private void Initialize(
string segmentedSource,
string segmentedTarget,
string alignment,
SegmentationMethod segmentationMethod,
string targetCode)
{
this.SegmentedSourceSentence = segmentedSource.Split(' ');
this.SegmentedTranslation = segmentedTarget.Split(' ');
this.SegmentedAlignmentSourceToTarget = TranslationPair.ParseAlignmentString(
alignment,
SegmentedSourceSentence.Length - 1,
SegmentedTranslation.Length - 1,
false);
this.SegmentedAlignmentTargetToSource = TranslationPair.ParseAlignmentString(
alignment,
SegmentedTranslation.Length - 1,
SegmentedSourceSentence.Length - 1,
true);
this.AlignmentString = alignment;
this.Segmentation = segmentationMethod;
}
static int Main(string[] args)
{
CanvasCommon.Utilities.LogCommandLine(args);
string inFile = null;
string outFile = null;
bool needHelp = false;
bool isGermline = false;
string bedPath = null;
double alpha = Segmentation.DefaultAlpha;
SegmentSplitUndo undoMethod = SegmentSplitUndo.None;
SegmentationMethod partitionMethod = SegmentationMethod.Wavelets;
int maxInterBinDistInSegment = 1000000;
OptionSet p = new OptionSet()
{
{ "i|infile=", "input file - usually generated by CanvasClean", v => inFile = v },
{ "o|outfile=", "text file to output", v => outFile = v },
{ "h|help", "show this message and exit", v => needHelp = v != null },
{ "a|alpha=", "alpha parameter to CBS. Default: " + alpha, v => alpha = float.Parse(v) },
{ "m|method=", "segmentation method (Wavelets/CBS). Default: " + partitionMethod, v => partitionMethod = (SegmentationMethod)Enum.Parse(typeof(SegmentationMethod), v) },
{ "s|split=", "CBS split method (None/Prune/SDUndo). Default: " + undoMethod, v => undoMethod = (SegmentSplitUndo)Enum.Parse(typeof(SegmentSplitUndo), v) },
{ "b|bedfile=", "bed file to exclude (don't span these intervals)", v => bedPath = v },
{ "g|germline", "flag indicating that input file represents germline genome", v => isGermline = v != null },
{ "d|maxInterBinDistInSegment=", "the maximum distance between adjacent bins in a segment (negative numbers turn off splitting segments after segmentation). Default: " + maxInterBinDistInSegment, v => maxInterBinDistInSegment = int.Parse(v) },
};
List <string> extraArgs = p.Parse(args);
if (needHelp)
{
ShowHelp(p);
return(0);
}
if (inFile == null || outFile == null)
{
ShowHelp(p);
return(0);
}
if (!File.Exists(inFile))
{
Console.WriteLine("CanvasPartition.exe: File {0} does not exist! Exiting.", inFile);
return(1);
}
if (!string.IsNullOrEmpty(bedPath) && !File.Exists(bedPath))
{
Console.WriteLine("CanvasPartition.exe: File {0} does not exist! Exiting.", bedPath);
return(1);
}
// no command line parameter for segmentation method
Segmentation SegmentationEngine = new Segmentation(inFile, bedPath, maxInterBinDistInSegment: maxInterBinDistInSegment);
SegmentationEngine.Alpha = alpha;
SegmentationEngine.UndoMethod = undoMethod;
SegmentationEngine.SegmentGenome(outFile, partitionMethod, isGermline);
return(0);
}
public TranslationPair(
string translation,
string segmentedSource,
string segmentedTarget,
string alignment,
SegmentationMethod segmentationMethod,
string targetCode)
{
this.Translation = translation;
this.Initialize(segmentedSource, segmentedTarget, alignment, segmentationMethod, targetCode);
}
public TranslationPair(
string segmentedSource,
string translationAndAlignment,
SegmentationMethod segmentationMethod,
string targetLanguage)
{
var lastSeparator = translationAndAlignment.LastIndexOf("|||");
var segmentedTranslation = translationAndAlignment.Substring(0, lastSeparator - 1);
this.RawTranslation = segmentedTranslation;
var alignment = translationAndAlignment.Substring(lastSeparator + "||| ".Length);
this.Initialize(segmentedSource, segmentedTranslation, alignment, segmentationMethod, targetLanguage);
}
public void SegmentGenome(string outPath, SegmentationMethod method, bool isGermline)
{
switch (method)
{
case SegmentationMethod.Wavelets:
default:// use Wavelets if CBS is not selected
Console.WriteLine("{0} Running Wavelet Partitioning", DateTime.Now);
this.Wavelets(isGermline, verbose: 2);
break;
case SegmentationMethod.CBS:
