Custom Calculator

TabularPivotModel

In TabularPivotModel, Calculator is implemented by two segments: summarization and context. Given a Calculator as an aggregation operation, the context refers to the intermediate result we keep track of while iterating over the operands (accumulation), while summarization refers to the computation from the context to the end result. For example, if we want to compute the AVERAGE over a list of numbers, we would likely keep track of their SUM and COUNT, and then retrieve the final result by dividing the two. In the above example, the context would be a data structure holding the sum and the count, while the summarization would be the division.

Define a Custom Calculator

To define a custom calculator, you must specify a ContextType so the model knows what to keep track of in accumulation phase.

public interface ContextualCalculator<C extends Context<C>> extends Calculator {
	
	public Number getResult(C context); // summarize the end result from the context
	
	public ContextType<C> getContextType(); // specify the context type
	
}

You can either use an existing ContextType, or create your own. The built-in context type are: TODO

It is encouraged to make ContextType as singletons, as Calculators can share Context if they are of the same ContextType.

public interface ContextType<C extends Context<C>> {
	
	public C create(); // ContextType has the responsibility as a Context factory
	
}
public interface Context<C extends Context<C>> {
	
	public void add(Object item); // what to do when iterating over raw data.
	
	public void merge(C ctx); // what to do when merging from contexts of a partition of its raw data set.
	
}

Example: Data Range

Say, we want to create a custom calculator to compute the number range (MAX minus MIN) of data, and luckily the StandardContextType.MIN_MAX already carries sufficient information to do so. We just need to define a ContextualCalculator based on the context type:

public class Range implements ContextualCalculator<MinMaxContext> {
	
	public static final Range INSTANCE = new Range();
	private Range() {}
	
	@Override
	@SuppressWarnings("unchecked")
	public ContextType<MinMaxContext> getContextType() {
		return StandardContextType.MIN_MAX;
	}
	@Override
	public Number getResult(MinMaxContext context) {
		return context.getMax().doubleValue() - context.getMin().doubleValue();
	}
	@Override
	public String getLabel() {
		return "Range";
	}
	@Override
	public String getLabelKey() {
		return "range";
	}
}

Example: Distinct Count

It may not be too obvious how to implement DISTINCT COUNT (which calculates the number of distinct items in the collection), but the best approach on context is a Set. As there is no similar built-in context, we have to implement our own:

public class SetContext implements Context<SetContext> {
	
	private final Set<Object> _set = new HashSet<Object>();
	@Override
	public void add(Object item) {
		_set.add(item);
	}
	@Override
	public void merge(SetContext ctx) {
		_set.addAll(ctx._set);
	}
	
	public Set<Object> getSet() {
		return _set;
	}
	
	public static final ContextType<SetContext> CONTEXT_TYPE = 
			new ContextType<SetContext>() {
		@Override
		public SetContext create() {
			return new SetContext();
		}
	};
}

Then the summarization part becomes quite obvious:

public class DistinctCount implements ContextualCalculator<SetContext> {
	
	public static final DistinctCount INSTANCE = new DistinctCount();
	private DistinctCount() {}
	
	@Override
	public ContextType<SetContext> getContextType() {
		return SetContext.CONTEXT_TYPE;
	}
	@Override
	public Number getResult(SetContext context) {
		return context.getSet().size();
	}
	@Override
	public String getLabel() {
		return "Distinct Count";
	}
	@Override
	public String getLabelKey() {
		return "distinctCount";
	}
}

Performance

The memory footprint size required for a Calculator is proportional to the footprint of its Context. And, for each context type, the add method will be called as many times as the raw data count. Thus, you should be aware of that contexts storing Set, Collection may cause a performance hit (caused by memory bound).

Version History

Version	Date	Content
2.0.0	June 2012	ContextualCalculator