internal int GetParaphraseCount(SLUFactory sluFactory)
{
AnnotatedQuestionActionEntry lastAction = null;
foreach (var action in ExplanationTurns)
{
if (lastAction != null && lastAction.Act.StartsWith("RequestExplanation"))
{
break;
}
lastAction = action;
}
if (lastAction == null)
{
throw new NotImplementedException();
}
if (isInform(lastAction, sluFactory))
{
return(1);
}
return(0);
}
private Dictionary <string, object>[] convertTurns(IEnumerable <AnnotatedQuestionActionEntry> turns, int turnOffset)
{
var sluFactory = new SLUFactory();
var result = new List <Dictionary <string, object> >();
var turnsArray = turns.ToArray();
for (var i = 0; i < turnsArray.Length; i += 2)
{
var machineAction = turnsArray[i];
var userAction = turnsArray[i + 1];
var representation = new Dictionary <string, object>();
result.Add(representation);
representation["turn_index"] = (i + turnOffset) / 2;
var output = new Dictionary <string, object>()
{
{ "time", machineAction.Time },
{ "transcript", machineAction.Text },
{ "dialog_acts", convertAct(machineAction.Act) }
};
var input = new Dictionary <string, object>()
{
{ "time", userAction.Time },
{ "chat", userAction.Text },
{ "chat_slu", getUserSlu(userAction, sluFactory) }
};
representation["output"] = output;
representation["input"] = input;
}
return(result.ToArray());
}
internal DialogContext(QuestionInfo topic, ResponseBase contextQuestion, IModel handlingModel, SLUFactory factory)
{
HandlingModel = handlingModel;
Factory = factory;
Topic = topic;
ContextQuestion = contextQuestion;
NextMachineOutput = contextQuestion;
}
internal int GetExplanationCount(SLUFactory sluFactory)
{
if (isInform(ExplanationTurns.Last(), sluFactory))
{
return(1);
}
return(0);
}
private bool isInform(AnnotatedQuestionActionEntry action, SLUFactory sluFactory)
{
var slu = sluFactory.GetBestDialogAct(UtteranceParser.Parse(action.Text));
if (slu is DontKnowAct || slu is NegateAct || slu is ChitChatAct || slu is AffirmAct)
{
return(false);
}
return(true);
}
static void MainParsing(string[] args)
{
var utterance = "yes, Obama is president of which state?";
var parsedUtterance = UtteranceParser.Parse(utterance);
var factory = new SLUFactory();
var bestAct = factory.GetBestDialogAct(parsedUtterance);
var acts = factory.GetDialogActs(parsedUtterance);
foreach (var act in acts)
{
Console.WriteLine(act);
}
}
private static void writeQuestionDataset()
{
var provider = Program.QuestionDialogProvider;
provider.Refresh();
var sluFactory = new SLUFactory();
var dialogCount = 0;
var turnCount = 0;
var explanationCount = 0;
var answerCount = 0;
var noAnswerCount = 0;
var parahpraseCount = 0;
var correctCount = 0;
var incorrectCount = 0;
var differentChoise = 0;
var invalid = 0;
var unwanted = 0;
var dialogIndexes = new List <int>();
var questions = new HashSet <string>();
for (var i = 0; i < provider.DialogCount; ++i)
{
var dialog = provider.GetDialog(i);
var annotation = dialog.Annotation;
switch (annotation)
{
case "invalid":
invalid += 1;
break;
case "unwanted_answer":
unwanted += 1;
break;
}
if (annotation == null || annotation == "invalid" || annotation == "unwanted_answer")
{
//we dont want this dialog
continue;
}
switch (annotation)
{
case "correct_answer":
correctCount += 1;
break;
case "different_choice":
differentChoise += 1;
incorrectCount += 1;
break;
case "no_answer":
noAnswerCount += 1;
break;
default:
incorrectCount += 1;
break;
}
questions.Add(dialog.Question);
dialogCount += 1;
turnCount += dialog.TurnCount;
explanationCount += dialog.GetExplanationCount(sluFactory);
parahpraseCount += dialog.GetParaphraseCount(sluFactory);
answerCount += dialog.GetAnswerCount();
dialogIndexes.Add(i);
}
Console.WriteLine("Dataset statistics");
w("Dialog count", dialogCount);
w("Turn count", turnCount);
w("Explanation count", explanationCount);
w("Answer count", answerCount);
w("Noanswer count", noAnswerCount);
w("Paraphrase count", parahpraseCount);
w("Correct count", correctCount);
w("Incorrect count", incorrectCount);
w("Different choice", differentChoise);
w("Invalid count", invalid);
w("Unwanted count", unwanted);
w("Question count", questions.Count);
Console.WriteLine();
var rnd = new Random(12345);
Shuffle(rnd, dialogIndexes);
var testRatio = 0.35;
var trainRatio = 0.5;
var testCount = (int)(dialogIndexes.Count * testRatio);
var trainCount = (int)(dialogIndexes.Count * trainRatio);
var devCount = dialogIndexes.Count - testRatio - trainRatio;
var testIndexes = dialogIndexes.Take(testCount).ToArray();
var trainIndexes = dialogIndexes.Skip(testCount).Take(trainCount).ToArray();
var devIndexes = dialogIndexes.Skip(testCount + trainCount).ToArray();
if (testIndexes.Intersect(trainIndexes).Any() || trainIndexes.Intersect(devIndexes).Any() || devIndexes.Intersect(testIndexes).Any())
{
throw new NotSupportedException("there is an error when spliting dataset");
}
writeQuestionDatasetTo("question_dialogs-test.json", testIndexes, provider);
writeQuestionDatasetTo("question_dialogs-train.json", trainIndexes, provider);
writeQuestionDatasetTo("question_dialogs-dev.json", devIndexes, provider);
Console.ReadLine();
}
private Dictionary <string, object> getUserSlu(AnnotatedQuestionActionEntry userAction, SLUFactory sluFactory)
{
var slu = sluFactory.GetBestDialogAct(UtteranceParser.Parse(userAction.Text));
if (slu is DontKnowAct || slu is NegateAct || slu is ChitChatAct || slu is AffirmAct)
{
return(slu.GetDialogAct().ToDictionary());
}
return(new Dictionary <string, object>()
{
{ "act", "Inform" },
{ "text", userAction.Text }
});
}
