private void Level1Classification(List <DocumentData> TrainData, List <DocumentData> TestData)
{
var selectedFeatures = AppDataCenter.Singleton.FeaturesSelected.Select((a, i) => new { item = a, index = i })
.Where(a => a.item.IsSelected).Select(a => a.index).ToArray();
// selectedFeatures = new int[] { 37 };
ClassifierService.Service.InitSelectFeatures(selectedFeatures);
ScalingFactors scales = null;
// if(AppDataCenter.Singleton.PathToScale != null)
// {
// scales = FeatureListHolder.loadScalesFromFile(ClassifierService.Service.LocalFeatures, AppDataCenter.Singleton.PathToScale);
// }
// AppDataCenter.Singleton.AddConsoleMessage("Building Model");
AppDataCenter.Singleton.ChosenFeatures = ClassifierService.Service.LocalFeatures;
AppDataCenter.Singleton.UpdateFeatureVertHorzSelection();
AppDataCenter.Singleton.EntireModelStats = ClassifierService.Service.CreateModel(AppDataCenter.Singleton.ChosenFields
, TrainData, TestData, Configuration, m_featuresCalculate, AppDataCenter.Singleton.AddConsoleMessage, TuneScale, AppDataCenter.Singleton.PathToScale);
// if (AppDataCenter.Singleton.PathToScale != null)
// {
// FeatureListHolder.SaveScalesToFile(ClassifierService.Service.LocalFeatures, AppDataCenter.Singleton.PathToScale,scales);
// }
double[] confidenceOut;
List <ReportResultItem> ReoprtResults = new List <ReportResultItem>();
Array.ForEach(AppDataCenter.Singleton.ChosenFields,
new Action <string>(a =>
{
ReoprtResults.Add(new ReportResultItem(a));
}));
ReportResultItem NullResult = new ReportResultItem("");
ReoprtResults.Add(NullResult);
AppDataCenter.Singleton.RejectedFields.Clear();
ClassifierService.Service.level1RunTestForLevel2(TrainData, AppDataCenter.Singleton.ChosenFields, AppDataCenter.Singleton.AddConsoleMessage);
// add printing of train data sucss
ConcurrentBag <FieldReportItem> fieldBag = new ConcurrentBag <FieldReportItem>();
AppDataCenter.Singleton.AddConsoleMessage("Running Test on : " + TestData.Count + " Pages");
int index = 0;
// int result;
foreach (DocumentData doc1 in TestData)
{
var startTimeFeatures = DateTime.Now;
// AppDataCenter.Singleton.AddConsoleMessage("Testing Page : " + docnum++);
// List<FieldData> fields = ClassifierService.Service.GetFieldsOfDoc(FieldsToTest, doc1);
TimeSpan FeatureCalcSpan = new TimeSpan(0);
/* foreach (CandidateData field in doc1.Candidates)
* {
* result = 0;
* startTimeFeatures = DateTime.Now;
* if (m_useLastRuntimeData == true)
* {
* foreach (var featureIndex in ClassifierService.Service.LocalFeatures
* .Select((x, i) => AppDataCenter.Singleton.IsFeatureCalculate(x.Name) == true ? i : -1)
* .Where(x => x >= 0))
* {
* field.Features[featureIndex] = ClassifierService.Service.GetFieldFeature(doc1, field, featureIndex);
* }
* }
* else
* {
* field.Features = ClassifierService.Service.GetFieldFeatures(doc1, field);
* }
* FeatureCalcSpan += (DateTime.Now - startTimeFeatures);
* result = ClassifierService.Service.GetDescition(field.Features, out confidenceOut);
*
*
* // field.AccordConfidance = (ClassifierService.Service.NormalizaedConfidence(confidenceOut));
* field.AccordConfidance = (confidenceOut);
* }*/
List <int> featursIndexFilter = new List <int>();
// Get the index of the feature we need to calculate - all other features will be use the value already exist in the field
