In [12]:

    

import numpy as np
import tensorflow as tf
import tensorboard_jupyter as tb
import matplotlib.pyplot as plt
from tensorflow.python.framework import ops
from sklearn import datasets

ops.reset_default_graph()
sess = tf.Session()

#2.6の回帰の例
batch_size = 25

# 平均1、標準偏差0.1の正規分布を作成
x_vals = np.random.normal(1, 0.1, 100)
with tf.name_scope('x'):
    x_data = tf.placeholder(shape=[None, 1], dtype=tf.float32)
# 10を100個
y_vals = np.repeat(10., 100)

with tf.name_scope('y'):
    y_target = tf.placeholder(shape=[None, 1], dtype=tf.float32)

# データをトレーニングセットとテストセットに8:2で分割
train_indices = np.random.choice(len(x_vals), round(len(x_vals) * 0.8), replace= False)
test_indices = np.array(list(set(range(len(x_vals))) - set(train_indices)))

x_vals_train = x_vals[train_indices]
x_vals_test = x_vals[test_indices]

y_vals_train = y_vals[train_indices]
y_vals_test = y_vals[test_indices]

# 予測結果の変数は最初はランダムで初期化
with tf.name_scope('A'):
    A = tf.Variable(tf.random_normal(shape=[1,1]))

# 計算グラフに乗算を追加する。
with tf.name_scope('model'):
    my_output = tf.matmul(x_data, A)
# L2損失関数(最小二乗法)
# 平均１のx_dataとAの乗算と10との差分のため、Aは10に収束するはず
with tf.name_scope('loss'):
    loss = tf.reduce_mean(tf.square(my_output - y_target))

# 変数最適化方法
my_opt = tf.train.GradientDescentOptimizer(learning_rate=0.02)
# L2損失を最小化するようにする
train_step = my_opt.minimize(loss)

# 変数初期化
init = tf.global_variables_initializer()
sess.run(init)

# トレーニング
for i in range(100):
    rand_index = np.random.choice(len(x_vals_train), size=batch_size)
    # 1行25列から25行1列へ変換
    rand_x = np.transpose([x_vals[rand_index]])
    rand_y = np.transpose([y_vals[rand_index]])
    sess.run(train_step, feed_dict={x_data: rand_x, y_target: rand_y})
    if (i+1)%25 == 0:
        print('Step #' + str(i+1) + ' A = ' + str(sess.run(A)))
        print('Loss = ' + str(sess.run(loss, feed_dict={x_data:rand_x, y_target: rand_y})))

mse_test = sess.run(loss, feed_dict={x_data: np.transpose([x_vals_test]), y_target: np.transpose([y_vals_test])})
mse_train = sess.run(loss, feed_dict={x_data: np.transpose([x_vals_train]), y_target: np.transpose([y_vals_train])})
print('MSE on test: ' + str(np.round(mse_test, 2)))
print('MSE on train: ' + str(np.round(mse_train, 2)))
        
tf.summary.FileWriter('./log/', sess.graph)
tb.show_graph(tf.get_default_graph().as_graph_def())


    

    




Step #25 A = [[ 6.20395088]]
Loss = 15.9325
Step #50 A = [[ 8.57587433]]
Loss = 1.88605
Step #75 A = [[ 9.45653248]]
Loss = 1.25016
Step #100 A = [[ 9.77085876]]
Loss = 0.994985
MSE on test: 1.27
MSE on train: 0.86






    






        
    

In [21]:

    

## 分類の例 ##
from tensorflow.python.framework import ops
ops.reset_default_graph()

sess = tf.Session()

# 平均-1と3のデータを50こずつ
x_vals = np.concatenate((np.random.normal(-1, 1, 50),
                        np.random.normal(2, 1, 50)))

with tf.name_scope('x_data') as scope:
    x_data = tf.placeholder(shape=[1, None], dtype=tf.float32)    
    
y_vals = np.concatenate([np.repeat(0., 50), np.repeat(1., 50)])
with tf.name_scope('y_target') as scope:
    y_target = tf.placeholder(shape=[1, None], dtype=tf.float32)

# データをトレーニングセットとテストセットに8:2で分割
train_indices = np.random.choice(len(x_vals), round(len(x_vals) * 0.8), replace= False)
test_indices = np.array(list(set(range(len(x_vals))) - set(train_indices)))

x_vals_train = x_vals[train_indices]
x_vals_test = x_vals[test_indices]

y_vals_train = y_vals[train_indices]
y_vals_test = y_vals[test_indices]
    
# 予測値は本来と離れた10で初期化
with tf.name_scope('A') as scope:
    A = tf.Variable(tf.random_normal(mean=10, shape=[1]))
    
# モデルは平均移動計算(sigmoid(x + A)だがsigmoidは損失関数が行う)
with tf.name_scope('Model') as scope:
    my_output = tf.add(x_data, A)

# シグモイド誤差エントロピー損失関数
xentropy = tf.nn.sigmoid_cross_entropy_with_logits(logits=my_output, labels=y_target)
# 最適化関数
my_opt = tf.train.GradientDescentOptimizer(0.05)
train_step = my_opt.minimize(xentropy)

init = tf.global_variables_initializer()
sess.run(init)

for i in range(1800):
    rand_index = np.random.choice(len(x_vals_train), size=batch_size)
    rand_x = [x_vals_train[rand_index]]
    rand_y = [y_vals_train[rand_index]]
    sess.run(train_step, feed_dict={x_data: rand_x, y_target: rand_y})
    if (i+1)%200==0:
        print('Step #' + str(i+1) + ' A = ' + str(sess.run(A)))
        print('Loss = ' + str(sess.run(xentropy, feed_dict={x_data: rand_x, y_target: rand_y})))


with tf.name_scope('prediction') as scope:
    y_prediction = tf.squeeze(tf.round(tf.nn.sigmoid(tf.add(x_data, A))))
    correct_prediction = tf.equal(y_prediction, y_target)
    accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
acc_value_test = sess.run(accuracy, feed_dict={x_data: [x_vals_test], y_target: [y_vals_test]})
acc_value_train = sess.run(accuracy, feed_dict={x_data: [x_vals_train], y_target: [y_vals_train]})
print('Accuracy on train set: ' + str(acc_value_train))
print('Accuracy on test set: ' + str(acc_value_test))
        
tf.summary.FileWriter('./log/', sess.graph)
tb.show_graph(tf.get_default_graph().as_graph_def())


