TensorBoard使用
為了更方便TensorFlow程序的理解、調試與優化,Google發布了一套叫做TensorBoard的可視化工具,可以用TensorBoard來展現TensorFlow的圖像,繪制圖像生成的定量指標圖以及附加數據。
TensorBoard設置完成之后的樣子應該如下圖:
image
其基本原理見TensorBoard中文手冊,內有詳細的介紹。
本文參考了放羊的水瓶的博文。
下面通過三個例程,來講解其使用:
例程1 矩陣相乘 tfboard1.py
import tensorflow as tf
with tf.name_scope('graph') as scope:
matrix1 = tf.constant([[3., 3.]], name = 'matrix') # 一行兩列
matrix2 = tf.constant([[2.], [2.]], name = 'matrix2') # 兩行一列
product = tf.matmul(matrix1, matrix2, name = 'product')
sess = tf.Session()
writer = tf.summary.FileWriter("logs1/", sess.graph)
init = tf.global_variables_initializer()
sess.run(init)
tf.name_scope函數是作用域名,上述代碼斯即在graph作用域op下,又有三個op(分別是matrix1,matrix2,product),用tf函數內部的name參數命名,這樣會在tensorboard中顯示。
運行上述代碼后,在項目所在目錄會生成"logs1"目錄(可以自定義名字),然后在命令行運行:
tensorboard --logdir logs1
即可在本機6006端口調用TensorBoard。可以通過瀏覽器打開使用。
例程2 線性擬合(一) tfboard2.py
例程1中沒有任何訓練過程,比較簡單,下面通過這個例子來畫出它的張量流動圖。
import tensorflow as tf
import numpy as np
# 準備原始數據
with tf.name_scope('data'):
x_data = np.random.rand(100).astype(np.float32)
y_data = 0.3*x_data + 0.1
# 參數設置
with tf.name_scope('parameters'):
weight = tf.Variable(tf.random_uniform([1], -1.0, 1.0))
bias = tf.Variable(tf.zeros([1]))
# 得到 y_prediction
with tf.name_scope('y_prediction'):
y_prediction = weight*x_data + bias
# 計算損失率compute the loss
with tf.name_scope('loss'):
loss = tf.reduce_mean(tf.square(y_data - y_prediction))
#
optimizer = tf.train.GradientDescentOptimizer(0.5)
with tf.name_scope('train'):
train = optimizer.minimize(loss)
with tf.name_scope('init'):
init = tf.global_variables_initializer()
sess = tf.Session()
writer = tf.summary.FileWriter("logs2/",sess.graph)
sess.run(init)
for step in range(101):
sess.run(train)
if step%10 == 0:
print(step, 'weight', sess.run(weight), 'bias:', sess.run(bias))
例程3 線性擬合(二) tfboard3.py
對例程二代碼進行修改,嘗試tensorboard的其他功能,例如scalars,distributions,histograms,這些功能對于分析學習算法的性能有很大幫助。
import tensorflow as tf
import numpy as np
with tf.name_scope('data'):
x_data = np.random.rand(100).astype(np.float32)
y_data = 0.3*x_data + 0.1
with tf.name_scope('paremeters'):
with tf.name_scope('weights'):
weight = tf.Variable(tf.random_uniform([1], -1.0, 1.0))
tf.summary.histogram('weight', weight)
with tf.name_scope('biases'):
bias = tf.Variable(tf.zeros([1]))
tf.summary.histogram('bias', bias)
with tf.name_scope('y_prediction'):
y_prediction = weight*x_data + bias
with tf.name_scope('loss'):
loss = tf.reduce_mean(tf.square(y_data - y_prediction))
tf.summary.scalar('loss', loss)
optimizer = tf.train.GradientDescentOptimizer(0.5)
with tf.name_scope('train'):
train = optimizer.minimize(loss)
with tf.name_scope('init'):
init = tf.global_variables_initializer()
sess = tf.Session()
merged = tf.summary.merge_all()
writer = tf.summary.FileWriter("logs3/", sess.graph)
sess.run(init)
for step in range(101):
sess.run(train)
rs = sess.run(merged)
writer.add_summary(rs, step)