An OpenCL kernel that is designed to calculate gradients and help train a neural network.
Inheritance: EncogKernel
コード例 #1
0
        /// <summary>
        /// Learn using the Manhattan update rule.
        /// </summary>
        ///
        /// <param name="learningRate">The learning rate.</param>
        public void LearnManhattan(double learningRate)
        {
            this.learningType = TrainFlatNetworkOpenCL.LEARN_MANHATTAN;
            this.learningRate = learningRate;

            IDictionary<String, String> options = GetOptions("LEARN_MANHATTAN");

            this.kernel = new KernelNetworkTrain(this.profile.Device,
                    this.network, this.training, 1);
            this.kernel.Compile(options, profile, this.network);

            this.kernel.TempDataArray[0] = (float)learningRate;
        }
コード例 #2
0
        /// <summary>
        /// Learn using RPROP with a custom initial update and max step.
        /// </summary>
        ///
        /// <param name="initialUpdate">The initial update value.</param>
        /// <param name="maxStep">The max step.</param>
        public void LearnRPROP(double initialUpdate, double maxStep)
        {
            this.learningType = TrainFlatNetworkOpenCL.LEARN_RPROP;
            this.initialUpdate = initialUpdate;
            this.maxStep = maxStep;

            IDictionary<String, String> options = GetOptions("LEARN_RPROP");

            this.kernel = new KernelNetworkTrain(this.profile.Device,
                    this.network, this.training,
                    this.network.Weights.Length * 2);

            this.kernel.Compile(options, profile, this.network);

            int weightLength = this.network.Weights.Length;

            for (int i = 0; i < weightLength; i++)
            {
                this.kernel.TempDataArray[i] = 0;
                this.kernel.TempDataArray[i + weightLength] = (float)this.initialUpdate;
            }

        }
コード例 #3
0
        /// <summary>
        /// Learn using backpropagation.
        /// </summary>
        ///
        /// <param name="learningRate">The learning rate.</param>
        /// <param name="momentum">The momentum.</param>
        public void LearnBPROP(double learningRate, double momentum)
        {
            this.learningType = TrainFlatNetworkOpenCL.LEARN_BPROP;
            this.momentum = momentum;
            this.learningRate = learningRate;

            this.learningType = TrainFlatNetworkOpenCL.LEARN_BPROP;

            IDictionary<String, String> options = GetOptions("LEARN_BPROP");

            this.kernel = new KernelNetworkTrain(this.profile.Device,
                    this.network, this.training,
                    this.network.Weights.Length + 2);
            this.kernel.Compile(options, profile, this.network);

            this.kernel.TempDataArray[0] = (float)learningRate;
            this.kernel.TempDataArray[1] = (float)momentum;
        }