ClassifierService.Service.getDocFeatures(doc1, null, null, AppDataCenter.Singleton.ChosenFields, featursIndexFilter, false, null, true, true);
double NumGolden = 0;
double NumMatched = 0;
foreach (var r in AppDataCenter.Singleton.ChosenFields.Select((x, i) => new { Value = x, Index = i }))
{
// if (doc1.Candidates.Where(a => a.NameFromTypist == r.Value).ToList().Count() <= 0) continue;
NumGolden++;
CandidateData fieldCandidate = doc1.Candidates.OrderByDescending(a => a.AccordConfidance[r.Index]).FirstOrDefault();
FieldReportItem fieldReportItem = new FieldReportItem(r.Value, r.Index, fieldCandidate, doc1);
if (fieldReportItem.IsMatch)
{
NumMatched++;
}
/* }
* else
* {
* fieldReportItem.isRejected = true;
*
* }*/
if (!fieldReportItem.IsMatch)
{
Trace.WriteLine("Doc : " + doc1.DocumentName + " ,Confidance : " + fieldCandidate.AccordConfidance[r.Index] + ",Field : " + r.Value + " ,Recognized : " + fieldCandidate.NameFromTypist + " content : " + fieldCandidate.Content + (fieldReportItem.IsMatch ? " True" : " False"));
}
fieldBag.Add(fieldReportItem);
}
AppDataCenter.Singleton.AddConsoleMessage(index++ + " Page : " + doc1.DocumentName + " Getting Features (sec) : " + FeatureCalcSpan.TotalSeconds + " success : " + Math.Round(NumMatched / NumGolden * 100) + "%");
}
AppDataCenter.Singleton.AddConsoleMessage("Update results");
AppDataCenter.Singleton.EntireModelStats.testresults = fieldBag.ToList();
ReportData.noOfFields = fieldBag.Count();
ReportData.NoOfPages = TestData.Count;
ReportData.fpCount = fieldBag.Where(a => a.IsFP).Count();
ReportData.matchCount = fieldBag.Where(a => a.IsMatch).Count();
ReportData.rejectCount = fieldBag.Where(a => a.IsRejected).Count();
foreach (FieldReportItem fieldresult in fieldBag.Where(a => a.IsMatch))
{
ReportData.QualityMatch.Add(fieldresult.MatchQuality);
}
List <FieldReportItem> FPList = fieldBag.Where(a => a.IsFP).ToList().OrderBy(b => b.Field.NameFromTypist ?? "").ThenByDescending(a => a.Field.AccordConfidance.Max()).ToList();
foreach (var r in AppDataCenter.Singleton.ChosenFields)
{
List <FieldReportItem> FPfromGolden = FPList.Where(a => a.ExpectedField.NameFromTypist == r).ToList();
List <FieldReportItem> FPnotFromGolden = FPList.Where(a => a.ExpectedField.NameFromTypist != r && a.MatchedName == r).ToList();
double success = (double)fieldBag.Where(a => a.Field.NameFromTypist == r && a.IsMatch).Count() / (double)fieldBag.Where(a => a.Field.NameFromTypist == r).Count() * 100;
int countNum = fieldBag.Where(a => a.Field.NameFromTypist == r).Count();
Trace.WriteLine("*********** The field : " + r + " success : " + Math.Round(success, 1) + "% Count : " + countNum + " *************");
}
AppDataCenter.Singleton.NotifyChange(NotifyGroup.ScatterData);
}
private void Level3Classification(List <DocumentData> TrainData, List <DocumentData> TestData)
{
AppDataCenter.Singleton.AddConsoleMessage("\nStarting Level3 ***************\n");
ClassifierService.Service.InitFeatureslevel3(AppDataCenter.Singleton.ChosenFields);
ClassifierService.Service.buildModelLevel3(AppDataCenter.Singleton.ChosenFields
, TrainData, Configuration, AppDataCenter.Singleton.AddConsoleMessage);
ConcurrentBag <FieldReportItem> fieldBag = new ConcurrentBag <FieldReportItem>();
