org.lenskit.eval.traintest

• All Known Implementing Classes:

public interface EvalTask

Interface for evaluation tasks. Each evaluation task performs some task with the trained model and measures the results. Performing a task on a recommender trained over a particular data set results is called a measurement.

TrainTestExperiment
• ### Method Summary

All Methods
Modifier and Type Method and Description
ConditionEvaluator createConditionEvaluator(AlgorithmInstance algorithm, DataSet dataSet, Recommender rec)
Set up a measurement of a single recommender.
void finish()
List<String> getGlobalColumns()
Get columns that will go in the aggregate output file.
Set<Class<?>> getRequiredRoots()
Get the root types required by this evaluation.
List<String> getUserColumns()
Get columns that will go in the per-user output file.
void start(ExperimentOutputLayout outputLayout)
Do initial setup for this eval task.
• ### Method Detail

• #### getGlobalColumns

List<String> getGlobalColumns()

Get columns that will go in the aggregate output file.

Returns:
The list of column names that this task will contribute to the aggregate output file.
• #### getUserColumns

List<String> getUserColumns()

Get columns that will go in the per-user output file.

Returns:
The list of column names that this task will contribute to the per-user output file.
• #### getRequiredRoots

Set<Class<?>> getRequiredRoots()

Get the root types required by this evaluation.

Returns:
The root types required by this evaluation.
• #### start

void start(ExperimentOutputLayout outputLayout)

Do initial setup for this eval task. This should create any per-task output files, etc.

Parameters:
outputLayout - The output layout for experiment results.
• #### finish

void finish()

Finalize this eval task. This should finish writing and close any per-task output files, etc.

• #### createConditionEvaluator

ConditionEvaluator createConditionEvaluator(AlgorithmInstance algorithm,
DataSet dataSet,
Recommender rec)

Set up a measurement of a single recommender.

Parameters:
algorithm - The algorithm being evaluated.
dataSet - The data set being evaluated.
rec - The recommender to measure.
Returns:
A condition evaluator that will measure the recommender’s performance on the algorithm and data set.