Console.WriteLine("{0} Running CBS Partitioning", DateTime.Now);
this.CBS(verbose: 2);
break;
}
Console.WriteLine("{0} Write CanvasPartition results:", DateTime.Now);
this.WriteCanvasPartitionResults(outPath);
Console.WriteLine("{0} CanvasPartition results written out", DateTime.Now);
}
public void SegmentGenome(string outPath, SegmentationMethod method, bool isGermline)
{
switch (method)
{
case SegmentationMethod.Wavelets:
default: // use Wavelets if CBS is not selected
Console.WriteLine("{0} Running Wavelet Partitioning", DateTime.Now);
this.Wavelets(isGermline, verbose: 2);
break;
case SegmentationMethod.CBS:
Console.WriteLine("{0} Running CBS Partitioning", DateTime.Now);
this.CBS(verbose: 2);
break;
}
Console.WriteLine("{0} Write CanvasPartition results:", DateTime.Now);
this.WriteCanvasPartitionResults(outPath);
Console.WriteLine("{0} CanvasPartition results written out", DateTime.Now);
}
internal static void WriteTranslationToDb(
string sourceText,
TranslationPair translation,
string model,
SegmentationMethod segmentationMethod,
string targetLanguage)
{
TranslationDbHelper.shortTermMtStorage.GetOrAdd(new Tuple <string, string>(sourceText, model), translation);
if (OpusCatMTEngineSettings.Default.CacheMtInDatabase)
{
try
{
TranslationDbHelper.WriteTranslationToSqliteDb(sourceText, translation, model, segmentationMethod, targetLanguage);
}
catch (Exception ex)
{
Log.Error(ex.ToString());
TranslationDbHelper.SetupTranslationDb();
}
}
}
private static void WriteTranslationToSqliteDb(
string sourceText,
TranslationPair translation,
string model,
SegmentationMethod segmentationMethod,
string targetLanguage)
{
var translationDb = new FileInfo(HelperFunctions.GetOpusCatDataPath(OpusCatMTEngineSettings.Default.TranslationDBName));
if (translationDb.Length == 0)
{
translationDb.Delete();
}
if (!translationDb.Exists)
{
TranslationDbHelper.CreateTranslationDb();
}
using (var m_dbConnection = new SQLiteConnection($"Data Source={translationDb};Version=3;"))
{
m_dbConnection.Open();
using (SQLiteCommand insert =
new SQLiteCommand(
"INSERT or REPLACE INTO translations (sourcetext, translation, segmentedsource, segmentedtranslation, alignment, model, additiondate, segmentationmethod, targetlanguage) VALUES (@sourcetext,@translation,@segmentedsource,@segmentedtranslation,@alignment,@model,CURRENT_TIMESTAMP,@segmentationmethod,@targetlanguage)", m_dbConnection))
{
insert.Parameters.Add(new SQLiteParameter("@sourcetext", sourceText));
insert.Parameters.Add(new SQLiteParameter("@translation", translation.Translation));
insert.Parameters.Add(new SQLiteParameter("@segmentedsource", String.Join(" ", translation.SegmentedSourceSentence)));
insert.Parameters.Add(new SQLiteParameter("@segmentedtranslation", String.Join(" ", translation.SegmentedTranslation)));
insert.Parameters.Add(new SQLiteParameter("@alignment", translation.AlignmentString));
insert.Parameters.Add(new SQLiteParameter("@model", model));
insert.Parameters.Add(new SQLiteParameter("@segmentationmethod", segmentationMethod.ToString()));
insert.Parameters.Add(new SQLiteParameter("@targetlanguage", targetLanguage));
insert.ExecuteNonQuery();
}
}
}
public MarianBatchTranslator(
string modelDir,
IsoLanguage sourceLang,
IsoLanguage targetLang,
SegmentationMethod segmentation,
bool includePlaceholderTags,
bool includeTagPairs)
{
this.SourceCode = sourceLang.OriginalCode;
this.TargetCode = targetLang.OriginalCode;
this.segmentation = segmentation;
this.includePlaceholderTags = includePlaceholderTags;
this.includeTagPairs = includeTagPairs;
this.modelDir = new DirectoryInfo(modelDir);
this.SystemName = $"{this.SourceCode}-{this.TargetCode}_" + this.modelDir.Name;
//Check if batch.yml exists, if not create it from decode.yml
var batchYaml = this.modelDir.GetFiles("batch.yml");
if (batchYaml.Length == 0)
{
var decoderYaml = this.modelDir.GetFiles("decoder.yml").Single();
var deserializer = new Deserializer();
var decoderSettings = deserializer.Deserialize <MarianDecoderConfig>(decoderYaml.OpenText());
decoderSettings.miniBatch = "16";
decoderSettings.log = Path.Combine(this.modelDir.FullName, "batch.log");