AppDataCenter.Singleton.AddConsoleMessage("Getting results from level 1 to be used in level 3");
TestData.AsParallel().ForAll(doc =>
{
doc.lastResultsCandidates = new Dictionary <string, CandidateData>();
for (int i = 0; i < AppDataCenter.Singleton.ChosenFields.Length; i++)
{
CandidateData releventCandidate = doc.Candidates.Where(a => a.AccordConfidance[i] == doc.Candidates.Select(b => b.AccordConfidance[i]).Max()).FirstOrDefault();
doc.lastResultsCandidates.Add(AppDataCenter.Singleton.ChosenFields[i], releventCandidate);
}
});
AppDataCenter.Singleton.AddConsoleMessage("Running Test on : " + TestData.Count + " Pages");
int index = 0;
int result;
double[] confidenceOut;
foreach (DocumentData doc1 in TestData)
{
var startTimeFeatures = DateTime.Now;
TimeSpan FeatureCalcSpan = new TimeSpan(0);
foreach (CandidateData field in doc1.CandidatesForStage3)
{
{
field.Features3 = ClassifierService.Service.GetFieldFeatures3(doc1, field);
}
FeatureCalcSpan += (DateTime.Now - startTimeFeatures);
result = ClassifierService.Service.GetDescition3(field.Features3, out confidenceOut);
// field.AccordConfidance = (ClassifierService.Service.NormalizaedConfidence(confidenceOut));
field.AccordConfidance3 = (confidenceOut);
}
double NumGolden = 0;
double NumMatched = 0;
foreach (var r in AppDataCenter.Singleton.ChosenFields.Select((x, i) => new { Value = x, Index = i }))
{
// if (doc1.Candidates.Where(a => a.NameFromTypist == r.Value).ToList().Count() <= 0) continue;
NumGolden++;
CandidateData fieldCandidate = doc1.CandidatesForStage3.OrderByDescending(a => a.AccordConfidance[r.Index]).Take(7).OrderByDescending(a => a.AccordConfidance3[r.Index]).FirstOrDefault();
FieldReportItem fieldReportItem = new FieldReportItem(r.Value, r.Index, fieldCandidate, doc1);
if (fieldReportItem.IsMatch)
{
NumMatched++;
}
/* }
* else
* {
* fieldReportItem.isRejected = true;
*
* }*/
if (!fieldReportItem.IsMatch)
{
Trace.WriteLine("Doc : " + doc1.DocumentName + " ,Confidance : " + fieldCandidate.AccordConfidance3[r.Index] + ",Field : " + r.Value + " ,Recognized : " + fieldCandidate.NameFromTypist + " content : " + fieldCandidate.Content + (fieldReportItem.IsMatch ? " True" : " False"));
}
fieldBag.Add(fieldReportItem);
}
AppDataCenter.Singleton.AddConsoleMessage(index++ + " Page : " + doc1.DocumentName + " Getting Features (sec) : " + FeatureCalcSpan.TotalSeconds + " success : " + Math.Round(NumMatched / NumGolden * 100, 1) + "%");
}
AppDataCenter.Singleton.AddConsoleMessage("Update results");
// AppDataCenter.Singleton.EntireModelStats.testresults = fieldBag.ToList();
double noOfFields = fieldBag.Count();
double fpCount = fieldBag.Where(a => a.IsFP).Count();
double matchCount = fieldBag.Where(a => a.IsMatch).Count();
double rejectCount = fieldBag.Where(a => a.IsRejected).Count();
List <FieldReportItem> FPList = fieldBag.Where(a => a.IsFP).ToList().OrderBy(b => b.Field.NameFromTypist ?? "").ThenByDescending(a => a.Field.AccordConfidance.Max()).ToList();
foreach (var r in AppDataCenter.Singleton.ChosenFields)
{
List <FieldReportItem> FPfromGolden = FPList.Where(a => a.ExpectedField.NameFromTypist == r).ToList();
List <FieldReportItem> FPnotFromGolden = FPList.Where(a => a.ExpectedField.NameFromTypist != r && a.MatchedName == r).ToList();
