/// <summary>
/// Constructor which takes a <see cref="PerceptronModel"/> and a <see cref="Stream"/> and prepares
/// itself to write the model to that writer.
/// </summary>
/// <param name="model">The <seealso cref="PerceptronModel"/> which is to be persisted.</param>
/// <param name="outputStream">The output stream.</param>
/// <exception cref="System.ArgumentNullException">outputStream</exception>
/// <exception cref="System.ArgumentException">Stream was not writable.</exception>
public PlainTextPerceptronModelWriter(AbstractModel model, Stream outputStream) : base(model) {
if (outputStream == null)
throw new ArgumentNullException("outputStream");
if (!outputStream.CanWrite)
throw new ArgumentException(@"Stream was not writable.", "outputStream");
writer = new StreamWriter(outputStream);
}
/// <para>
/// <c>index 0</c>: <see cref="T:Context[]"/> containing the model parameters.
/// <c>index 1</c>: <see cref="T:IndexHashTable{string}"/> containing the mapping of model predicates to unique integers.
/// <c>index 2</c>: <see cref="T:string[]"/> containing the names of the outcomes, stored in the index of the array which represents their unique ids in the model.
/// <c>index 3</c>: <see cref="T:double"/> containing the value of the models correction constant.
/// <c>index 4</c>: <see cref="T:double"/> containing the value of the models correction parameter.
/// </para>
///
protected GISModelWriter(AbstractModel model) {
var data = model.GetDataStructures();
PARAMS = (Context[]) data[0];
var map = (IndexHashTable<string>) data[1];
OUTCOME_LABELS = (string[]) data[2];
CORRECTION_CONSTANT = (int) data[3]; // string[] ??????????
CORRECTION_PARAM = (Double) data[4];
PRED_LABELS = new String[map.Size];
map.ToArray(PRED_LABELS);
}
public static void WriteModel(AbstractModel model, Stream outStream) {
if (model == null)
throw new ArgumentNullException("model");
if (outStream == null)
throw new ArgumentNullException("outStream");
if (!outStream.CanWrite)
throw new ArgumentException("Stream was not writeable.");
var writer = new GenericModelWriter(model, outStream);
writer.Persist();
}
/// <summary>
/// Initializes a new instance of the <see cref="GenericModelWriter" /> class.
/// </summary>
/// <param name="model">The model.</param>
/// <param name="outputStream">The output stream.</param>
/// <exception cref="System.InvalidOperationException">Invalid model type.</exception>
public GenericModelWriter(AbstractModel model, Stream outputStream) {
switch (model.ModelType) {
case ModelType.Maxent:
writer = new BinaryGISModelWriter(model, outputStream);
break;
case ModelType.MaxentQn:
writer = new BinaryQNModelWriter(model, outputStream);
break;
case ModelType.Perceptron:
writer = new BinaryPerceptronModelWriter(model, outputStream);
break;
default:
throw new InvalidOperationException("Invalid model type");
}
}
/// <summary>
/// Initializes a new instance of the <see cref="GenericModelWriter" /> class.
/// </summary>
/// <param name="model">The model.</param>
/// <param name="outputStream">The output stream.</param>
/// <exception cref="System.InvalidOperationException">Invalid model type.</exception>
public GenericModelWriter(AbstractModel model, Stream outputStream)
{
switch (model.ModelType)
{
case ModelType.Maxent:
writer = new BinaryGISModelWriter(model, outputStream);
break;
case ModelType.MaxentQn:
writer = new BinaryQNModelWriter(model, outputStream);
break;
case ModelType.Perceptron:
writer = new BinaryPerceptronModelWriter(model, outputStream);
break;
default:
throw new InvalidOperationException("Invalid model type");
}
}
/// <summary>
/// Constructor which takes a <see cref="PerceptronModel"/> and a File and prepares itself to
/// write the model to that file. Detects whether the file is GZipped or not based on whether
/// the suffix contains ".gz".
/// </summary>
/// <param name="model">The PerceptronModel which is to be persisted.</param>
/// <param name="fileName">The filename in which the model is to be persisted.</param>
public PlainTextPerceptronModelWriter(AbstractModel model, string fileName) : base(model) {
writer = fileName.EndsWith(".gz")
? new StreamWriter(new GZipStream(new FileStream(fileName, FileMode.Create), CompressionMode.Compress))
: new StreamWriter(new FileStream(fileName, FileMode.Create));
}
protected void ValidatePOSDictionary(POSDictionary posDic, AbstractModel posModel) {
if (!posModel.ContainsOutcomes(posDic)) {
throw new InvalidFormatException("Tag dictionary contains tags which are unknown by the model!");
}
}
/// <summary>
/// Creates a new event array based on the outcomes predicted by the specified parameters for the specified sequence.
/// </summary>
/// <param name="sequence">The sequence to be evaluated.</param>
/// <param name="model">The model.</param>
/// <returns>The event array.</returns>
public Event[] UpdateContext(Sequence sequence, AbstractModel model) {
var tagger =
new NameFinderME(
new TokenNameFinderModel("x-unspecified", model, new Dictionary<string, object>(), null));
var sentence = sequence.GetSource<NameSample>().Sentence;
var tags = seqCodec.Encode(tagger.Find(sentence), sentence.Length);
return NameFinderEventStream.GenerateEvents(sentence, tags, pcg).ToArray();
}
/// <summary>
/// Creates a new event array based on the outcomes predicted by the specified parameters for the specified sequence.
/// </summary>
/// <param name="sequence">The sequence to be evaluated.</param>
/// <param name="model">The model.</param>
/// <returns>The event array.</returns>
/// <remarks>Always return null.</remarks>
public Event[] UpdateContext(Sequence sequence, AbstractModel model) {
// TODO: Should be implemented for Perceptron sequence learning ...
return null;
}
/// <summary>
/// Initializes a new instance of the <see cref="BinaryGISModelWriter"/> class.
/// </summary>
/// <param name="model">The GIS model.</param>
/// <param name="outputStream">The output stream.</param>
public BinaryGISModelWriter(AbstractModel model, Stream outputStream) : base(model) {
writer = new BinaryFileDataWriter(outputStream);
}
/// <summary>
/// Creates a new event array based on the outcomes predicted by the specified parameters for the specified sequence.
/// </summary>
/// <param name="sequence">The sequence to be evaluated.</param>
/// <param name="model">The model.</param>
/// <returns>The event array.</returns>
public Event[] UpdateContext(Sequence sequence, AbstractModel model) {
var tagger = new POSTaggerME(new POSModel("x-unspecified", model, null, new POSTaggerFactory()));
var sample = sequence.GetSource<POSSample>();
var tags = tagger.Tag(sample.Sentence);
return POSSampleEventStream.GenerateEvents(
sample.Sentence,
tags,
Array.ConvertAll(sample.AdditionalContext, input => (object) input),
contextGenerator).ToArray();
}
public BinaryPerceptronModelWriter(AbstractModel model, Stream outStream) : base(model) {
writer = new BinaryFileDataWriter(outStream);
}
