/// <summary>
/// Runs the standard active learning procedure in parallel on an array of model instances and an input data set.
/// </summary>
/// <param name="data">The data.</param>
/// <param name="modelName">The model name.</param>
/// <param name="runType">The model run type.</param>
/// <param name="model">The model instance.</param>
/// <param name="taskSelectionMethod">The method for selecting tasks (Random / Entropy).</param>
/// <param name="workerSelectionMethod">The method for selecting workers (only Random is implemented).</param>
/// <param name="numIncremData">The number of data points to add at each iteration.</param>
/// <param name="communityCount">The number of communities (only for CBCC).</param>
/// <param name="initialNumLabelsPerTask">The initial number of exploratory labels that are randomly selected for each task.</param>
public static void RunParallelActiveLearning(IList<Datum> data,
string[] modelName,
RunType[] runType,
BCC[] model,
TaskSelectionMethod[] taskSelectionMethod,
WorkerSelectionMethod[] workerSelectionMethod,
int communityCount = -1,
int initialNumLabelsPerTask = 1,
int numIncremData = 1)
{
int numModels = runType.Length;
Stopwatch stopWatch = new Stopwatch();
int totalLabels = data.Count();
// Dictionary keyed by task Id, with randomly order labelings
var groupedRandomisedData =
data.GroupBy(d => d.TaskId).
Select(g =>
{
var arr = g.ToArray();
int cnt = arr.Length;
var perm = Rand.Perm(cnt);
return new
{
key = g.Key,
arr = g.Select((t, i) => arr[perm[i]]).ToArray()
};
}).ToDictionary(a => a.key, a => a.arr);
// Dictionary keyed by task Id, with label counts
Dictionary<string, int> totalCounts = groupedRandomisedData.ToDictionary(kvp => kvp.Key, kvp => kvp.Value.Length);
Dictionary<string, int> currentCounts = groupedRandomisedData.ToDictionary(kvp => kvp.Key, kvp => initialNumLabelsPerTask);
// Keyed by task, value is a HashSet containing all the remaining workers with a label - workers are removed after adding a new datum
Dictionary<string, HashSet<string>> remainingWorkersPerTask = groupedRandomisedData.ToDictionary(kvp => kvp.Key, kvp => new HashSet<string>(kvp.Value.Select(dat => dat.WorkerId)));
int numTaskIds = totalCounts.Count();
int totalInstances = data.Count - initialNumLabelsPerTask * numTaskIds;
//throw an exception if the totalInstances is less than or equals to zero
if (totalInstances <= 0)
{
throw new System.Exception("The variable 'totalInstances' should be greater than zero");
}
//only creates accuracy list when it's null (for GUI Use)
if (accuracyArray == null)
{
accuracyArray = Util.ArrayInit<List<double>>(numModels, i => new List<double>());
}
List<double>[] avgRecallArray = Util.ArrayInit(numModels, i => new List<double>());
taskValueListArray = Util.ArrayInit(numModels, i => new List<ActiveLearningResult>());
int[] indexArray = new int[numModels];
Debug.WriteLine("Parallel Active Learning");
Debug.WriteLine("\tModel\tAcc\tAvgRec");
// Get initial data
//make the results variable be global for GUi
results = Util.ArrayInit<Results>(numModels, i => new Results());
List<Datum> subData = GetSubdata(groupedRandomisedData, currentCounts, remainingWorkersPerTask);
List<Datum>[] subDataArray = Util.ArrayInit<List<Datum>>(numModels, i => new List<Datum>(subData));
List<Datum>[] nextData = new List<Datum>[numModels];
ActiveLearning[] activeLearning = new ActiveLearning[numModels];
isExperimentCompleted = false;
// Main loop
for (int iter = 0; ; iter++)
{
bool calculateAccuracy = true;
bool doSnapShot = iter % 100 == 0; // Frequency of snapshots
//stop Active Learning if the user requests to stop
if (isExperimentCompleted) {
return;
}
// Run all the models
for (int indexModel = 0; indexModel < numModels; indexModel++ )
{
if (subDataArray[indexModel] != null || nextData[indexModel] != null)
{
switch (runType[indexModel])
{
case RunType.VoteDistribution:
results[indexModel].RunMajorityVote(subDataArray[indexModel], data, calculateAccuracy, true);
break;
case RunType.MajorityVote:
results[indexModel].RunMajorityVote(subDataArray[indexModel], data, calculateAccuracy, false);
break;
case RunType.DawidSkene:
results[indexModel].RunDawidSkene(subDataArray[indexModel], data, calculateAccuracy);
