add visualize embeddings codes

code/tensorflow-tools/log/checkpoint

@@ -0,0 +1,2 @@
+model_checkpoint_path: "model.ckpt-0"
+all_model_checkpoint_paths: "model.ckpt-0"

code/tensorflow-tools/log/metadata.tsv

@@ -0,0 +1,100 @@
+7
+2
+1
+0
+4
+1
+4
+9
+5
+9
+0
+6
+9
+0
+1
+5
+9
+7
+3
+4
+9
+6
+6
+5
+4
+0
+7
+4
+0
+1
+3
+1
+3
+4
+7
+2
+7
+1
+2
+1
+1
+7
+4
+2
+3
+5
+1
+2
+4
+4
+6
+3
+5
+5
+6
+0
+4
+1
+9
+5
+7
+8
+9
+3
+7
+4
+6
+4
+3
+0
+7
+0
+2
+9
+1
+7
+3
+2
+9
+7
+7
+6
+2
+7
+8
+4
+7
+3
+6
+1
+3
+6
+9
+3
+1
+4
+1
+7
+6
+9

code/tensorflow-tools/log/projector_config.pbtxt

@@ -0,0 +1,9 @@
+embeddings {
+  tensor_name: "embedding:0"
+  metadata_path: "metadata.tsv"
+  sprite {
+    image_path: "mnist_10k_sprite.png"
+    single_image_dim: 28
+    single_image_dim: 28
+  }
+}

code/tensorflow-tools/tensorflow_embeddings.py

@@ -0,0 +1,43 @@
+#-*- coding: utf-8 -*-
+# Author: Lawlite
+# Date: 2017/07/26
+# Associate Blog: http://lawlite.me/2017/06/24/Tensorflow学习-工具相关/#1、可视化embedding
+# License: MIT
+
+import numpy as np
+import tensorflow as tf
+from tensorflow.contrib.tensorboard.plugins import projector
+from tensorflow.examples.tutorials.mnist import input_data
+import os
+
+MNIST_DATA_PATH = 'MNIST_data'
+LOG_DIR = 'log'
+SPRITE_IMAGE_FILE = 'mnist_10k_sprite.png'
+META_DATA_FILE = 'metadata.tsv'
+IMAGE_NUM = 100
+
+mnist = input_data.read_data_sets(MNIST_DATA_PATH, one_hot=False)
+plot_array = mnist.test.images[:IMAGE_NUM]   # 取前100个图片
+np.savetxt(os.path.join(LOG_DIR, META_DATA_FILE), mnist.test.labels[:IMAGE_NUM], fmt='%d')  # label 保存为metadata.tsv文件
+
+
+'''可视化embedding, 3个步骤'''
+with tf.Session() as sess:
+    '''1、 将2D矩阵放入Variable中国'''
+    embeddings_var = tf.Variable(plot_array, name='embedding')
+    tf.global_variables_initializer().run()
+
+    '''2、 保存到文件中'''
+    saver = tf.train.Saver([embeddings_var])
+    sess.run(embeddings_var.initializer)
+    saver.save(sess, os.path.join(LOG_DIR, "model.ckpt"), global_step=0)
+
+    '''3、 关联metadata和sprite图片'''
+    summary_writer = tf.summary.FileWriter(LOG_DIR)
+    config = projector.ProjectorConfig()
+    embedding = config.embeddings.add()
+    embedding.tensor_name = embeddings_var.name
+    embedding.metadata_path = META_DATA_FILE
+    embedding.sprite.image_path = SPRITE_IMAGE_FILE
+    embedding.sprite.single_image_dim.extend([28, 28])
+    projector.visualize_embeddings(summary_writer, config)

code/triplet-loss/test_triplet_loss.py

@@ -264,11 +264,11 @@ def test_batch_hard_triplet_loss():
 
 #test_pairwise_distances()
 #test_gradients_pairwise_distances()
-#test_anchor_positive_triplet_mask()
-#test_anchor_negative_triplet_mask()
 #test_triplet_mask()
 #test_pairwise_distances_are_positive()
-test_batch_all_triplet_loss()
+#test_batch_all_triplet_loss()
+#test_anchor_positive_triplet_mask()
+#test_anchor_negative_triplet_mask()
 #test_batch_hard_triplet_loss()
 
 

code/triplet-loss/triplet_loss.py

@@ -66,7 +66,8 @@ def _get_triplet_mask(labels):
 
 def batch_all_triplet_loss(labels, embeddings, margin, squared=False):
     '''
-       triplet loss of a batch
+       triplet loss of a batch, 注意这里的loss一般不是收敛的，因为是计算的semi-hard和hard的距离均值，因为每次是先选择出semi-hard和hard
+       triplet, 那么上次优化后的可能就选择不到了，所以loss并不会收敛，但是fraction_postive_triplets是收敛的，因为随着优化占的比例是越来越少的
        -------------------------------
        Args:
           labels:     标签数据，shape = （batch_size,）