/// <summary>
/// Gets conditional probability table plot data.
/// </summary>
/// <param name="array">
/// The array representing the conditional probability table.
/// </param>
/// <param name="mapping">
/// The data mapping.
/// </param>
/// <param name="rowLabel">
/// The row label.
/// </param>
/// <param name="columnLabel">
/// The column label.
/// </param>
/// <returns>
/// The conditional probability table plot data.
/// </returns>
public static Dictionary <string, Dictionary <string, List <Tweet> > > GetTweetMatrix(
List <Tweet>[,] array,
CrowdDataMapping mapping,
string rowLabel,
string columnLabel)
{
Debug.Assert(
array.GetLength(0) == mapping.LabelCount && array.GetLength(1) == mapping.LabelCount,
"Inconsistent arguments");
var result = new Dictionary <string, Dictionary <string, List <Tweet> > >();
for (var i = 0; i < mapping.LabelCount; i++)
{
var rowName = $"{mapping.LabelValueToString[mapping.LabelIndexToValue[i]]} ({rowLabel})";
var row = new Dictionary <string, List <Tweet> >();
for (var j = 0; j < mapping.LabelCount; j++)
{
var colName = $"{mapping.LabelValueToString[mapping.LabelIndexToValue[j]]} ({columnLabel})";
row[colName] = array[i, j];
}
result[rowName] = row;
}
return(result);
}
/// <summary>
/// Gets confusion matrix or conditional probability table plot data.
/// </summary>
/// <param name="array">
/// The array representing the conditional probability table.
/// </param>
/// <param name="mapping">
/// The data mapping.
/// </param>
/// <param name="rowLabel">
/// The row label.
/// </param>
/// <param name="columnLabel">
/// The column label.
/// </param>
/// <param name="asPercentages">Entries as percentages.</param>
/// <returns>
/// The conditional probability table plot data.
/// </returns>
public static Dictionary <string, List <object> > GetConfusionMatrix(
double[,] array,
CrowdDataMapping mapping,
string rowLabel,
string columnLabel,
bool asPercentages = false)
{
Debug.Assert(
array.GetLength(0) == mapping.LabelCount && array.GetLength(1) == mapping.LabelCount,
"Inconsistent arguments");
var result = new Dictionary <string, List <object> >();
var labelIndexToString = mapping.LabelIndexToString;
var rowSums = new double[mapping.LabelCount];
var namesColumn = new List <object>();
for (var row = 0; row < mapping.LabelCount; row++)
{
var rowName = $"{labelIndexToString[row]} ({rowLabel})";
namesColumn.Add(rowName);
for (var col = 0; col < mapping.LabelCount; col++)
{
rowSums[row] += array[row, col];
}
}
result[string.Empty] = namesColumn;
for (var col = 0; col < mapping.LabelCount; col++)
{
var columnName = $"{labelIndexToString[col]} ({columnLabel})";
var column = new List <object>();
for (var row = 0; row < mapping.LabelCount; row++)
{
var value = array[row, col];
if (asPercentages)
{
column.Add(new Percentage(value / rowSums[row]));
}
else
{
column.Add(value);
}
}
result[columnName] = column;
}
return(result);
}
/// <summary>
/// Gets conditional probability table plot data.
/// </summary>
/// <param name="workerCpts">
/// The worker conditional probability tables.
/// </param>
/// <param name="mapping">
/// The data mapping.
/// </param>
/// <param name="asPercentages">Entries as percentages</param>
/// <returns>
/// The conditional probability table plot data.
/// </returns>
public static Dictionary <string, Dictionary <string, List <object> > > GetWorkerCpts(
Dictionary <string, Dirichlet[]> workerCpts,
CrowdDataMapping mapping,
bool asPercentages = false)
{
return(workerCpts.ToDictionary(
kvp => kvp.Key,
kvp =>
{
var labelCount = kvp.Value.Length;
var meanConfusionMatrix = kvp.Value.Select(cm => cm.GetMean()).ToArray();
return GetConfusionMatrix(
Util.ArrayInit(labelCount, labelCount, (i, j) => meanConfusionMatrix[i][j]),
mapping,
ConfusionMatrixRowLabel,
ConfusionMatrixColLabel,
asPercentages);
}));
}
/// <summary>
/// Gets conditional probability table plot data.
/// </summary>
/// <param name="cpt">
/// The array representing the conditional probability table.
/// </param>
/// <param name="mapping">
/// The data mapping.
/// </param>
/// <param name="rowLabel">
/// The row label.
/// </param>
/// <param name="columnLabel">
/// The column label.
/// </param>
/// <returns>
/// The conditional probability table plot data.
/// </returns>
public static Dictionary <string, PointWithBounds <string>[]> GetCptWithBounds(
Dirichlet[] cpt,
CrowdDataMapping mapping,
string rowLabel,
string columnLabel)
{
Debug.Assert(
cpt.Length == mapping.LabelCount && cpt.All(mat => mat.Dimension == mapping.LabelCount),
"Inconsistent arguments");
var result = new Dictionary <string, PointWithBounds <string>[]>();
var labelIndexToString = mapping.LabelIndexToString;
for (var i = 0; i < mapping.LabelCount; i++)
{
var rowName = $"{labelIndexToString[i]} ({rowLabel})";
result[rowName] = DirichletWithErrorBars(cpt[i], labelIndexToString);
}
return(result);
}
/// <summary>
/// Gets the prominent workers for a data set - i.e. those who have given many labels.
