/// <summary>
/// Learns a program to extract a single region using two examples in two different files.
/// Learning from different files is similar to learning with multiple examples from a single file.
/// Demonstrates how to learn with examples from different files.
/// </summary>
private static void LearnRegionUsingMultipleFiles()
{
var input1 = RegionLearner.CreateStringRegion("Carrie Dodson 100");
var input2 = RegionLearner.CreateStringRegion("Leonard Robledo 75");
var examples = new[] {
new CorrespondingMemberEquals <StringRegion, StringRegion>(input1, input1.Slice(7, 13)), // "Carrie Dodson 100" => "Dodson"
new CorrespondingMemberEquals <StringRegion, StringRegion>(input2, input2.Slice(8, 15)) // "Leonard Robledo 75" => "Robledo"
};
RegionProgram topRankedProg = RegionLearner.Instance.Learn(examples);
if (topRankedProg == null)
{
Console.Error.WriteLine("Error: Learning fails!");
return;
}
var testInput = RegionLearner.CreateStringRegion("Margaret Cook 320"); // expect "Cook"
StringRegion output = topRankedProg.Run(testInput);
if (output == null)
{
Console.Error.WriteLine("Error: Extracting fails!");
return;
}
Console.WriteLine("\"{0}\" => \"{1}\"", testInput, output);
}
/// <summary>
/// Learns to serialize and deserialize Extraction.Text program.
/// </summary>
private static void SerializeProgram()
{
var input = RegionLearner.CreateStringRegion("Carrie Dodson 100");
var examples = new[] {
new CorrespondingMemberEquals <StringRegion, StringRegion>(input, input.Slice(7, 13)) // "Carrie Dodson 100" => "Dodson"
};
RegionProgram topRankedProg = RegionLearner.Instance.Learn(examples);
if (topRankedProg == null)
{
Console.Error.WriteLine("Error: Learning fails!");
return;
}
string serializedProgram = topRankedProg.Serialize();
RegionProgram deserializedProgram = Loader.Instance.Region.Load(serializedProgram);
var testInput = RegionLearner.CreateStringRegion("Leonard Robledo 75"); // expect "Robledo"
StringRegion output = deserializedProgram.Run(testInput);
if (output == null)
{
Console.Error.WriteLine("Error: Extracting fails!");
return;
}
Console.WriteLine("\"{0}\" => \"{1}\"", testInput, output);
}
/// <summary>
/// Learns a program to extract a single region from a file.
/// </summary>
private static void LearnRegion()
{
var input = RegionLearner.CreateStringRegion("Carrie Dodson 100");
// Only one example because we extract one region from one file.
// Position specifies the location between two characters in the file. It starts at 0 (the beginning of the file).
// An example is identified by a pair of start and end positions.
var examples = new[] {
new CorrespondingMemberEquals <StringRegion, StringRegion>(input, input.Slice(7, 13)) // "Carrie Dodson 100" => "Dodson"
};
RegionProgram topRankedProg = RegionLearner.Instance.Learn(examples);
if (topRankedProg == null)
{
Console.Error.WriteLine("Error: Learning fails!");
return;
}
var testInput = RegionLearner.CreateStringRegion("Leonard Robledo 75"); // expect "Robledo"
StringRegion output = topRankedProg.Run(testInput);
if (output == null)
{
Console.Error.WriteLine("Error: Extracting fails!");
return;
}
Console.WriteLine("\"{0}\" => \"{1}\"", testInput, output);
}
/// <summary>
/// Learns top-ranked 3 region programs.
/// Demonstrates access to lower-ranked programs.
/// </summary>
private static void LearnTop3RegionPrograms()
{
var input = RegionLearner.CreateStringRegion("Carrie Dodson 100");
var examples = new[] {
new CorrespondingMemberEquals <StringRegion, StringRegion>(input, input.Slice(14, 17)) // "Carrie Dodson 100" => "Dodson"
};
IEnumerable <RegionProgram> topKPrograms = RegionLearner.Instance.LearnTopK(examples, 3);
var i = 0;
StringRegion[] otherInputs =
{
input, RegionLearner.CreateStringRegion("Leonard Robledo NA"),
RegionLearner.CreateStringRegion("Margaret Cook 320")
};
foreach (var prog in topKPrograms)
{
Console.WriteLine("Program {0}:", ++i);
foreach (var str in otherInputs)
{
var r = prog.Run(str);
Console.WriteLine(r != null ? r.Value : "null");
}
}
}
/// <summary>
/// Learns all region programs that satisfy the examples (advanced feature).
/// Demonstrates access to the entire program set.
/// </summary>
private static void LearnAllRegionPrograms()
{
var input = RegionLearner.CreateStringRegion("Carrie Dodson 100");
var examples = new[] {
new CorrespondingMemberEquals <StringRegion, StringRegion>(input, input.Slice(14, 17)) // "Carrie Dodson 100" => "Dodson"
};
ProgramSet allPrograms = RegionLearner.Instance.LearnAll(examples);
IEnumerable <ProgramNode> topKPrograms =
allPrograms.TopK(RegionLearner.Instance.ScoreFeature, 3); // "Score" is the ranking feature
var i = 0;
StringRegion[] otherInputs =
{
input, RegionLearner.CreateStringRegion("Leonard Robledo NA"),
RegionLearner.CreateStringRegion("Margaret Cook 320")
};
foreach (var prog in topKPrograms)
{
Console.WriteLine("Program {0}:", ++i);
foreach (var str in otherInputs)
{
State inputState = State.Create(Language.Grammar.InputSymbol, str); // Create Microsoft.ProgramSynthesis input state
object r = prog.Invoke(inputState); // Invoke Microsoft.ProgramSynthesis program node on the input state
Console.WriteLine(r != null ? (r as StringRegion).Value : "null");
}
}
}
/// <summary>
/// Learns a program to extract a single region using another region that appears after it as reference (i.e.,
/// succeeding sibling region).
/// Demonstrates how sibling referencing works.
/// </summary>
private static void LearnRegionReferencingSucceedingSibling()
{
var input = RegionLearner.CreateStringRegion("Carrie Dodson 100\nLeonard Robledo 75\nMargaret Cook 320");
StringRegion[] records = { input.Slice(0, 17), input.Slice(18, 36), input.Slice(37, 54) };
StringRegion[] numbers = { input.Slice(14, 17), input.Slice(34, 36), input.Slice(51, 54) };
// Suppose we want to extract the first name w.r.t the number
var examples = new[] {
new CorrespondingMemberEquals <StringRegion, StringRegion>(numbers[0], records[0].Slice(0, 6)), // "Carrie" => "100"
new CorrespondingMemberEquals <StringRegion, StringRegion>(numbers[1], records[1].Slice(18, 25)) // "Leonard" => "75"
};
RegionProgram topRankedProg = RegionLearner.Instance.Learn(examples);
if (topRankedProg == null)
{
Console.Error.WriteLine("Error: Learning fails!");
return;
}
foreach (StringRegion number in numbers)
{
string output = topRankedProg.Run(number)?.Value ?? "null";
Console.WriteLine("\"{0}\" => \"{1}\"", number, output);
}
}
/// <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 input = RegionLearner.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 "***".
var examples = new[] {
new CorrespondingMemberEquals <StringRegion, StringRegion>(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.
RegionProgram topRankedProg = RegionLearner.Instance.Learn(examples, new[] { records.Skip(1) });
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);
}
}
/// <summary>
/// Learns a program to extract a region with both positive and negative examples.
/// Demonstrates the use of negative examples.
/// </summary>
private static void LearnRegionWithNegativeExamples()
{
var input = RegionLearner.CreateStringRegion("Carrie Dodson 100\nLeonard Robledo NA\nMargaret Cook 320");
StringRegion[] records = { input.Slice(0, 17), input.Slice(18, 36), input.Slice(37, 54) };
// Suppose we want to extract "100", "320".
var constraints = new Constraint <IEnumerable <StringRegion>, IEnumerable <StringRegion> >[] {
new CorrespondingMemberEquals <StringRegion, StringRegion>(records[0], records[0].Slice(14, 17)), // "Carrie Dodson 100" => "100"
new CorrespondingMemberDoesNotIntersect <StringRegion>(records[1], records[1]) // no extraction in "Leonard Robledo NA"
};
// Extraction.Text will find a program whose output does not OVERLAP with any of the negative examples.
RegionProgram topRankedProg = RegionLearner.Instance.Learn(constraints);
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);
}
}
private static void TestTextTransformationBenchmark(Grammar grammar, string benchmark, int exampleCount = 2)
{
string[] lines = File.ReadAllLines($"benchmarks/{benchmark}.tsv");
Tuple <string, string>[] data = lines.Select(l =>
{
var parts = l.Split(new[] { "\t" }, StringSplitOptions.RemoveEmptyEntries);
return(Tuple.Create(parts[0], parts[1]));
}).ToArray();
var examples =
data.Take(exampleCount)
.ToDictionary(
t => State.Create(grammar.InputSymbol, RegionLearner.CreateStringRegion(t.Item1)),
t => (object)RegionLearner.CreateStringRegion(t.Item2));
var spec = new ExampleSpec(examples);
ProgramNode program = Learn(grammar, spec,
new Substrings.RankingScore(grammar),
new Substrings.WitnessFunctions(grammar));
foreach (Tuple <string, string> row in data.Skip(exampleCount))
{
State input = State.Create(grammar.InputSymbol,
RegionLearner.CreateStringRegion(row.Item1));
var output = program.Invoke(input);
WriteColored(ConsoleColor.DarkCyan, $"{row.Item1} => {output}");
}
}
public static List <StringRegion> LoadBenchmark(string filename, out StringRegion document)
{
string content = File.ReadAllText(filename);
Match[] examples = ExampleRegex.Matches(content).Cast <Match>().ToArray();
document = RegionLearner.CreateStringRegion(content.Replace("}", "").Replace("{", ""));
var result = new List <StringRegion>();
for (int i = 0, shift = -1; i < examples.Length; i++, shift -= 2)
{
int start = shift + examples[i].Index;
int end = start + examples[i].Length;
result.Add(document.Slice((uint)start, (uint)end));
}
return(result);
}
private static void LoadAndTestSubstrings()
{
var grammar = LoadGrammar("ProseSample.Substrings.grammar");
if (grammar == null)
{
return;
}
ProgramNode p = ProgramNode.Parse(@"SubStr(v, PosPair(AbsPos(v, -4), AbsPos(v, -1)))",
grammar, ASTSerializationFormat.HumanReadable);
StringRegion data = RegionLearner.CreateStringRegion("Microsoft PROSE SDK");
State input = State.Create(grammar.InputSymbol, data);
Console.WriteLine(p.Invoke(input));
StringRegion sdk = data.Slice(data.End - 3, data.End);
Spec spec = ShouldConvert.Given(grammar).To(data, sdk);
Learn(grammar, spec,
new Substrings.RankingScore(grammar), new Substrings.WitnessFunctions(grammar));
TestTextTransformationBenchmark(grammar, "emails");
}