/// <summary>
/// Learns a program to extract a region and provides other references to help find the intended program.
/// Demonstrates the use of additional references.
/// </summary>
private static void LearnRegionWithAdditionalReferences()
{
var session = new RegionSession();
StringRegion input =
RegionSession.CreateStringRegion("Carrie Dodson 100\nLeonard Robledo 75\nMargaret Cook ***");
StringRegion[] records = { input.Slice(0, 17), input.Slice(18, 36), input.Slice(37, 54) };
// Suppose we want to extract "100", "75", and "***".
session.AddConstraints(new RegionExample(records[0], records[0].Slice(14, 17)));
// "Carrie Dodson 100" => "100"
// Additional references help Extraction.Text observe the behavior of the learnt programs on unseen data.
// In this example, if we do not use additional references, Extraction.Text may learn a program that extracts the first number.
// On the contrary, if other references are present, it knows that this program is not applicable on the third record "Margaret Cook ***",
// and promotes a more applicable program.
session.AddInputs(records.Skip(1));
RegionProgram topRankedProg = session.Learn();
if (topRankedProg == null)
{
Console.Error.WriteLine("Error: Learning fails!");
return;
}
foreach (StringRegion record in records)
{
string output = topRankedProg.Run(record)?.Value ?? "null";
Console.WriteLine("\"{0}\" => \"{1}\"", record, output);
}
}
public static async Task <StructureExtractor> TrainExtractorAsync(IEnumerable <Tuple <string, uint, uint> > examples, IEnumerable <string> noneLabeledExamples = null)
{
if (null == examples || !examples.Any())
{
throw new AggregateException($"{nameof(examples)} must not be null or empty");
}
var regionSession = new RegionSession();
foreach (var example in examples)
{
var stringRegion = new StringRegion(example.Item1, Semantics.Tokens);
var field = stringRegion.Slice(example.Item2, example.Item3);
regionSession.AddConstraints(new RegionExample(stringRegion, field));
}
if (noneLabeledExamples?.Any() == true)
{
regionSession.AddInputs(noneLabeledExamples.Select(e => new StringRegion(e, Semantics.Tokens)));
}
var program = await regionSession.LearnAsync();
if (null == program)
{
throw new Exception("No program found");
}
return(new StructureExtractor(program));
}