/// </summary>
/// <param name="data">
/// The data.
/// </param>
/// <param name="maxNumberWorkers">
/// The maximum number of workers for which to get metrics.
/// </param>
/// <returns>
/// The metrics.
/// </returns>
private static HashSet <string> GetProminentWorkers(CrowdData data, int maxNumberWorkers = 20)
{
var mapping = new CrowdDataMapping(
data,
LabelValuesToString);
var labelCounts = new Dictionary <string, object>();
var labelsGroupedByWorker = data.CrowdLabels.GroupBy(cd => cd.WorkerId);
foreach (var worker in labelsGroupedByWorker)
{
var workerId = worker.Key;
var workerLabels = worker.Distinct(CrowdData.WorkerTweetEqualityComparer.Instance)
.ToDictionary(lab => lab.TweetId, lab => lab.WorkerLabel);
var workerMetrics = GetMetrics(mapping, workerLabels);
labelCounts[workerId] = workerMetrics[Metric.Count];
}
return(new HashSet <string>(
labelCounts.OrderByDescending(kvp => (int)kvp.Value).Take(maxNumberWorkers).Select(kvp => kvp.Key)));
}
/// <summary>
/// Initializes a new instance of the <see cref="BiasedWorkerModelRunner"/> class.
/// </summary>
/// <param name="dataMapping">
/// The data mapping.
/// </param>
/// <param name="model">
/// The model.
/// </param>
/// <param name="trainingRunner">
/// The training runner.
/// </param>
public BiasedWorkerModelRunner(CrowdDataMapping dataMapping, BiasedWorkerModel model, ModelRunnerBase trainingRunner = null)
: base(dataMapping, model, trainingRunner)
{
}
/// <inheritdoc />
/// <param name="dataMapping">
/// The data mapping
/// </param>
/// <param name="model">
/// The model.
/// </param>
/// <param name="trainingRunner">
/// The training Runner. This should be null if
/// (a) we are running training, or
/// (b) we are running validation and the model has no training
/// </param>
protected ModelRunnerBase(CrowdDataMapping dataMapping, ModelBase model, ModelRunnerBase trainingRunner = null)
: base(dataMapping)
{
this.Model = model;
this.TrainingRunner = trainingRunner;
}
/// <summary>
/// Initializes a new instance of the <see cref="HonestWorkerRunner"/> class.
/// </summary>
/// <param name="dataMapping">
/// The data mapping.
/// </param>
/// <param name="model">
/// The model.
/// </param>
/// <param name="trainingRunner">
/// The training runner.
/// </param>
public HonestWorkerRunner(CrowdDataMapping dataMapping, HonestWorkerModel model, ModelRunnerBase trainingRunner = null)
: base(dataMapping, model, trainingRunner)
{
}
/// <summary>
/// Initializes a new instance of the <see cref="RunnerBase"/> class.
/// </summary>
/// <param name="dataMapping">
/// The mapping between data and indices.
/// </param>
protected RunnerBase(CrowdDataMapping dataMapping)
{
this.DataMapping = dataMapping;
}
/// <summary>
/// Gets the metrics.
/// </summary>
/// <param name="dataMapping">
/// The data mapping.
/// </param>
/// <param name="predictions">
/// The predictions.
/// </param>
/// <param name="trueLabels">
/// The true labels. If null, then the gold labels are used.
/// </param>
/// <returns>
/// The dictionary of metric values.
/// </returns>
public static Dictionary <Metric, object> GetMetrics(CrowdDataMapping dataMapping, Dictionary <string, int> predictions, Dictionary <string, int> trueLabels = null)
{
var result = new Dictionary <Metric, object>();
var labelCount = dataMapping.LabelCount;
var data = dataMapping.Data as CrowdDataWithText;
var confusionMatrix = Util.ArrayInit(labelCount, labelCount, (i, j) => 0.0);
var tweetMatrix = Util.ArrayInit(labelCount, labelCount, (i, j) => new List <Tweet>());
var correct = 0.0;
var trueLabelCount = 0;
if (trueLabels == null)
{
trueLabels = dataMapping.Data.GoldLabels;
}
foreach (var kvp in trueLabels)
{
var trueLabel = dataMapping.LabelValueToIndex[kvp.Value];
if (predictions.ContainsKey(kvp.Key))
{
trueLabelCount++;
var predictedLabel = dataMapping.LabelValueToIndex[predictions[kvp.Key]];
confusionMatrix[trueLabel, predictedLabel] = confusionMatrix[trueLabel, predictedLabel] + 1.0;
if (data != null)
{
if (data.Tweets.ContainsKey(kvp.Key))
{
tweetMatrix[trueLabel, predictedLabel].Add(data.Tweets[kvp.Key]);
}
}
if (trueLabel == predictedLabel)
{
correct++;
}
}
}
result[Metric.Count] = trueLabelCount;
result[Metric.Accuracy] = correct / trueLabelCount;
result[Metric.ConfusionMatrix] = confusionMatrix;
result[Metric.TweetMatrix] = tweetMatrix;
// Average recall
double sumRec = 0;
for (var i = 0; i < labelCount; i++)
{
double classSum = 0;
for (var j = 0; j < labelCount; j++)
{
classSum += confusionMatrix[i, j];
}
sumRec += confusionMatrix[i, i] / classSum;
}
result[Metric.AverageRecall] = sumRec / labelCount;
return(result);
}