// Licensed to the .NET Foundation under one or more agreements.
// The .NET Foundation licenses this file to you under the MIT license.
// See the LICENSE file in the project root for more information.
 
using Microsoft.ML.Data;
using Microsoft.ML.Internal.Utilities;
 
namespace Microsoft.ML.Transforms.TimeSeries
{
    /// <summary>
    /// The base container class for the forecast result on a sequence of type <typeparamref name="T"/>.
    /// </summary>
    /// <typeparam name="T">The type of the elements in the sequence</typeparam>
    internal abstract class ForecastResultBase<T>
    {
        public VBuffer<T> PointForecast;
    }
 
    /// <summary>
    /// The standard interface for modeling a sequence.
    /// </summary>
    /// <typeparam name="TInput">The type of the elements in the input sequence</typeparam>
    /// <typeparam name="TOutput">The type of the elements in the output sequence</typeparam>
    internal abstract class SequenceModelerBase<TInput, TOutput> : ICanSaveModel
    {
        private protected SequenceModelerBase()
        {
        }
 
        /// <summary>
        /// Initializes the state of the modeler
        /// </summary>
        internal abstract void InitState();
 
        /// <summary>
        /// Consumes one element from the input sequence.
        /// </summary>
        /// <param name="input">An element in the sequence</param>
        /// <param name="updateModel">determines whether the sequence model should be updated according to the input</param>
        internal abstract void Consume(ref TInput input, bool updateModel = false);
 
        /// <summary>
        /// Trains the sequence model on a given sequence.
        /// </summary>
        /// <param name="data">The input sequence used for training</param>
        internal abstract void Train(FixedSizeQueue<TInput> data);
 
        /// <summary>
        /// Trains the sequence model on a given sequence. The method accepts an object of RoleMappedData,
        /// and assumes the input column is the 'Feature' column of type TInput.
        /// </summary>
        /// <param name="data">The input sequence used for training</param>
        internal abstract void Train(RoleMappedData data);
 
        /// <summary>
        /// Forecasts the next 'horizon' elements in the output sequence.
        /// </summary>
        /// <param name="result">The forecast result for the given horizon along with optional information depending on the algorithm</param>
        /// <param name="horizon">The forecast horizon</param>
        internal abstract void Forecast(ref ForecastResultBase<TOutput> result, int horizon = 1);
 
        /// <summary>
        /// Predicts the next element in the output sequence.
        /// </summary>
        /// <param name="output">The output ref parameter the will contain the prediction result</param>
        internal abstract void PredictNext(ref TOutput output);
 
        /// <summary>
        /// Creates a clone of the model.
        /// </summary>
        /// <returns>A clone of the object</returns>
        internal abstract SequenceModelerBase<TInput, TOutput> Clone();
 
        /// <summary>
        /// Implementation of <see cref="ICanSaveModel.Save(ModelSaveContext)"/>.
        /// </summary>
        void ICanSaveModel.Save(ModelSaveContext ctx) => SaveModel(ctx);
 
        private protected abstract void SaveModel(ModelSaveContext ctx);
    }
}