    

    




Step #200 A = [-0.69047976]
Loss = [[ 0.16616148  0.65559411  0.89609718  0.05554588  0.08533639  0.46616307
   0.07652663  0.10405916  0.02700546  0.11825523  0.95115638  0.03222379
   0.08100253  0.4198207   0.11449566  0.11449566  0.10343304  0.08100253
   0.10612128  0.42004126  0.4198207   0.19641592  0.07536426  0.07536426
   0.10612128]]
Step #400 A = [-0.56424481]
Loss = [[ 0.24569601  0.1195773   0.06775134  0.52938318  0.1195773   0.12276029
   0.08553281  0.05267632  0.29887959  0.43578029  0.15737741  0.4063105
   0.09197678  0.05463177  0.21694562  0.12276029  0.46515566  0.07558858
   0.1195773   0.10910847  0.05534232  0.38110635  0.33341065  0.06278943
   0.06079619]]
Step #600 A = [-0.6002931]
Loss = [[ 0.06295894  0.05455804  0.11334316  0.08955632  0.26493016  1.11108971
   0.13227196  0.11636318  0.43350312  0.07015296  0.05731658  0.05455804
   0.05584619  0.41891679  0.10110906  0.11266427  0.15220828  0.43464652
   0.17198761  0.46298206  0.10512213  1.11108971  0.12293167  0.11266427
   0.11636318]]
Step #800 A = [-0.6427415]
Loss = [[ 0.21864513  0.11751396  0.13938749  0.05818432  0.03074447  0.22911347
   0.31312305  0.4474602   0.31975704  0.26314947  0.13333322  0.07197441
   0.11103227  0.46421763  0.26314947  0.11103227  0.13938749  0.09220471
   0.08532066  0.14856726  0.10799613  0.13214977  0.30573937  0.92213017
   0.14354172]]
Step #1000 A = [-0.53698671]
Loss = [[ 0.8597796   0.16384301  0.1226868   0.23965625  0.05172425  0.25046194
   0.37254855  0.1054479   0.29190826  0.1226868   0.08428752  0.18035971
   0.24031614  0.63877755  0.14716148  0.09382833  0.23212351  0.44505167
   0.06498085  0.08332601  0.06598832  0.05389354  0.13658904  0.04105734
   0.21168759]]
Step #1200 A = [-0.66117775]
Loss = [[ 0.4301883   0.37497833  0.03309396  0.47110409  0.13445054  0.05836788
   0.13107556  0.18185802  0.40434626  0.4299635   0.18185802  0.93327552
   0.04739198  0.07326578  0.30819991  0.93327552  0.11607544  0.05263273
   0.14109612  0.02779742  0.58206779  0.93327552  0.11607544  0.30091578
   0.23660219]]
Step #1400 A = [-0.6949929]
Loss = [[ 0.10361405  0.4184956   0.46784705  0.28768548  0.46784705  0.16685373
   0.02241759  0.03201102  0.03201102  0.20859651  0.05530255  0.39112297
   0.15594549  0.05981818  0.09071283  0.16685373  0.06203062  0.07569257
   0.10299046  0.12695804  0.89342856  0.13671103  0.48394245  0.16685373
   0.29934195]]
Step #1600 A = [-0.75468701]
Loss = [[ 0.25704432  0.20718922  0.29018718  0.06390999  0.12164412  0.61209434
   0.04804517  0.12002506  0.1025841   0.07309439  0.03018377  0.27744466
   0.10375654  0.68698418  0.08140018  0.12164412  0.2842378   0.09074595
   0.13056147  0.37218708  0.03432439  0.08725557  0.3030504   0.3218323
   0.12004825]]
Step #1800 A = [-0.71812469]
Loss = [[ 0.11752354  0.02627871  0.31190363  0.07858957  1.03360128  0.07739706
   0.31190363  0.29361987  0.07739706  0.21260592  0.41042888  0.19154167
   0.24887322  0.12422799  0.37949657  0.21260592  0.40222523  0.09907759
   0.4928804   0.34095863  0.28642654  0.27243623  0.10337221  0.40222523
   0.12425194]]
Accuracy on train set: 0.9625
Accuracy on test set: 0.95






    






        
    

In [24]:

    

# 可視化
A_result = -sess.run(A)
bins = np.linspace(-5, 5, 50)
plt.hist(x_vals[0:50], bins, alpha = 0.5, label='N(-1,1)', color='gray')
plt.hist(x_vals[50:100], bins[0:50], alpha = 0.5, label='N(2,1)', color='red')
plt.plot((A_result, A_result), (0,8), 'k--', linewidth=3, label='A = ' + str(np.round(A_result, 2)))
plt.legend(loc='upper right')
plt.title('Binary Classifier Accuracy = ' + str(np.round(acc_value_test, 2)))
plt.show()


    

In [ ]: