mnist_tensorboard_embedding_visualization.py

import os
from keras.datasets import mnist
import matplotlib.pyplot as plt
from keras import backend as K
import numpy as np
import tensorflow as tf
from tensorflow.contrib.tensorboard.plugins import projector

batch_size = 128
num_classes = 10
epochs = 12

#建立資料夾，可以將callback儲存的log丟進來
log_dir = 'C:\\Users\\Lido_Lee\\Downloads\\mnist_callbacks'

# input image dimensions
img_rows, img_cols = 28, 28

# 直接從 Keras data 庫讀取 MNIST data
(x_train, y_train), (x_test, y_test) = mnist.load_data()

#判斷 Keras 後端讀取資料格式
if K.image_data_format() == 'channels_first':
    x_train = x_train.reshape(x_train.shape[0], 1, img_rows, img_cols)
    x_test = x_test.reshape(x_test.shape[0], 1, img_rows, img_cols)
    input_shape = (1, img_rows, img_cols)
else:
    x_train = x_train.reshape(x_train.shape[0], img_rows, img_cols, 1)
    x_test = x_test.reshape(x_test.shape[0], img_rows, img_cols, 1)
    input_shape = (img_rows, img_cols, 1)

#數據預處理，轉格float32格式，且值在0~1之間
x_train = x_train.astype('float32')
x_test = x_test.astype('float32')
x_test = x_test.reshape((10000,28*28))
print('x_train shape:', x_train.shape)
print(x_train.shape[0], 'train samples')
print(x_test.shape[0], 'test samples')

embed_count = 1600
x_test = x_test[:embed_count] / 255
y_test = y_test[:embed_count]

# setup the write and embedding tensor
summary_writer = tf.summary.FileWriter(log_dir)
embedding_var = tf.Variable(x_test, name='mnist_embedding')
config = projector.ProjectorConfig()
embedding = config.embeddings.add()
embedding.tensor_name = embedding_var.name

embedding.metadata_path = os.path.join(log_dir, 'metadata.tsv')
embedding.sprite.image_path = os.path.join(log_dir, 'sprite.png')
embedding.sprite.single_image_dim.extend([28, 28])

projector.visualize_embeddings(summary_writer, config)

# run the sesion to create the model check point

with tf.Session() as sesh:
    sesh.run(tf.global_variables_initializer())
    saver = tf.train.Saver()
    saver.save(sesh, os.path.join(log_dir, 'model.ckpt'))


# create the sprite image and the metadata file

rows = 28
cols = 28

label = ['0', '1', '2', '3', '4',
         '5', '6', '7', '8', '9']

sprite_dim = int(np.sqrt(x_test.shape[0]))

sprite_image = np.ones((cols * sprite_dim, rows * sprite_dim))

index = 0
labels = []
for i in range(sprite_dim):
    for j in range(sprite_dim):
        labels.append(label[int(y_test[index])])

        sprite_image[
        i * cols: (i + 1) * cols,
        j * rows: (j + 1) * rows
        ] = x_test[index].reshape(28, 28) * -1 + 1

        index += 1

with open(embedding.metadata_path, 'w') as meta:
    meta.write('Index\tLabel\n')
    for index, label in enumerate(labels):
        meta.write('{}\t{}\n'.format(index, label))

plt.imsave(embedding.sprite.image_path, sprite_image, cmap='gray')
plt.imshow(sprite_image, cmap='gray')
plt.show()