06.tensorboard01_example

In [1]:

import numpy as np
import pandas as pd
import tensorflow as tf

In [2]:

data = pd.read_csv("./datas/classification_ex.csv", dtype="float32").values

In [3]:

x_data = data[:, 0:2]
y_data = data[:, 2:]

model save¶

• global_step -> trainable=False

In [4]:

global_step = tf.Variable(0, trainable=False, name="global_step")

model setting¶

In [5]:

# ground truth
X = tf.placeholder(tf.float32)
Y = tf.placeholder(tf.float32)

# weight, bias
W1 = tf.Variable(tf.random_normal([2, 10],  mean=0, stddev=1))
W2 = tf.Variable(tf.random_normal([10, 20], mean=0, stddev=1))
W3 = tf.Variable(tf.random_normal([20, 3],  mean=0, stddev=1))

b1 = tf.zeros([10])
b2 = tf.zeros([20])
b3 = tf.zeros([3])

# Layer 1
with tf.name_scope("Layer_1"):
    L1 = tf.add(tf.matmul(X, W1), b1)
    L1 = tf.nn.sigmoid(L1)

# Layer 2
with tf.name_scope("Layer_2"):
    L2 = tf.add(tf.matmul(L1, W2), b2)
    L2 = tf.nn.sigmoid(L2)

# Layer 3
with tf.name_scope("Layer_3"):
    model = tf.add(tf.matmul(L2, W3), b3)
    model = tf.nn.softmax(model)

    
# cost function
with tf.name_scope("optimizer"):
    cost = tf.reduce_mean(-tf.reduce_sum(Y*tf.log(model) + (1-Y)*tf.log(1-model)))
    optimizer = tf.train.AdamOptimizer(learning_rate=0.01)
    train_op = optimizer.minimize(cost, global_step=global_step)
    tf.summary.scalar("cost", cost)
    tf.summary.histogram("Weight1", W1)
    tf.summary.histogram("Weight2", W2)
    tf.summary.histogram("Weight3", W3)
    tf.summary.histogram("bais1", b1)
    tf.summary.histogram("bais2", b2)
    tf.summary.histogram("bais3", b3)

create session¶

In [6]:

sess = tf.Session()
saver = tf.train.Saver(tf.global_variables())

tf.global_variables()는 앞서 정의한 변수들을 가져오는 함수

make checkpoint¶

In [7]:

ckpt = tf.train.get_checkpoint_state("./model")
if ckpt and tf.train.checkpoint_exists(ckpt.model_checkpoint_path):
    saver.restore(sess, ckpt.model_checkpoint_path)
    
else:
    sess.run(tf.global_variables_initializer())

tf.summary.merge_all 함수로 앞서 지정한 텐서들을 수집한 다음 tf.summary.FileWriter 함수를 이용해 그래프와 텐서들의 값을 저장할 디렉토리를 설정

In [8]:

merged = tf.summary.merge_all()
writer = tf.summary.FileWriter("./logs", sess.graph)

running model¶

In [9]:

for step in range(1000):
    sess.run(train_op, feed_dict={X:x_data, Y:y_data})
    
    if (step+1) % 50 == 0:    
        print("step: {}, cost: {:.5f}".\
              format(sess.run(global_step), 
                     sess.run(cost, feed_dict={X:x_data, Y:y_data})))
        
    summary = sess.run(merged, feed_dict={X:x_data, Y:y_data})
    writer.add_summary(summary, global_step=sess.run(global_step))

step: 50, cost: 6.33244
step: 100, cost: 2.69545
step: 150, cost: 0.85424
step: 200, cost: 0.35161
step: 250, cost: 0.19246
step: 300, cost: 0.12306
step: 350, cost: 0.08628
step: 400, cost: 0.06428
step: 450, cost: 0.04998
step: 500, cost: 0.04012
step: 550, cost: 0.03300
step: 600, cost: 0.02768
step: 650, cost: 0.02359
step: 700, cost: 0.02037
step: 750, cost: 0.01779
step: 800, cost: 0.01568
step: 850, cost: 0.01393
step: 900, cost: 0.01246
step: 950, cost: 0.01122
step: 1000, cost: 0.01016

In [10]:

saver.save(sess, "./model/dnn.ckpt", global_step=global_step)

Out[10]:

'./model/dnn.ckpt-1000'

In [11]:

prediction = tf.argmax(model, 1)
target = tf.argmax(Y, 1)
print("prediction: \t{}".format(sess.run(prediction, feed_dict={X: x_data})))
print("target: \t{}".format(sess.run(target, feed_dict={Y:y_data})))

is_correct = tf.equal(prediction, target)
accuracy = tf.reduce_mean(tf.cast(is_correct, tf.float32))
print("\naccuracy: \t{:.3f}%".format(sess.run(accuracy*100, feed_dict={X: x_data, Y: y_data})))

prediction: 	[2 0 1 2 2 1]
target: 	[2 0 1 2 2 1]

accuracy: 	100.000%

In [12]:

# tensorboard --logdir=./logs

In [13]:

from IPython.core.display import display, HTML

display(HTML("<style> .container{width:100% !important;}</style>"))