_grid()'>paradox::generate_design_grid()</a></code>.
The points of the grid are evaluated in a random order.
In order to support general termination criteria and parallelization,
we evaluate points in a batch-fashion of size <code>batch_size</code>.
Larger batches mean we can parallelize more, smaller batches imply a more fine-grained checking
of termination criteria.