double success = (double)fieldBag.Where(a => a.Field.NameFromTypist == r && a.IsMatch).Count() / (double)fieldBag.Where(a => a.Field.NameFromTypist == r).Count() * 100;
int countNum = fieldBag.Where(a => a.Field.NameFromTypist == r).Count();
Trace.WriteLine("*********** The field : " + r + " success : " + Math.Round(success, 1) + "% Count : " + countNum + " *************");
}
AppDataCenter.Singleton.AddConsoleMessage("Total result level 3 , sucsss : " + Math.Round(matchCount / noOfFields * 100, 2) + "%");
}
public void Level2Classification(List <DocumentData> TrainData, List <DocumentData> TestData)
{
List <int[]> groups = ClassifierService.Service.BuildGroupsOfFeatures(AppDataCenter.Singleton.ChosenFields);
int[] optionsPerSol = new int[AppDataCenter.Singleton.ChosenFields.Length];
for (int i = 0; i < optionsPerSol.Length; i++)
{
if (groups.SelectMany(a => a).Contains(i))
{
optionsPerSol[i] = 2;
}
else
{
optionsPerSol[i] = 1;
}
}
ClassifierService.Service.InitFeatureslevel2(groups);
ClassifierService.Service.Prepareraindatafalsegrouping(TrainData, optionsPerSol, 0, AppDataCenter.Singleton.AddConsoleMessage);
// prepare candidates for not group with test Data and move them into the build
ClassifierService.Service.buildModelLevel2(AppDataCenter.Singleton.ChosenFields, TrainData, groups, Configuration, AppDataCenter.Singleton.AddConsoleMessage);
ClassifierService.Service.PrepareCandidateslevel2Test(TestData, optionsPerSol, 0, AppDataCenter.Singleton.AddConsoleMessage);
TestData.ForEach(doc =>
{
solutionData chosenSolution = null;
double lastProb = -100;
doc.solutionCandidates.ToList().ForEach(
sol =>
{
double conf;
ClassifierService.Service.GetDescitionLevel2(sol.features, out conf);
sol.Confidance = conf;
if (conf > lastProb)
{
chosenSolution = sol;
lastProb = conf;
}
}
);
doc.chosenSolution = chosenSolution;
});
ConcurrentBag <FieldReportItem> fieldBag = new ConcurrentBag <FieldReportItem>();
int index = 0;
double TotalGolden = 0;
double TotalMatched = 0;
foreach (DocumentData doc1 in TestData)
{
var startTimeFeatures = DateTime.Now;
TimeSpan FeatureCalcSpan = new TimeSpan(0);
double NumGolden = 0;
double NumMatched = 0;
foreach (var r in AppDataCenter.Singleton.ChosenFields.Select((x, i) => new { Value = x, Index = i }))
{
// if (doc1.Candidates.Where(a => a.NameFromTypist == r.Value).ToList().Count() <= 0) continue;
NumGolden++;
FieldReportItem fieldReportItem = new FieldReportItem(r.Value, r.Index, doc1.chosenSolution.offeredSolution[r.Index], doc1);
if (fieldReportItem.IsMatch)
{
NumMatched++;
}
/* }
* else
* {
* fieldReportItem.isRejected = true;
*
* }*/
Trace.WriteLine("Doc : " + doc1.DocumentName + ",Field : " + r.Value + " ,Recognized : " + fieldReportItem.MatchedName + " content : " + doc1.chosenSolution.offeredSolution[r.Index].Content + "Match : " + (fieldReportItem.IsMatch ? " True" : " False"));
fieldBag.Add(fieldReportItem);
}
TotalGolden += NumGolden;
TotalMatched += NumMatched;
AppDataCenter.Singleton.AddConsoleMessage(index++ + " Page : " + doc1.DocumentName + " Getting Features (sec) : " + FeatureCalcSpan.TotalSeconds + " success : " + Math.Round(NumMatched / NumGolden * 100) + "%");
}
AppDataCenter.Singleton.AddConsoleMessage(" Total Matched : " + Math.Round(TotalMatched / TotalGolden * 100, 1) + "%");
// calculate success
// print result
}
