NaiveBayesClassifier.cs

/*****************************************************************************
   Copyright 2018 The TensorFlow.NET Authors. All Rights Reserved.

   Licensed under the Apache License, Version 2.0 (the "License");
   you may not use this file except in compliance with the License.
   You may obtain a copy of the License at

       http://www.apache.org/licenses/LICENSE-2.0

   Unless required by applicable law or agreed to in writing, software
   distributed under the License is distributed on an "AS IS" BASIS,
   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
   See the License for the specific language governing permissions and
   limitations under the License.
******************************************************************************/

using System;
using System.Collections.Generic;
using System.IO;
using Tensorflow;
using Tensorflow.Keras.Utils;
using Tensorflow.NumPy;
using static Tensorflow.Binding;

namespace TensorFlowNET.Examples
{
    /// <summary>
    /// https://github.com/nicolov/naive_bayes_tensorflow
    /// </summary>
    public class NaiveBayesClassifier : SciSharpExample, IExample
    {
        public NDArray X, y;
        public Normal dist { get; set; }

        public ExampleConfig InitConfig()
            => Config = new ExampleConfig
            {
                Name = "Naive Bayes Classifier",
                Enabled = true,
                IsImportingGraph = false
            };

        public bool Run()
        {
            tf.enable_eager_execution();

            PrepareData();

            fit(X, y);

            // Create a regular grid and classify each point 
            float x_min = np.amin(X, 0)[0] - 0.5f;
            float y_min = np.amin(X, 0)[1] - 0.5f;
            float x_max = np.amax(X, 0)[1] + 0.5f;
            float y_max = np.amax(X, 0)[1] + 0.5f;

            var (xx, yy) = np.meshgrid(np.linspace(x_min, x_max, 30), np.linspace(y_min, y_max, 30));
            var array = np.Load<double[,]>(Path.Join("nb", "nb_example.npy"));
            var samples = np.array(array).astype(np.float32);
            var Z = predict(samples);

            return true;
        }

        public void fit(NDArray X, NDArray y)
        {
            var (unique_y, _) = np.unique(y);

            var dic = new Dictionary<int, List<List<float>>>();
            // Init uy in dic
            foreach (int uy in unique_y.ToArray<int>())
            {
                dic.Add(uy, new List<List<float>>());
            }
            // Separate training points by class 
            // Shape : nb_classes * nb_samples * nb_features
            int maxCount = 0;
            for (int i = 0; i < (int)y.size; i++)
            {
                var curClass = y[i];
                var l = dic[curClass];
                var pair = new List<float>();
                pair.Add(X[i, 0]);
                pair.Add(X[i, 1]);
                l.Add(pair);
                if (l.Count > maxCount)
                {
                    maxCount = l.Count;
                }
                dic[curClass] = l;
            }
            float[,,] points = new float[dic.Count, maxCount, X.shape[1]];
            foreach (KeyValuePair<int, List<List<float>>> kv in dic)
            {
                int j = kv.Key;
                for (int i = 0; i < maxCount; i++)
                {
                    for (int k = 0; k < X.shape[1]; k++)
                    {
                        points[j, i, k] = kv.Value[i][k];
                    }
                }

            }
            var points_by_class = np.array(points);
            // estimate mean and variance for each class / feature
            // shape : nb_classes * nb_features
            var cons = tf.constant(points_by_class);
            var (mean, variance) = tf.nn.moments(cons, new int[] { 1 });

            // Create a 3x2 univariate normal distribution with the 
            // Known mean and variance           
            dist = tf.distributions.Normal(mean, tf.sqrt(variance));
        }

        public Tensor predict(NDArray X)
        {
            if (dist == null)
            {
                throw new ArgumentNullException("cant not find the model (normal distribution)!");
            }
            int nb_classes = (int)dist.scale().shape[0];
            int nb_features = (int)dist.scale().shape[1];

            // Conditional probabilities log P(x|c) with shape
            // (nb_samples, nb_classes)
            var t1 = ops.convert_to_tensor(X, TF_DataType.TF_FLOAT);
            var t2 = ops.convert_to_tensor(new int[] { 1, nb_classes });
            Tensor tile = tf.tile(t1, t2);
            var t3 = ops.convert_to_tensor(new int[] { -1, nb_classes, nb_features });
            Tensor r = tf.reshape(tile, t3);
            var cond_probs = tf.reduce_sum(dist.log_prob(r), 2);
            // uniform priors
            float[] tem = new float[nb_classes];
            for (int i = 0; i < tem.Length; i++)
            {
                tem[i] = 1.0f / nb_classes;
            }
            var priors = np.log(np.array(tem));

