Tensorflow可視化程式設計

安裝Tensoflow1.0

Linux/ubuntu:

• python2.7:

pip install https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-1.0.1-cp27-none-linux_x86_64.whl
           

• python3.5:

pip3 install https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-1.0.1-cp35-cp35m-linux_x86_64.whl 
           

Maxos:

• python2:

pip install https://storage.googleapis.com/tensorflow/mac/cpu/tensorflow-1.0.1-py2-none-any.whl
           

• python3:

pip3 install https://storage.googleapis.com/tensorflow/mac/cpu/tensorflow-1.0.1-py3-none-any.whl
           

Tensorflow完成加法

import tensorflow as tf
# 消除警告(使用源碼安裝可自動消除)
import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'

a = tf.constant(3.0)
b = tf.constant(4.0)

with tf.Session() as sess:
    a_b = tf.add(a, b)
    print("相加後的類型為")
    print(a_b)
    print("真正的結果為:")
    print(sess.run(a_b))
           

tf_add

将加法運算以圖形化方式展示

• 在會話中添加記錄檔案的語句

import tensorflow as tf
# 消除警告(使用源碼安裝可自動消除)
import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'

a = tf.constant(3.0)
b = tf.constant(4.0)

with tf.Session() as sess:
    a_b = tf.add(a, b)
    print("相加後的類型為")
    print(a_b)
    print("真正的結果為:")
    print(sess.run(a_b))
    # 添加board記錄檔案
    file_write = tf.summary.FileWriter('/Users/lijianzhao/tensorBoard/', graph=sess.graph)
           

• 在終端運作

  tensorboard --logdir="/Users/lijianzhao/tensorBoard/"

在終端運作tensorboard

• 根據終端提示,在浏覽器鍵入

  http://192.168.199.213:6006

tensorboard主界面

• 選擇GRAPHS

實作簡單的線性回歸

import tensorflow as tf
# 消除警告(使用源碼安裝可自動消除)
import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'

# 回歸函數
def my_regression():

    # 準備10000 條資料x的平均值為5.0 标準差為1.0
    x = tf.random_normal([100, 1], mean = 5.0, stddev=1.0, name="x")
    # 真實的關系為 y = 0.7x + 0.6
    y_true = tf.matmul(x, [[0.7]]) + 0.6

    # 建立權重變量
    weight = tf.Variable(tf.random_normal([1, 1], mean=1.0, stddev=0.1), name="weight")

    # 建立偏置變量,初始值為1
    bias = tf.Variable(1.0, name="bias")

    # 預測結果
    y_predict = tf.matmul(x, weight) + bias

    # 計算損失
    loss = tf.reduce_mean(tf.square(y_predict - y_true))

    # 梯度下降減少損失,每次的學習率為0.1
    train_op = tf.train.GradientDescentOptimizer(0.01).minimize(loss)

    # 收集變量
    tf.summary.scalar("losses", loss)
    tf.summary.histogram("weightes", weight)

    # 合并變量
    merged = tf.summary.merge_all()

    # 初始化變量
    init_op = tf.global_variables_initializer()

    # 梯度下降優化損失
    with tf.Session() as sess:
        sess.run(init_op)

        print("初始的權重為{}, 初始的偏置為{}".format(weight.eval(), bias.eval()))

        # 添加board記錄檔案
        file_write = tf.summary.FileWriter('/Users/lijianzhao/tensorBoard/my_regression', graph=sess.graph)


        # 循環訓練線性回歸模型
        for i in range(20000):
            sess.run(train_op)
            print("訓練第{}次的權重為{}, 偏置為{}".format(i,weight.eval(), bias.eval()))

            # 觀察每次值的變化
            # 運作merge
            summery = sess.run(merged)
            # 每次收集到的值添加到檔案中
            file_write.add_summary(summery, i)


if __name__ == '__main__':
    my_regression()
           

運作結果

程式流程圖

損失值降低

權重逐漸接近真實值

為程式添加作用域

import tensorflow as tf
# 消除警告(使用源碼安裝可自動消除)
import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'

# 回歸函數
def my_regression():

    # 準備資料
    with tf.variable_scope("data"):
        # 準備10000 條資料x的平均值為5.0 标準差為1.0
        x = tf.random_normal([100, 1], mean = 5.0, stddev=1.0, name="x")
        # 真實的關系為 y = 0.7x + 0.6
        y_true = tf.matmul(x, [[0.7]]) + 0.6

    # 建立模型
    with tf.variable_scope ("model"):
        # 建立權重變量
        weight = tf.Variable(tf.random_normal([1, 1], mean=1.0, stddev=0.1), name="weight")

        # 建立偏置變量,初始值為1
        bias = tf.Variable(1.0, name="bias")

        # 預測結果
        y_predict = tf.matmul(x, weight) + bias

    # 計算損失
    with tf.variable_scope ("loss"):
        # 計算損失
        loss = tf.reduce_mean(tf.square(y_predict - y_true))

    # 減少損失
    with tf.variable_scope("optimizer"):
        # 梯度下降減少損失,每次的學習率為0.1
        train_op = tf.train.GradientDescentOptimizer(0.01).minimize(loss)

    # 收集變量
    tf.summary.scalar("losses", loss)
    tf.summary.histogram("weightes", weight)

    # 合并變量
    merged = tf.summary.merge_all()

    # 初始化變量
    init_op = tf.global_variables_initializer()

    # 梯度下降優化損失
    with tf.Session() as sess:
        sess.run(init_op)
        print("初始的權重為{}, 初始的偏置為{}".format(weight.eval(), bias.eval()))
        # 添加board記錄檔案
        file_write = tf.summary.FileWriter('/Users/lijianzhao/tensorBoard/my_regression', graph=sess.graph)
        # 循環訓練線性回歸模型
        for i in range(20000):
            sess.run(train_op)
            print("訓練第{}次的權重為{}, 偏置為{}".format(i,weight.eval(), bias.eval()))
            # 觀察每次值的變化
            # 運作merge
            summery = sess.run(merged)
            # 每次收集到的值添加到檔案中
            file_write.add_summary(summery, i)

if __name__ == '__main__':
    my_regression()

           

添加作用域

模型的儲存與恢複(儲存會話資源)

• 建立儲存模型的saver

saver = tf.train.Saver()
           

• 儲存模型

saver.save(sess, "./tmp/ckpt/test")
           

• 恢複模型

save.restore(sess, "./tmp/ckpt/test")