break;
default: // Run BCC models
results[indexModel].RunBCC(modelName[indexModel], subDataArray[indexModel], data, model[indexModel], RunMode.ClearResults, calculateAccuracy, communityCount, false);
break;
}
} //end for running all the data
if (activeLearning[indexModel] == null)
{
activeLearning[indexModel] = new ActiveLearning(data, model[indexModel], results[indexModel], communityCount);
}
else
{
activeLearning[indexModel].UpdateActiveLearningResults(results[indexModel]);
}
// Select next task
Dictionary<string, ActiveLearningResult> TaskUtility = new Dictionary<string, ActiveLearningResult>();
switch (taskSelectionMethod[indexModel])
{
case TaskSelectionMethod.EntropyTask:
TaskUtility = activeLearning[indexModel].EntropyTrueLabel();
break;
case TaskSelectionMethod.RandomTask:
TaskUtility = data.GroupBy(d => d.TaskId).ToDictionary(a => a.Key, a => new ActiveLearningResult
{
TaskValue = Rand.Double()
});
break;
case TaskSelectionMethod.UniformTask:
//add task value according to the count left
TaskUtility = currentCounts.OrderBy(kvp => kvp.Value).ToDictionary(a => a.Key, a => new ActiveLearningResult
{
TaskValue = 1
});
break;
default: // Entropy task selection
TaskUtility = activeLearning[indexModel].EntropyTrueLabel();
break;
}
// We create a list of worker utilities
Dictionary<string, double> WorkerAccuracy = null;
// Best worker selection is only allowed for methods that infer worker confusion matrices.
if (results[indexModel].WorkerConfusionMatrix == null)
workerSelectionMethod[indexModel] = WorkerSelectionMethod.RandomWorker;
switch (workerSelectionMethod[indexModel])
{
case WorkerSelectionMethod.BestWorker:
// Assign worker accuracies to the maximum value on the diagonal of the confusion matrix (conservative approach).
// Alternative ways are also possible.
WorkerAccuracy = results[indexModel].WorkerConfusionMatrixMean.ToDictionary(
kvp => kvp.Key,
kvp => Results.GetConfusionMatrixDiagonal(kvp.Value).Max());
break;
case WorkerSelectionMethod.RandomWorker:
// Assign worker accuracies to random values
WorkerAccuracy = results[indexModel].FullMapping.WorkerIdToIndex.ToDictionary(kvp => kvp.Key, kvp => Rand.Double());
break;
default:
throw new ApplicationException("No worker selection method selected");
}
// Create a list of tuples (TaskId, WorkerId, ActiveLearningResult)
List<Tuple<string, string, ActiveLearningResult>> LabelValue = new List<Tuple<string, string, ActiveLearningResult>>();
foreach (var kvp in TaskUtility)
{
foreach (var workerId in remainingWorkersPerTask[kvp.Key])
{
var labelValue = new ActiveLearningResult
{
WorkerId = workerId,
TaskId = kvp.Key,
TaskValue = kvp.Value.TaskValue,
WorkerValue = WorkerAccuracy[workerId]
};
LabelValue.Add(Tuple.Create(labelValue.TaskId, labelValue.WorkerId, labelValue));
}
}
// Increment tha active set with new data
nextData[indexModel] = GetNextData(groupedRandomisedData, LabelValue, currentCounts, totalCounts, remainingWorkersPerTask, numIncremData);
if (nextData[indexModel] == null || nextData[indexModel].Count == 0)
break;
indexArray[indexModel] += nextData[indexModel].Count;
subDataArray[indexModel].AddRange(nextData[indexModel]);
// Logs
if (calculateAccuracy)
{
accuracyArray[indexModel].Add(results[indexModel].Accuracy);
avgRecallArray[indexModel].Add(results[indexModel].AvgRecall);
if (TaskUtility == null)
{
var sortedLabelValue = LabelValue.OrderByDescending(kvp => kvp.Item3.TaskValue).ToArray();
taskValueListArray[indexModel].Add(sortedLabelValue.First().Item3);
}
else
{
//Adding WorkerId into taskValueListArray
ActiveLearningResult nextTaskValueItem = TaskUtility[nextData[indexModel].First().TaskId];
nextTaskValueItem.WorkerId = nextData[indexModel].First().WorkerId;
nextTaskValueItem.TaskId = nextData[indexModel].First().TaskId;
taskValueListArray[indexModel].Add(nextTaskValueItem);
}
if (doSnapShot)
{
Debug.WriteLine("{0} of {1}:\t{2}\t{3:0.000}\t{4:0.0000}", indexArray[indexModel], totalInstances, modelName[indexModel], accuracyArray[indexModel].Last(), avgRecallArray[indexModel].Last());
}
}
}//end of models
}//end for all data
}
/// <summary>
/// Runs the standard active learning procedure on a model instance and an input data set.
/// </summary>
/// <param name="data">The data.</param>
/// <param name="modelName">The model name.</param>
/// <param name="runType">The model run type.</param>
/// <param name="model">The model instance.</param>
/// <param name="taskSelectionMethod">The method for selecting tasks (Random / Entropy).</param>
/// <param name="workerSelectionMethod">The method for selecting workers (only Random is implemented).</param>
/// <param name="resultsDir">The directory to save the log files.</param>
/// <param name="communityCount">The number of communities (only for CBCC).</param>
/// <param name="initialNumLabelsPerTask">The initial number of exploratory labels that are randomly selected for each task.</param>
/// <param name="numIncremData">The number of data points to add at each round.</param>
public static void RunActiveLearning(
IList<Datum> data,
string modelName,
RunType runType,
BCC model,
TaskSelectionMethod taskSelectionMethod,
WorkerSelectionMethod workerSelectionMethod,
string resultsDir,
int communityCount = -1,
int initialNumLabelsPerTask = 1,
int numIncremData = 1)
{
//Count elapsed time
Stopwatch stopWatchTotal = new Stopwatch();
stopWatchTotal.Start();
int totalLabels = data.Count();
// Dictionary keyed by task Id, with randomly order labelings
var groupedRandomisedData =
data.GroupBy(d => d.TaskId).
Select(g =>
{
var arr = g.ToArray();
int cnt = arr.Length;
var perm = Rand.Perm(cnt);
return new
{
key = g.Key,
arr = g.Select((t, i) => arr[perm[i]]).ToArray()
};
}).ToDictionary(a => a.key, a => a.arr);
// Dictionary keyed by task Id, with label counts
Dictionary<string, int> totalCounts = groupedRandomisedData.ToDictionary(kvp => kvp.Key, kvp => kvp.Value.Length);
// Keyed by task, value is a HashSet containing all the remaining workers with a label - workers are removed after adding a new datum
Dictionary<string, HashSet<string>> remainingWorkersPerTask = groupedRandomisedData.ToDictionary(kvp => kvp.Key, kvp => new HashSet<string>(kvp.Value.Select(dat => dat.WorkerId)));
int numTaskIds = totalCounts.Count();
int totalInstances = initialNumLabelsPerTask > 0 ? data.Count - initialNumLabelsPerTask * numTaskIds : data.Count - numIncremData;
//throw an exception if the totalInstances is less than or equals to zero
if (totalInstances <= 0)
{
throw new System.Exception("The variable 'totalInstances' should be greater than zero");
}
string[] WorkerIds = data.Select(d => d.WorkerId).Distinct().ToArray();
//only creat accuracy list when it's null (for GUI Use)
if (accuracy == null)
{
accuracy = new List<double>();
}
List<double> avgRecall = new List<double>();
//List<ActiveLearningResult> taskValueList = new List<ActiveLearningResult>();
taskValueList = new List<ActiveLearningResult>();
int index = 0;
Console.WriteLine("Active Learning: {0}", modelName);
Console.WriteLine("\t\t\t\t\t\tAcc\tAvgRec");
// Get initial data
Results results = new Results();
Dictionary<string, int> currentCounts = groupedRandomisedData.ToDictionary(kvp => kvp.Key, kvp => initialNumLabelsPerTask);
List<Datum> subData = GetSubdata(groupedRandomisedData, currentCounts, remainingWorkersPerTask);
var s = remainingWorkersPerTask.Select(w => w.Value.Count).Sum();
List<Datum> nextData = null;
ActiveLearning activeLearning = null;
isExperimentCompleted = false;
for (int iter = 0; ; iter++) //run until data run out
{
bool calculateAccuracy = true;
bool doSnapShot = iter % 1 == 0;
if (subData != null || nextData != null)
{
switch (runType)
{
case RunType.VoteDistribution:
results.RunMajorityVote(subData, data, calculateAccuracy, true);
break;
case RunType.MajorityVote:
results.RunMajorityVote(subData, data, calculateAccuracy, false);
break;
case RunType.DawidSkene:
results.RunDawidSkene(subData, data, calculateAccuracy);
break;
default: // Run BCC models
results.RunBCC(modelName, subData, data, model, RunMode.ClearResults, calculateAccuracy, communityCount, false);
break;
}
}
if (activeLearning == null)
{
activeLearning = new ActiveLearning(data, model, results, communityCount);
}
else
{
activeLearning.UpdateActiveLearningResults(results);
}
// We create a list of task utilities
// TaskValue: Dictionary keyed by task, the value is an active learning result.
