public static void UsersTimelineTest(PredictionEngine <SentenceData, PredictionData> engine, string userName, string language, int count = 100)
{
Console.WriteLine($"Getting @{userName}'s recent tweets..");
var timeline = _token.Statuses.UserTimeline(screen_name: userName, count: count, exclude_replies: true);
int lan = 0;
List <SentenceData> testData = new List <SentenceData>();
foreach (var item in timeline)
{
var text = Regex.Replace(item.Text, @"http[^\s]+", "");
var tweet = new SentenceData()
{
Sentence = text
};
var prediction = engine.Predict(tweet);
if (prediction.PredictedLabel.Equals(language))
{
lan++;
}
tweet.Label = prediction.PredictedLabel;
testData.Add(tweet);
}
// Expected + Predicted + Sentence
File.AppendAllLines("tweet.txt", testData.Select(i => language + '\t' + i.Label + '\t' + i.Sentence));
Console.WriteLine($"Last {lan} tweets from {timeline.Count} are in '{language}' class ");
}
static void Train()
{
string dataPath = "train.txt", testDataPath = "test.txt";
var context = new MLContext();
// Create textloader for our structure
var textLoader = context.Data.CreateTextReader(new TextLoader.Arguments()
{
Separator = "\t",
Column = new[] {
new TextLoader.Column("Label", DataKind.Text, 0),
new TextLoader.Column("Sentence", DataKind.Text, 1)
}
});
var trainDataView = textLoader.Read(dataPath);
var testDataView = textLoader.Read(testDataPath);
// Create data process pipeline
// First we have to change label value into ML.NET KeyType
var dataProcessPipeline = context.Transforms.Conversion.MapValueToKey("Label")
// Then, we have to normalize text
.Append(context.Transforms.Text.NormalizeText("Sentence", "NormalizedSentence"))
// Featurize the given text with n-grams
.Append(context.Transforms.Text.FeaturizeText("NormalizedSentence", "Features"))
// Give the naive bayes algorithm
.Append(context.MulticlassClassification.Trainers.NaiveBayes())
// Convert back the label value to it's origin
.Append(context.Transforms.Conversion.MapKeyToValue("PredictedLabel"));
// Create our model with train data
var model = dataProcessPipeline.Fit(trainDataView);
// Transform our model with test data
var testPredictions = model.Transform(testDataView);
// Evaluate the model and Print the results
Evaluate(context, testPredictions);
// Create single prediction engine
var predictionEngine = model.CreatePredictionEngine <SentenceData, PredictionData>(context);
// Create the testing data
var testData = new SentenceData()
{
Sentence = "Enfunda tu espada, saca tu baraja y prepárate para disfrutar con Hearthstone, un trepidante juego de cartas de estrategia, fácil de aprender y salvajemente divertido. Inicia una partida gratuita y utiliza tus mejores cartas para lanzar hechizos, invocar criaturas y dar órdenes a los héroes de Warcraft en épicos y estratégicos duelos."
};
// Predict the testing data
var result = predictionEngine.Predict(testData);
Console.WriteLine("Predicted language {0}", result.PredictedLabel);
TweetTest.UsersTimelineTest(predictionEngine, "NetflixES", "es", 400);
TweetTest.UsersTimelineTest(predictionEngine, "netflix", "en", 400);
TweetTest.UsersTimelineTest(predictionEngine, "netflixturkiye", "tr", 400);
}
