/// <summary>
/// Learns a program to extract a sequence of regions from a file.
/// </summary>
private static void LearnSequence()
{
var session = new SequenceSession();
// It is advised to learn a sequence with at least 2 examples because generalizing a sequence from a single element is hard.
// Also, we need to give positive examples continuously (i.e., we cannot skip any example).
var input =
SequenceSession.CreateStringRegion(
"United States\n Carrie Dodson 100\n Leonard Robledo 75\n Margaret Cook 320\n" +
"Canada\n Concetta Beck 350\n Nicholas Sayers 90\n Francis Terrill 2430\n" +
"New Zealand\n Nettie Pope 50\n Mack Beeson 1070");
// Suppose we want to extract all last names from the input string.
session.AddConstraints(
new SequenceExample(input, new[]
{
input.Slice(15, 21), // input => "Carrie"
input.Slice(34, 41), // input => "Leonard"
})
);
SequenceProgram topRankedProg = session.Learn();
if (topRankedProg == null)
{
Console.Error.WriteLine("Error: Learning fails!");
return;
}
foreach (StringRegion r in topRankedProg.Run(input))
{
string output = r != null ? r.Value : "null";
Console.WriteLine(output);
}
}
public List <string> RunSequence(string programAsString, string input)
{
//programAsString = programAsString.Replace("\\", "");
var program = Loader.Instance.Sequence.Load(programAsString);
var inputRegion = SequenceSession.CreateStringRegion(input);
var result = program.Run(inputRegion);
return(result.Select(region => region.Value).ToList());
}
private void HandleSequenceInitPacket(SequenceInitPacket packet)
{
var seqSession = new SequenceSession(packet);
if (packet.IsOperation)
{
seqSession.IsOperation = true;
seqSession.OperationGuid = packet.OperationGuid;
}
lock (_lockObj)
_activeSequenceSessions.Add(packet.SequenceGuid, seqSession);
}
public override void Run()
{
if (!Directory.Exists(ReportDirPath))
{
Directory.CreateDirectory(ReportDirPath);
}
Console.WriteLine($"Learning Extraction.Text sequence program for {SubBenchmarkName}");
while (!Success && TryGetNextExample(out SequenceExample nextConstraint))
{
var session = new SequenceSession();
UsedExamples.Add(nextConstraint);
session.Constraints.Add(UsedExamples);
Program = session.Learn();
Success = Program != null && AllExamples.All(e => Valid(e, Program));
}
RecordResult();
}
/// <summary>
/// Learns a program to extract a sequence of regions using its preceding sibling as reference.
/// </summary>
private static void LearnSequenceReferencingSibling()
{
var session = new SequenceSession();
var input =
SequenceSession.CreateStringRegion(
"United States\n Carrie Dodson 100\n Leonard Robledo 75\n Margaret Cook 320\n" +
"Canada\n Concetta Beck 350\n Nicholas Sayers 90\n Francis Terrill 2430\n" +
"New Zealand\n Nettie Pope 50\n Mack Beeson 1070");
// areas = { "United States", "Canada", "New Zealand" }
StringRegion[] areas = { input.Slice(0, 13), input.Slice(72, 78), input.Slice(140, 151) };
// Suppose we want to extract all last names from the input string.
session.AddConstraints(
new SequenceExample(areas[0], new[]
{
input.Slice(15, 21), // "United States" => "Carrie"
input.Slice(34, 41), // "United States" => "Leonard"
})
);
SequenceProgram topRankedProg = session.Learn();
if (topRankedProg == null)
{
Console.Error.WriteLine("Error: Learning fails!");
return;
}
// Note: we can't use SequenceProgram.Run(StringRegion) because of sibling referencing.
// Read the documentation for more information.
IEnumerable <IEnumerable <StringRegion> > outputSeq = topRankedProg.Run(areas);
foreach (Record <IEnumerable <StringRegion>, StringRegion> tup in outputSeq.ZipWith(areas))
{
foreach (StringRegion output in tup.Item1)
{
Console.WriteLine("\"{0}\" => \"{1}\"", tup.Item2, output == null ? "null" : output.Value);
}
}
}
public string LearnSequence(List <TextExtractExample> textExtractExamples)
{
var session = new SequenceSession();
var sequenceExamples = new List <SequenceExample>();
foreach (var textExtractExample in textExtractExamples)
{
var inputRegion = SequenceSession.CreateStringRegion(textExtractExample.text);
var exampleRegions = new List <StringRegion>();
foreach (var textExtractSelection in textExtractExample.selections)
{
var exampleRegion = inputRegion.Slice((uint)textExtractSelection.startPos, (uint)textExtractSelection.endPos);
exampleRegions.Add(exampleRegion);
}
sequenceExamples.Add(new SequenceExample(inputRegion, exampleRegions));
}
session.AddConstraints(sequenceExamples);
var program = session.Learn();
return(program.Serialize());
}
private static void Main(string[] args)
{
/*LearnRegion();
*
* LearnRegionUsingMultipleFiles();
*
* LearnRegionWithNegativeExamples();
*
* LearnRegionWithAdditionalReferences();
*
* LearnRegionReferencingParent();
*
* LearnRegionReferencingPrecedingSibling();
*
* LearnRegionReferencingSucceedingSibling();
*
* LearnTop3RegionPrograms();
*
* LearnAllRegionPrograms();
*
* LearnRegionWithRegexes();
*
* SerializeProgram();*/
// Learning sequence is similar to learning region.
// We only illustrate some API usages. Other sequence learning APIs are similar to their region APIs counterpart.
// Note: we need to give positive examples continuously.
// For instance, suppose we learn a list of {A, B, C, D, E}.
// {A, B} is a valid set of examples, while {A, C} is not.
// In case of { A, C}, Extraction.Text assumes that B is a negative example.
// This helps our learning converge more quickly.
//LearnSequence();
//LearnSequenceReferencingSibling();
var session = new SequenceSession();
// It is advised to learn a sequence with at least 2 examples because generalizing a sequence from a single element is hard.
// Also, we need to give positive examples continuously (i.e., we cannot skip any example).
var input =
SequenceSession.CreateStringRegion(
"We have some programs to showcase now. Example: Hello. Now do you see? Example: Yes. Now we move on to another definition. Example: what is up. This is getting so? Example: what. tiring. What is the time? Example: It detects!.");
// Suppose we want to extract all last names from the input string.
session.AddConstraints(
new SequenceExample(input, new[]
{
input.Slice(48, 53), // input => "Carrie"
input.Slice(80, 83), // input => "Leonard"
input.Slice(132, 142),
input.Slice(173, 177),
})
);
SequenceProgram topRankedProg = session.Learn();
if (topRankedProg == null)
{
Console.Error.WriteLine("Error: Learning fails!");
return;
}
foreach (StringRegion r in topRankedProg.Run(input))
{
string output = r != null ? r.Value : "null";
Console.WriteLine(output);
}
Console.WriteLine("\n\nDone.");
}
