private static void Kaggle_MNIST_Test(SimpleNeuralNet net)
        {
            Console.WriteLine("Testing...");
            string fname = "../kaggle_mnist_test.csv";
            //string fname = "../my_test_data.csv";
            //string fname = "../mnist_train.csv";

            var lines      = File.ReadAllLines(fname);
            var train_data = new Vector <double> [lines.Length - 1];
            var j          = -1;

            foreach (var line in lines)
            {
                if (line.Contains("pixel"))
                {
                    continue;
                }
                if (string.IsNullOrEmpty(line))
                {
                    continue;
                }

                var      words = line.Split(',');
                string   label = words[0];
                double[] nums  = new double[words.Length];
                for (int i = 0; i < words.Length; i++)
                {
                    nums[i] = Convert.ToDouble(words[i]);
                    nums[i] = (nums[i] / 255.0) * 0.99 + 0.01; // normalize the nums to be between 0.01 - 1 (inclusive)
                }
                train_data[++j] = DenseVector.OfArray(nums);
            }
            lines = null;

            int    right        = 0;
            string fileContents = "ImageId,Label\n";

            for (int i = 0; i < train_data.Length; i++)
            {
                var answer = net.Query(train_data[i]);
                int max    = 0;
                for (int k = 1; k < answer.Count; k++)
                {
                    if (answer[k] > answer[max])
                    {
                        max = k;
                    }
                }
                fileContents += (i + 1) + "," + max + "\n";
            }
            File.WriteAllText("../kaggle_mnist_answer.csv", fileContents);
        }
        private static void MNIST_Test(SimpleNeuralNet net)
        {
            Console.WriteLine("Testing...");
            string fname = "../mnist_test.csv";
            //string fname = "../my_test_data.csv";
            //string fname = "../mnist_train.csv";

            var lines       = File.ReadAllLines(fname);
            var train_data  = new Vector <double> [lines.Length];
            var target_data = new Vector <double> [lines.Length];
            var j           = -1;

            foreach (var line in lines)
            {
                var      words = line.Split(',');
                string   label = words[0];
                double[] nums  = new double[words.Length - 1];
                for (int i = 1; i < words.Length; i++)
                {
                    nums[i - 1] = Convert.ToDouble(words[i]);
                    nums[i - 1] = (nums[i - 1] / 255.0) * 0.99 + 0.01; // normalize the nums to be between 0.01 - 1 (inclusive)
                }
                train_data[++j] = DenseVector.OfArray(nums);
                var target = new double[10];
                for (int i = 0; i < 10; i++)
                {
                    target[i] = 0.01;
                }
                target[Convert.ToInt32(label)] = 0.99;
                target_data[j] = DenseVector.OfArray(target);
            }
            lines = null;

            int right = 0;

            for (int i = 0; i < train_data.Length; i++)
            {
                Console.WriteLine("Test #" + (i + 1));
                int right_answer = 0;
                for (int k = 0; k < target_data[i].Count; k++)
                {
                    if (target_data[i][k] > 0.98)
                    {
                        right_answer = k;
                        break;
                    }
                }

                var answer = net.Query(train_data[i]);
                int max    = 0;
                for (int k = 1; k < answer.Count; k++)
                {
                    if (answer[k] > answer[max])
                    {
                        max = k;
                    }
                }

                right += (max == right_answer ? 1 : 0);
                Console.WriteLine(right_answer + "  NI: " + max + " " + (max == right_answer ? "R" : "W"));
            }

            Console.WriteLine("Accuracy: " + ((double)right / (double)train_data.Length));
        }