利用tensorboard将数据可视化(tf1.x 和 tf2.x)

注：代码是网上下载的，但是找不到原始出处了，侵权则删

先写出visual类：

tf1.x:

class TF_visualizer(object):
    def __init__(self, dimension, vecs_file, metadata_file, output_path):
        self.dimension = dimension
        self.vecs_file = vecs_file
        self.metadata_file = metadata_file
        self.output_path = output_path
        
        self.vecs = []
        with open(self.vecs_file, 'r') as vecs:
        #with open(self.vecs_file, 'rb') as vecs:
            for i, line in enumerate(vecs):
                if line != '': self.vecs.append(line)

    def visualize(self):
        # adding into projector
        config = projector.ProjectorConfig()

        placeholder = np.zeros((len(self.vecs), self.dimension))
        
        for i, line in enumerate( self.vecs ):   
            placeholder[i] = np.fromstring(line, sep=',')
        #for i,line in enumerate(self.vecs):
        #    placeholder[i] = np.fromstring(line)

        embedding_var = tf.Variable(placeholder, trainable=False, name='amazon')

        embed = config.embeddings.add()
        embed.tensor_name = embedding_var.name
        embed.metadata_path = self.metadata_file

        # define the model without training
        sess = tf.InteractiveSession()
        
        tf.global_variables_initializer().run()
        saver = tf.train.Saver()
        
        saver.save(sess, os.path.join(self.output_path, 'w2x_metadata.ckpt'))

        writer = tf.summary.FileWriter(self.output_path, sess.graph)
        projector.visualize_embeddings(writer, config)
        sess.close()
        print('Run `tensorboard --logdir={0}` to run visualize result on tensorboard'.format(self.output_path))

tf2.x:

class TF_visualizer(object):
    def __init__(self, dimension, vecs_file, metadata_file, output_path):
        self.dimension = dimension
        self.vecs_file = vecs_file
        self.metadata_file = metadata_file
        self.output_path = output_path
        
        self.vecs = []
        with open(self.vecs_file, 'r') as vecs:
        #with open(self.vecs_file, 'rb') as vecs:
            for i, line in enumerate(vecs):
                if line != '': self.vecs.append(line)

    def visualize(self):
        # adding into projector
        config = projector.ProjectorConfig()

        placeholder = np.zeros((len(self.vecs), self.dimension))
        
        for i, line in enumerate( self.vecs ):   
            placeholder[i] = np.fromstring(line, sep=',')
        #for i,line in enumerate(self.vecs):
        #    placeholder[i] = np.fromstring(line)

        embedding_var = tf.Variable(placeholder, trainable=False, name='kmeans')
        embed = config.embeddings.add()
        embed.tensor_name = "embedding/.ATTRIBUTES/VARIABLE_VALUE"
        embed.metadata_path = self.metadata_file

        checkpoint = tf.train.Checkpoint(embedding=embedding_var)
        checkpoint.save(os.path.join(self.output_path, 'w2x_metadata.ckpt'))

        projector.visualize_embeddings(self.output_path,config)

        #sess.run(tf.compat.v1.global_variables_initializer())
        #saver = tf.compat.v1.train.Saver()
        #saver.save(sess, os.path.join(self.output_path, 'w2x_metadata.ckpt'),STEP)
        #train_writer = tf.summary.create_file_writer('./logs/1/train')
        #writer = tf.summary.FileWriter(self.output_path, sess.graph)
        #writer = tf.summary.create_file_writer(self.output_path)
        #with writer.as_default():
        #    tf.summary.projector(tensor=embedding_var,labels=self.vecs,step=STEP,name='desired name')
        #writer.write(sess.graph)
        #projector.visualize_embeddings(writer, config)
        #sess.close()
        print('Run `tensorboard --logdir={0}` to run visualize result on tensorboard'.format(self.output_path))

然后调用类：

output = '/home/xx'

# create a new tensor board visualizer
visualizer = TF_visualizer(dimension = 768,
                           vecs_file = os.path.join(output, 'amazon_vec.tsv'),
                           #vecs_file = os.path.join(output, 'mnist_10k_784d_tensors.bytes'),
                           metadata_file = os.path.join(output, 'amazon.tsv'),
                           output_path = output)
visualizer.visualize()

其中，amazon_vec.tsv中存放向量（包括词向量，句子向量...）,amazon.tsv中存放原始数据，格式为id,label,title，id和title可以随意定义，label则为对应向量的标识，两个文件是 一一对应的（即amazon_vec中的第一行数据对应amazon中第一行数据）

注：amazon.tsv举例：

0,快乐,0
1,幸福,1
2,哪里,2
3,剪子,3
4,鼠标,4

amazon_vec.tsv举例：

0.21729235351085663,-0.21714180707931519,0.10137219727039337,0.530093789100647,-0.16228507459163666

1.43824303150177,0.40661126375198364,-1.3043369054794312,0.4775696396827698,-0.21097205579280853

-0.3607819080352783,0.9494154453277588,1.102367877960205,1.2270256280899048,0.4637971818447113

0.7370161414146423,-1.6456717252731323,-2.1842262744903564,2.0185391902923584,1.6656044721603394

1.03994882106781,-1.7641232013702393,1.042765736579895,2.6722264289855957,1.6226638555526733

最后，命令行输入

tensorboard --logdir=/home/xx

在浏览器输入http://xx-desktop:6006即可看到可视化的数据（6006是默认端口）