decoderSettings.alignment = "hard";
var serializer = new Serializer();
var configPath = Path.Combine(this.modelDir.FullName, "batch.yml");
using (var writer = File.CreateText(configPath))
{
serializer.Serialize(writer, decoderSettings, typeof(MarianDecoderConfig));
}
}
}
internal static WordsearchSolutionEvaluator EvaluateWordsearchBitmap(Bitmap wordsearchBitmap, string[] wordsToFind,
Dictionary<string, List<WordPosition>> correctSolutions, SegmentationAlgorithm segmentationAlgorithm,
bool segmentationRemoveSmallRowsAndCols, SegmentationMethod segmentationMethod,
Classifier probabilisticRotationCorrectionClassifier, Classifier classifier, Solver wordsearchSolver)
{
/*
* Wordsearch Segmentation
*/
Segmentation segmentation = segmentationAlgorithm.Segment(wordsearchBitmap);
//Remove erroneously small rows and columns from the segmentation if that option is specified
if(segmentationRemoveSmallRowsAndCols)
{
segmentation = segmentation.RemoveSmallRowsAndCols();
}
/*
* Wordsearch Rotation Correction
*/
WordsearchRotation originalRotation;
//If we're using fixed row & col width
if (segmentationMethod == SegmentationMethod.FixedWidth)
{
originalRotation = new WordsearchRotation(wordsearchBitmap, segmentation.NumRows, segmentation.NumCols);
}
else //Otherwise we're using varied row/col width segmentation, use the Segmentation object
{
originalRotation = new WordsearchRotation(wordsearchBitmap, segmentation);
}
WordsearchRotation rotatedWordsearch = WordsearchRotationCorrection.CorrectOrientation(originalRotation, probabilisticRotationCorrectionClassifier);
Bitmap rotatedImage = rotatedWordsearch.Bitmap;
//If the wordsearch has been rotated
if (rotatedImage != wordsearchBitmap)
{
//Update the segmentation
//If the wordsearch rotation won't have been passed a segmentation
if (segmentationMethod == SegmentationMethod.FixedWidth)
{
//Make a new fixed width segmentation from the WordsearchRotation
segmentation = new Segmentation(rotatedWordsearch.Rows, rotatedWordsearch.Cols,
rotatedImage.Width, rotatedImage.Height);
}
else
{
//Use the rotated segmentation
segmentation = rotatedWordsearch.Segmentation;
}
}
/*
* Classification
*/
//Split image up into individual characters
Bitmap[,] rawCharImgs = null;
//If we're using fixed row & col width
if (segmentationMethod == SegmentationMethod.FixedWidth)
{
ResizeBicubic resize = new ResizeBicubic(Constants.CHAR_WITH_WHITESPACE_WIDTH * segmentation.NumCols,
Constants.CHAR_WITH_WHITESPACE_HEIGHT * segmentation.NumRows);
Bitmap resizedImage = resize.Apply(rotatedImage);
rawCharImgs = SplitImage.Grid(resizedImage, segmentation.NumRows, segmentation.NumCols);
//Resized image no longer required
resizedImage.Dispose();
}
else //Otherwise we're using varied row/col width segmentation
{
rawCharImgs = SplitImage.Segment(rotatedImage, segmentation);
//If the Segmentation Method is to resize the raw char imgs, resize them
if (segmentationMethod == SegmentationMethod.VariedWidthWithResize)
{
ResizeBicubic resize = new ResizeBicubic(Constants.CHAR_WITH_WHITESPACE_WIDTH, Constants.CHAR_WITH_WHITESPACE_HEIGHT);
for (int i = 0; i < rawCharImgs.GetLength(0); i++)
{
for (int j = 0; j < rawCharImgs.GetLength(1); j++)
{
//Only do the resize if it isn't already that size
if (rawCharImgs[i, j].Width != Constants.CHAR_WITH_WHITESPACE_WIDTH
|| rawCharImgs[i, j].Height != Constants.CHAR_WITH_WHITESPACE_HEIGHT)
{
Bitmap orig = rawCharImgs[i, j];
rawCharImgs[i, j] = resize.Apply(orig);
//Remove the now unnecessary original/not resized image
orig.Dispose();
}
}
}
}
}
//Full sized rotated image no longer required
rotatedImage.Dispose();
//Get the part of the image that actually contains the character (without any whitespace)
Bitmap[,] charImgs = CharImgExtractor.ExtractAll(rawCharImgs);
//Raw char img's are no longer required
rawCharImgs.ToSingleDimension().DisposeAll();
//Perform the classification on all of the images (returns probabilities for each possible class)
double[][][] classifierOutput = classifier.Classify(charImgs);
//Actual images of the characters are no longer required
charImgs.ToSingleDimension().DisposeAll();
