show_graph()

​tf.reset_default_graph()
import tensorflow as tf
import numpy as np
import tensorboard as tb

#placeholder宣言
x = tf.placeholder(tf.float32, name='x')
t = tf.placeholder(tf.float32, name='t')

# 変数宣言
with tf.name_scope('variables'):
    W = tf.Variable(np.random.uniform(low=-0.07, high=0.07, size=(2, 1)).astype('float32'), name='W') # 2行1列
    b = tf.Variable(np.zeros(1).astype('float32'), name='b')

# モデルの設定
with tf.name_scope('model'):
    y = tf.nn.sigmoid(tf.matmul(x, W) + b, name='y')

# E = 1/N・Σn=0->N-1{t_n・logy_n + (1 -t_n)log(1 -y_n)}
with tf.name_scope('train'):
    cost = -tf.reduce_mean(t*tf.log(tf.clip_by_value(y, 1e-10, 1.0)) + (1 - t)*tf.log(tf.clip_by_value(1 - y, 1e-10, 1.0)))

with tf.name_scope('gradients'):
    dW, db = tf.gradients(cost, [W, b])
    updates = [
         W.assign_add(-0.01*dW),
         b.assign_add(-0.01*db)
    ]

train = tf.group(*updates)
# OR(どちらかが１なら１、どちらも０のときのみ０)
train_X = np.array([[0, 1], [1, 0], [0, 0], [1, 1]])
train_y = np.array([[1], [1], [0], [1]])

# 学習（10000回の繰り返し、1000回ごとに誤差を表示)
with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())

    for i in range(10000):
        _cost, _ = sess.run([cost, train], feed_dict={x: train_X, t: train_y})
        if (i+1)%1000==0:
            print(_cost)
  
    tf.summary.FileWriter('./log/', sess.graph) # Tensorboad用に logフォルダにログを作成

tb.show_graph(tf.get_default_graph().as_graph_def()) # グラフの表示