Dictionary<string, ActiveLearningResult> TaskUtility = null;
switch (taskSelectionMethod)
{
case TaskSelectionMethod.EntropyTask:
TaskUtility = activeLearning.EntropyTrueLabel();
break;
case TaskSelectionMethod.RandomTask:
TaskUtility = data.GroupBy(d => d.TaskId).ToDictionary(a => a.Key, a => new ActiveLearningResult
{
TaskValue = Rand.Double()
});
break;
case TaskSelectionMethod.UniformTask:
// Reproduce uniform task selection by picking the task with the lowest number of current labels. That is, minus the current count.
TaskUtility = currentCounts.OrderBy(kvp => kvp.Value).ToDictionary(a => a.Key, a => new ActiveLearningResult
{
TaskId = a.Key,
TaskValue = -a.Value
});
break;
default:
TaskUtility = activeLearning.EntropyTrueLabel();
break;
}
// We create a list of worker utilities.
Dictionary<string, double> WorkerAccuracy = null;
// Best worker selection is only allowed for methods that infer worker confusion matrices.
if (results.WorkerConfusionMatrix == null)
workerSelectionMethod = WorkerSelectionMethod.RandomWorker;
switch (workerSelectionMethod)
{
case WorkerSelectionMethod.BestWorker:
// Assign worker accuracies to the maximum value on the diagonal of the confusion matrix (conservative approach).
// Alternative ways are also possible.
WorkerAccuracy = results.WorkerConfusionMatrixMean.ToDictionary(
kvp => kvp.Key,
kvp => Results.GetConfusionMatrixDiagonal(kvp.Value).Max());
break;
case WorkerSelectionMethod.RandomWorker:
// Assign worker accuracies to random values
WorkerAccuracy = results.FullMapping.WorkerIdToIndex.ToDictionary(kvp => kvp.Key, kvp => Rand.Double());
break;
default:
throw new ApplicationException("No worker selection method selected");
}
// Create a list of tuples (TaskIds, WorkerId, ActiveLearningResult).
List<Tuple<string, string, ActiveLearningResult>> LabelValue = new List<Tuple<string,string,ActiveLearningResult>>();
foreach (var kvp in TaskUtility)
{
foreach (var workerId in remainingWorkersPerTask[kvp.Key])
{
var labelValue = new ActiveLearningResult
{
WorkerId = workerId,
TaskId = kvp.Key,
TaskValue = kvp.Value.TaskValue,
WorkerValue = WorkerAccuracy[workerId]
};
LabelValue.Add(Tuple.Create(labelValue.TaskId, labelValue.WorkerId, labelValue));
}
}
// Increment tha active set with new data
nextData = GetNextData(groupedRandomisedData, LabelValue, currentCounts, totalCounts, remainingWorkersPerTask, numIncremData);
if (nextData == null || nextData.Count == 0)
break;
index += nextData.Count;
subData.AddRange(nextData);
// Logs
if (calculateAccuracy)
{
accuracy.Add(results.Accuracy);
avgRecall.Add(results.AvgRecall);
if (TaskUtility == null)
{
var sortedLabelValue = LabelValue.OrderByDescending(kvp => kvp.Item3.TaskValue).ToArray();
taskValueList.Add(sortedLabelValue.First().Item3);
}
else
{
//Adding WorkerId into taskValueList
ActiveLearningResult nextTaskValueItem = TaskUtility[nextData.First().TaskId];
nextTaskValueItem.WorkerId = nextData.First().WorkerId;
//add taskID
nextTaskValueItem.TaskId = nextData.First().TaskId;
taskValueList.Add(nextTaskValueItem);
}
if (doSnapShot)
{
Console.WriteLine("{0} (label {1} of {2}):\t{3:0.000}\t{4:0.0000}", modelName, index, totalInstances, accuracy.Last(), avgRecall.Last());
//DoSnapshot(accuracy, nlpd, avgRecall, taskValueList, results, modelName, "interim", resultsDir, initialNumLabelsPerTask);
}
}//end if logs
}//end for all data
isExperimentCompleted = true;
stopWatchTotal.Stop();
DoSnapshot(accuracy, avgRecall, taskValueList, results, modelName, "final", resultsDir, initialNumLabelsPerTask);
ResetAccuracyList();
Console.WriteLine("Elapsed time: {0}\n", stopWatchTotal.Elapsed);
}
