KMeansClustering.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.Diagnostics;
using Tensorflow;
using Tensorflow.NumPy;
using static Tensorflow.Binding;
using System;

namespace TensorFlowNET.Examples
{
    /// <summary>
    /// Implement K-Means algorithm with TensorFlow.NET, and apply it to classify
    /// handwritten digit images.
    /// https://github.com/aymericdamien/TensorFlow-Examples/blob/master/examples/2_BasicModels/kmeans.py
    /// </summary>
    public class KMeansClustering : SciSharpExample, IExample
    {
        public int? train_size = null;
        public int validation_size = 5000;
        public int? test_size = null;
        public int batch_size = 1024; // The number of samples per batch

        Datasets<MnistDataSet> mnist;
        NDArray full_data_x;
        int num_steps = 20; // Total steps to train
        int k = 25; // The number of clusters
        int num_classes = 10; // The 10 digits

        float accuray_test = 0f;

        public ExampleConfig InitConfig()
            => Config = new ExampleConfig
            {
                Name = "K-means Clustering",
                Enabled = false,
                IsImportingGraph = true
            };

        public bool Run()
        {
            PrepareData();
            var graph = ImportGraph();
            using (var sess = tf.Session(graph))
            {
                Train(sess);
            }

            return accuray_test > 0.70;
        }

        public override void PrepareData()
        {
            var loader = new MnistModelLoader();

            var setting = new ModelLoadSetting
            {
                TrainDir = ".resources/mnist",
                OneHot = true,
                TrainSize = train_size,
                ValidationSize = validation_size,
                TestSize = test_size,
                ShowProgressInConsole = true
            };

            mnist = loader.LoadAsync(setting).Result;

            full_data_x = mnist.Train.Data;

            // download graph meta data
            string url = "https://raw.githubusercontent.com/SciSharp/TensorFlow.NET/master/graph/kmeans.meta";
            loader.DownloadAsync(url, ".resources/graph", "kmeans.meta").Wait();
        }

        public override Graph ImportGraph()
        {
            var graph = tf.Graph().as_default();

            tf.train.import_meta_graph(".resources/graph/kmeans.meta");

            return graph;
        }

        public void Train(Session sess)
        {
            var graph = tf.Graph();

            // Input images
            Tensor X = graph.get_operation_by_name("Placeholder"); // tf.placeholder(tf.float32, shape: new TensorShape(-1, num_features));
            //  Labels (for assigning a label to a centroid and testing)
            Tensor Y = graph.get_operation_by_name("Placeholder_1"); // tf.placeholder(tf.float32, shape: new TensorShape(-1, num_classes));

            // K-Means Parameters
            //var kmeans = new KMeans(X, k, distance_metric: KMeans.COSINE_DISTANCE, use_mini_batch: true);

            // Build KMeans graph
            //var training_graph = kmeans.training_graph();

            var init_vars = tf.global_variables_initializer();
            Tensor init_op = graph.get_operation_by_name("cond/Merge");
            var train_op = graph.get_operation_by_name("group_deps");
            Tensor avg_distance = graph.get_operation_by_name("Mean");
            Tensor cluster_idx = graph.get_operation_by_name("Squeeze_1");
            NDArray[] result = null;

            sess.run(init_vars, new FeedItem(X, full_data_x));
            sess.run(init_op, new FeedItem(X, full_data_x));

            // Training
            var sw = new Stopwatch();

            foreach (var i in range(1, num_steps + 1))
            {
                sw.Restart();
                result = sess.run(new ITensorOrOperation[] { train_op, avg_distance, cluster_idx }, new FeedItem(X, full_data_x));
                sw.Stop();

                if (i % 4 == 0 || i == 1)
                    print($"Step {i}, Avg Distance: {result[1]} Elapse: {sw.ElapsedMilliseconds}ms");
            }

            var idx = result[2].ToArray<int>();

            // Assign a label to each centroid
            // Count total number of labels per centroid, using the label of each training
            // sample to their closest centroid (given by 'idx')
            var counts = np.zeros((k, num_classes), np.float32);

            sw.Start();
            foreach (var i in range(idx.Length))
            {
                var x = mnist.Train.Labels[i];
                // counts[idx[i]] += x;
                throw new NotImplementedException("");
            }

            sw.Stop();
            print($"Assign a label to each centroid took {sw.ElapsedMilliseconds}ms");

            // Assign the most frequent label to the centroid
            var labels_map_array = np.argmax(counts, 1);
            var labels_map = tf.convert_to_tensor(labels_map_array);

            // Evaluation ops
            // Lookup: centroid_id -> label
            var cluster_label = tf.nn.embedding_lookup(labels_map, cluster_idx);

            // Compute accuracy
            var correct_prediction = tf.equal(cluster_label, tf.cast(tf.math.argmax(Y, 1), tf.int32));
            var cast = tf.cast(correct_prediction, tf.float32);
            var accuracy_op = tf.reduce_mean(cast);

            // Test Model
            var (test_x, test_y) = (mnist.Test.Data, mnist.Test.Labels);
            // result = sess.run(accuracy_op, new FeedItem(X, test_x), new FeedItem(Y, test_y));
            // accuray_test = result;
            print($"Test Accuracy: {accuray_test}");
        }
    }
}