            // posterior log probability, log P(c) + log P(x|c)
            var joint_likelihood = ops.convert_to_tensor(priors, TF_DataType.TF_FLOAT) + cond_probs;
            // normalize to get (log)-probabilities

            var norm_factor = tf.reduce_logsumexp(joint_likelihood, new int[] { 1 }, keepdims: true);
            var log_prob = joint_likelihood - norm_factor;
            // exp to get the actual probabilities
            return tf.exp(log_prob);
        }

        public override void PrepareData()
        {
            #region Training data
            X = np.array(new float[,] {
                {5.1f, 3.5f}, {4.9f, 3.0f}, {4.7f, 3.2f}, {4.6f, 3.1f}, {5.0f, 3.6f}, {5.4f, 3.9f},
                {4.6f, 3.4f}, {5.0f, 3.4f}, {4.4f, 2.9f}, {4.9f, 3.1f}, {5.4f, 3.7f}, {4.8f, 3.4f},
                {4.8f, 3.0f}, {4.3f, 3.0f}, {5.8f, 4.0f}, {5.7f, 4.4f}, {5.4f, 3.9f}, {5.1f, 3.5f},
                {5.7f, 3.8f}, {5.1f, 3.8f}, {5.4f, 3.4f}, {5.1f, 3.7f}, {5.1f, 3.3f}, {4.8f, 3.4f},
                {5.0f, 3.0f}, {5.0f, 3.4f}, {5.2f, 3.5f}, {5.2f, 3.4f}, {4.7f, 3.2f}, {4.8f, 3.1f},
                {5.4f, 3.4f}, {5.2f, 4.1f}, {5.5f, 4.2f}, {4.9f, 3.1f}, {5.0f, 3.2f}, {5.5f, 3.5f},
                {4.9f, 3.6f}, {4.4f, 3.0f}, {5.1f, 3.4f}, {5.0f, 3.5f}, {4.5f, 2.3f}, {4.4f, 3.2f},
                {5.0f, 3.5f}, {5.1f, 3.8f}, {4.8f, 3.0f}, {5.1f, 3.8f}, {4.6f, 3.2f}, {5.3f, 3.7f},
                {5.0f, 3.3f}, {7.0f, 3.2f}, {6.4f, 3.2f}, {6.9f, 3.1f}, {5.5f, 2.3f}, {6.5f, 2.8f},
                {5.7f, 2.8f}, {6.3f, 3.3f}, {4.9f, 2.4f}, {6.6f, 2.9f}, {5.2f, 2.7f}, {5.0f, 2.0f},
                {5.9f, 3.0f}, {6.0f, 2.2f}, {6.1f, 2.9f}, {5.6f, 2.9f}, {6.7f, 3.1f}, {5.6f, 3.0f},
                {5.8f, 2.7f}, {6.2f, 2.2f}, {5.6f, 2.5f}, {5.9f, 3.0f}, {6.1f, 2.8f}, {6.3f, 2.5f},
                {6.1f, 2.8f}, {6.4f, 2.9f}, {6.6f, 3.0f}, {6.8f, 2.8f}, {6.7f, 3.0f}, {6.0f, 2.9f},
                {5.7f, 2.6f}, {5.5f, 2.4f}, {5.5f, 2.4f}, {5.8f, 2.7f}, {6.0f, 2.7f}, {5.4f, 3.0f},
                {6.0f, 3.4f}, {6.7f, 3.1f}, {6.3f, 2.3f}, {5.6f, 3.0f}, {5.5f, 2.5f}, {5.5f, 2.6f},
                {6.1f, 3.0f}, {5.8f, 2.6f}, {5.0f, 2.3f}, {5.6f, 2.7f}, {5.7f, 3.0f}, {5.7f, 2.9f},
                {6.2f, 2.9f}, {5.1f, 2.5f}, {5.7f, 2.8f}, {6.3f, 3.3f}, {5.8f, 2.7f}, {7.1f, 3.0f},
                {6.3f, 2.9f}, {6.5f, 3.0f}, {7.6f, 3.0f}, {4.9f, 2.5f}, {7.3f, 2.9f}, {6.7f, 2.5f},
                {7.2f, 3.6f}, {6.5f, 3.2f}, {6.4f, 2.7f}, {6.8f, 3.0f}, {5.7f, 2.5f}, {5.8f, 2.8f},
                {6.4f, 3.2f}, {6.5f, 3.0f}, {7.7f, 3.8f}, {7.7f, 2.6f}, {6.0f, 2.2f}, {6.9f, 3.2f},
                {5.6f, 2.8f}, {7.7f, 2.8f}, {6.3f, 2.7f}, {6.7f, 3.3f}, {7.2f, 3.2f}, {6.2f, 2.8f},
                {6.1f, 3.0f}, {6.4f, 2.8f}, {7.2f, 3.0f}, {7.4f, 2.8f}, {7.9f, 3.8f}, {6.4f, 2.8f},
                {6.3f, 2.8f}, {6.1f, 2.6f}, {7.7f, 3.0f}, {6.3f, 3.4f}, {6.4f, 3.1f}, {6.0f, 3.0f},
                {6.9f, 3.1f}, {6.7f, 3.1f}, {6.9f, 3.1f}, {5.8f, 2.7f}, {6.8f, 3.2f}, {6.7f, 3.3f},
                {6.7f, 3.0f}, {6.3f, 2.5f}, {6.5f, 3.0f}, {6.2f, 3.4f}, {5.9f, 3.0f}, {5.8f, 3.0f}});

            y = np.array(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
                        0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
                        0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
                        1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
                        1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
                        2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
                        2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2);


            string url = "https://raw.githubusercontent.com/SciSharp/TensorFlow.NET/master/data/nb_example.npy";
            Web.Download(url, "nb", "nb_example.npy");
            #endregion
        }
    }
}