net.simplace.sim.control.iterators.ShrinkParameterspaceIterator

Iterator that shrinks the parameter space, i. e. reduces the range where a parameter can vary.

Documentation

In each iteration step, the range of a parameter is shrunk by cShrinkFactor and centered around the the actual value of the parameter.

Initial step

Values for each cal variable's min and max or sigma and mean are taken from initial configuration.

Subsequent steps

If the cal variable has a value, calculated by a simulation and selected by a selector in the previous iteration, then the new range will be centered around this value. Otherwise the new range will be centered around the midpoint of the previous range.

If cDrift is set to true then the new range can exceed the initial min and max values. Otherwise the new range will be cut to fit inside the original range. If it is false, then min and max properties have to be give, even if you wan to use only mean and sigma.

Calculation for iteration step \(i \gt 1\): \[ \begin{eqnarray} mean_i & = & \begin{cases} parametervalue & \text{if parametervalue is set} \\ \frac{max_{i-1} + min_{i-1}}{2} & \text{else} \end{cases} \\ radius_i & = & \frac{max_{i-1} - min_{i-1}}{2} \cdot cShrinkFactor \\ min_i & = & \begin{cases} mean_i - radius_i & \text{if } cDrift \text{ is true} \\ \max(mean_i - radius_i,min_1) & \text{else} \end{cases} \\ max_i & = & \begin{cases} mean_i + radius_i & \text{if } cDrift \text{ is true} \\ \min(mean_i + radius_i,max_1) & \text{else} \end{cases} \\ sigma_i & = & sigma_{i-1} \cdot cShrinkFactor \end{eqnarray} \]

Stop condition

The iterator runs maximum maxCount times. Iteration will stop earlier, if the constraint rule becomes false.

Example

Defining and configuring an iterator

|  <iterator id="Iterator" class="net.simplace.sim.control.iterators.ShrinkParameterspaceIterator">
|      <input id="cShrinkFactor" datatype="DOUBLE">.5</input>
|      <input id="cDrift" datatype="BOOLEAN">false</input>
|      <input id="maxCount" datatype="INT">5</input>
|      <constraint description="" rule="${Selector.BestErrorValue} &gt; 46" />
|      <cal id="SLA" datatype="DOUBLE" target="vSLA">
|          <input id="min" datatype="DOUBLE">0.01</input>
|          <input id="max" datatype="DOUBLE">0.1</input>
|          <input id="sigma" datatype="DOUBLE">0.0025</input>
|          <input id="mean" datatype="DOUBLE">0.02</input>
|      </cal>
|      <cal id="LUE" datatype="DOUBLE" target="vLUE">
|          <input id="min" datatype="DOUBLE">1</input>
|          <input id="max" datatype="DOUBLE">5</input>
|          <input id="sigma" datatype="DOUBLE">0.21</input>
|          <input id="mean" datatype="DOUBLE">3.0</input>
|      </cal>
|  </iterator>

Iterator calculates min/max/mean/sigma for two parameters vSLA and vLUE.

Initial interval for vLUE is from 1.0 to 5.0. In each interation step, the size of the interval will be half of the previous size. E.g. if after iteration step 1 the selected LUE will be 3.7, then the new intervall will range from 2.7 to 4.7. If the selected LUE after iteration step 2 will be 3.8, then the new intervall will range from 3.3 to 4.3 etc... If the selected LUE after iteration step 2 will be 4.6, then the new interval will range from 4.1 to 5.0, as the designed upper bound 5.1 is bigger than initial max value 5.0.

Using the iterator by a generator

|  <generator  id="Generator" class="net.simplace.client.simulation.lap.generators.RandomSimGenerator">
|    <input id="cCount" datatype="INT">100</input>
|    <cal id="LUE" datatype="DOUBLE" target="vLUE">
|	  <input id="min" datatype="DOUBLE" source="vLUE_min" />
|	  <input id="max" datatype="DOUBLE" source="vLUE_max" />
|      <input id="sigma" datatype="DOUBLE" source="Iterator.LUE.sigma" />
|      <input id="mean" datatype="DOUBLE" source="Iterator.LUE.mean" />
|      <input id="distribution" datatype="CHAR">Normal</input>
|    </cal>
|    <cal id="SLA" datatype="DOUBLE" target="vSLA">
|      <input id="min" datatype="DOUBLE" source="vSLA_min" />
|      <input id="max" datatype="DOUBLE" source="vSLA_max" />
|      <input id="distribution" datatype="CHAR">Uniform</input>
|    </cal>
|  </generator>

The generator will generate 100 simulations in each, chosing vLUE from a normal distribution centered around the mean and sigma values provided by the iterator. As sigma becomes smaller, the range of the values will be smaller.

Notice that for vSLA min and max are independent from the iterator, so in each step the values will be randomly chosen from the same range.



public class ShrinkParameterspaceIterator extends DefaultSimulationIterator {
// Public Constructors
public ShrinkParameterspaceIterator(Element aSimulationIteratorElement,
FWSimSession aSession);

// Public Instance Methods
public HashMap createVariables(); // Defines
net.simplace.sim.control.FWSimulationIterator

public void init(); // Defines net.simplace.sim.control.FWSimulationIterator


// Protected Instance Methods
protected boolean iterate(String aProjectID);
// Overrides DefaultSimulationIterator


}



Hierarchy: java.lang.Object - net.simplace.sim.control.FWSimulationControlContainer (net.simplace.sim.util.FWSimFieldContainer) - net.simplace.sim.control.FWSimulationIterator - DefaultSimulationIterator - ShrinkParameterspaceIterator