using System;
using System.Collections.Generic;
using System.IO;
using Microsoft.ML;
 
namespace Samples.Dynamic
{
    public static class ApplyCustomWordEmbedding
    {
        public static void Example()
        {
            // Create a new ML context, for ML.NET operations. It can be used for
            // exception tracking and logging, as well as the source of randomness.
            var mlContext = new MLContext();
 
            // Create an empty list as the dataset. The 'ApplyWordEmbedding' does
            // not require training data as the estimator ('WordEmbeddingEstimator')
            // created by 'ApplyWordEmbedding' API is not a trainable estimator.
            // The empty list is only needed to pass input schema to the pipeline.
            var emptySamples = new List<TextData>();
 
            // Convert sample list to an empty IDataView.
            var emptyDataView = mlContext.Data.LoadFromEnumerable(emptySamples);
 
            // Write a custom 3-dimensional word embedding model with 4 words.
            // Each line follows '<word> <float> <float> <float>' pattern.
            // Lines that do not confirm to the pattern are ignored.
            var pathToCustomModel = @".\custommodel.txt";
            using (StreamWriter file = new StreamWriter(pathToCustomModel, false))
            {
                file.WriteLine("great 1.0 2.0 3.0");
                file.WriteLine("product -1.0 -2.0 -3.0");
                file.WriteLine("like -1 100.0 -100");
                file.WriteLine("buy 0 0 20");
            }
 
            // A pipeline for converting text into a 9-dimension word embedding
            // vector using the custom word embedding model. The 
            // 'ApplyWordEmbedding' computes the minimum, average and maximum values
            // for each token's embedding vector. Tokens in 'custommodel.txt' model
            // are represented as 3-dimension vector. Therefore, the output is of
            // 9 -dimension [min, avg, max].
            //
            // The 'ApplyWordEmbedding' API requires vector of text as input.
            // The pipeline first normalizes and tokenizes text then applies word
            // embedding transformation.
            var textPipeline = mlContext.Transforms.Text.NormalizeText("Text")
                .Append(mlContext.Transforms.Text.TokenizeIntoWords("Tokens",
                    "Text"))
                .Append(mlContext.Transforms.Text.ApplyWordEmbedding("Features",
                    pathToCustomModel, "Tokens"));
 
            // Fit to data.
            var textTransformer = textPipeline.Fit(emptyDataView);
 
            // Create the prediction engine to get the embedding vector from the
            // input text/string.
            var predictionEngine = mlContext.Model.CreatePredictionEngine<TextData,
                TransformedTextData>(textTransformer);
 
            // Call the prediction API to convert the text into embedding vector.
            var data = new TextData()
            {
                Text = "This is a great product. I would " +
                "like to buy it again."
            };
            var prediction = predictionEngine.Predict(data);
 
            // Print the length of the embedding vector.
            Console.WriteLine($"Number of Features: {prediction.Features.Length}");
 
            // Print the embedding vector.
            Console.Write("Features: ");
            foreach (var f in prediction.Features)
                Console.Write($"{f:F4} ");
 
            //  Expected output:
            //   Number of Features: 9
            //   Features: -1.0000 0.0000 -100.0000 0.0000 34.0000 -25.6667 1.0000 100.0000 20.0000
        }
 
        private class TextData
        {
            public string Text { get; set; }
        }
 
        private class TransformedTextData : TextData
        {
            public float[] Features { get; set; }
        }
    }
}