/*
* Solve Wordsearch
*/
Solution solution = wordsearchSolver.Solve(classifierOutput, wordsToFind);
/*
* Evaluate the Proposed Solution
*/
WordsearchSolutionEvaluator evaluator = new WordsearchSolutionEvaluator(solution, correctSolutions);
return evaluator;
}
private static double Evaluate(List<Image> images, SegmentationAlgorithm detectionSegmentationAlgorithm,
bool detectionSegmentationRemoveSmallRowsAndCols, SegmentationAlgorithm segmentationAlgorithm,
bool segmentationRemoveSmallRowsAndCols, SegmentationMethod segmentationMethod,
Classifier probabilisticRotationCorrectionClassifier, Classifier classifier, Solver wordsearchSolver)
{
DefaultLog.Info("Evaluating Full System . . .");
int numCorrect = 0;
List<WordsearchSolutionEvaluator> evaluators = new List<WordsearchSolutionEvaluator>();
foreach(Image image in images)
{
//Register an interest in the Bitmap of the image
image.RegisterInterestInBitmap();
/*
* Wordsearch Detection
*/
Tuple<List<IntPoint>, Bitmap> wordsearchImageTuple = DetectionAlgorithm.ExtractBestWordsearch(image.Bitmap, detectionSegmentationAlgorithm, detectionSegmentationRemoveSmallRowsAndCols);
//Original wordsearch image is no longer required
image.DeregisterInterestInBitmap();
//If the system failed to find anything remotely resembling a wordsearch, fail now
if(wordsearchImageTuple == null)
{
continue;
}
//Get the words to look for later from this image & the correct solutions
string[] wordsToFind = null; //Requires default, but won't even get used
Dictionary<string, List<WordPosition>> correctSolutions = null;
//If the image contains more than one wordsearch, we need to work out which one has been found
if(image.WordsearchImages.Length > 1)
{
List<IntPoint> coordinates = wordsearchImageTuple.Item1;
bool found = false;
//Select the wordsearch found using the algorithm for checking if the returned wordsearch is correct in EvaluateWordsearchDetection
foreach(WordsearchImage wordsearchImage in image.WordsearchImages)
{
//If it's this wordsearch
if(EvaluateWordsearchDetection.IsWordsearch(coordinates, wordsearchImage))
{
wordsToFind = wordsearchImage.Wordsearch.Words;
correctSolutions = wordsearchImage.Wordsearch.Solutions;
found = true;
break;
}
}
//If this isn't one of the wordsearches in the image, then fail now
if(!found)
{
//Clean up
wordsearchImageTuple.Item2.Dispose();
continue;
}
}
else //Otherwise just use the one wordsearch that's in the image
{
wordsToFind = image.WordsearchImages[0].Wordsearch.Words;
correctSolutions = image.WordsearchImages[0].Wordsearch.Solutions;
}
Bitmap extractedImage = wordsearchImageTuple.Item2;
/*
* Image Segmentation onwards happen in EvaluateWordsearchBitmap
*/
WordsearchSolutionEvaluator evaluator = EvaluateWordsearchBitmap(extractedImage, wordsToFind, correctSolutions,
segmentationAlgorithm, segmentationRemoveSmallRowsAndCols, segmentationMethod,
probabilisticRotationCorrectionClassifier, classifier, wordsearchSolver);
//Clean up
extractedImage.Dispose();
//Log Evaluation
evaluators.Add(evaluator);
DefaultLog.Info(evaluator.ToString());
if(evaluator.Correct)
{
numCorrect++;
}
}
DefaultLog.Info("System found all words correctly for {0} / {1} Images correctly", numCorrect, images.Count);
//Calculate some extra statistics
int numWordsearchesNoWordsFound = 0;
int numDidntReachEvaluation = images.Count - evaluators.Count;
double fMeasureSum = 0;
int numValidFMeasures = 0;
foreach (WordsearchSolutionEvaluator evaluator in evaluators)
{
//If no words were found correctly
if(evaluator.TruePositive == 0)
{
numWordsearchesNoWordsFound++;
}
//If there was a valid F-Measure
if(!double.IsNaN(evaluator.FMeasure))
{
fMeasureSum += evaluator.FMeasure;
numValidFMeasures++;
}
}
DefaultLog.Info("In {0} wordsearches no words were found correctly at all", numWordsearchesNoWordsFound);
DefaultLog.Info("{0} wordsearch images got discarded before reaching the evaluation stage", numDidntReachEvaluation);
DefaultLog.Info("Average F-Measure (when not NaN): {0}", fMeasureSum / numValidFMeasures);
DefaultLog.Info("Full System Evaluation Completed");
return (double)numCorrect / images.Count;
}