利用tensorboard将数据可视化

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：

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))

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）

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）