TensorFlow入门－Tianic数据集训练

import pandas as pd
import tensorflow as tf
from sklearn.model_selection import train_test_split
import numpy as np

train_step = 5
train_path = 'train.csv'
is_train = False
learn_rate = 0.0001
epochs = 10

data = pd.read_csv(train_path)

# 取部分特征字段用于分类，并将所有缺失的字段填充为0
data['Sex'] = data['Sex'].apply(lambda s: 1 if s == 'male' else 0)
data = data.fillna(0)
dataset_X = data[['Sex', 'Age', 'Pclass', 'SibSp', 'Parch', 'Fare']]
dataset_X = dataset_X.as_matrix()

# 两种分类分别是幸存和死亡，'Survived'字段是其中一种分类的标签
# 新增'Deceased'字段表示第二种分类的标签，取值为'Survived'字段取非
data['Deceased'] = data['Survived'].apply(lambda s: int(not s))
dataset_Y = data[['Deceased', 'Survived']]
dataset_Y = dataset_Y.as_matrix()

# 使用sklearn的train_test_split函数将标记数据切分为‘训练数据集和验证数据集’
# 将全部标记数据随机洗牌后切分，其中验证数据占20%，由test_size参数指定
X_train, X_test, Y_train, Y_test = train_test_split(dataset_X, dataset_Y,
                                                    test_size=0.2, random_state=42)
# 声明输入数据点位符
X = tf.placeholder(tf.float32, shape=[None, 6])
Y = tf.placeholder(tf.float32, shape=[None, 2])
# 声明变量（参数）
W = tf.Variable(tf.random_normal([6, 2]), name='weights')
b = tf.Variable(tf.zeros([2]), name='bias')
# 构造前向传播计算图
y_pred = tf.nn.softmax(tf.matmul(X, W) + b)

# 使用交叉熵作为代价函数 Y * log(y_pred + e-10),程序中e-10,防止y_pred十分接近0或者1时，
# 计算（log0）会得到无穷，导致非法，进一步导致无法计算梯度，迭代陷入崩溃。
cross_entropy = -tf.reduce_sum(Y * tf.log(y_pred + 1e-10), reduction_indices=1)
# 批量样本的代价为所有样本交叉熵的平均值
cost = tf.reduce_mean(cross_entropy)
# 使用随机梯度下降算法优化器来最小化代价，系统自动构建反向传播部分的计算图
train_op = tf.train.GradientDescentOptimizer(learn_rate).minimize(cost)

saver = tf.train.Saver()
if is_train:
    with tf.Session() as sess:
        writer = tf.summary.FileWriter('logfile', sess.graph)
        # 初始化所有变量，必须最先执行
        tf.global_variables_initializer().run()
        # 以下为训练迭代，迭代10轮
        for epoch in range(10):
            total_loss = 0
            for i in range(len(X_train)):
                _, loss = sess.run([train_op, cost], feed_dict={X:[X_train[i]], Y:[Y_train[i]]})
                total_loss += loss
            print('Epoch: %04d, total loss=%.9f' % (epoch + 1, total_loss))
            # 保存model
            if (epoch + 1) % train_step == 0:
                save_path = saver.save(sess, './model/model.ckpt', global_step=epoch + 1)
        print('Training complete!')
        pred = sess.run(y_pred, feed_dict={X: X_test})
        # np.argmax的axis＝1表示第2轴最大值的索引（这里表示列与列对比，最大值的索引）
        correct = np.equal(np.argmax(pred, axis=1), np.argmax(Y_test, axis=1))
        accuracy = np.mean(correct.astype(np.float32))
        print("Accuracy on validation set: %.9f" % accuracy)
else:
    # 恢复model,继续训练
    with tf.Session() as sess1:
        # 从'checkpoint'文件中读出最新存档的路径
        ckpt = tf.train.get_checkpoint_state('./model')
        if ckpt and ckpt.model_checkpoint_path:
            saver.restore(sess1, ckpt.model_checkpoint_path)
            print('restore model sucess!')
        else:
            sys(0)
        print('continue train …………')
        for epoch in range(epochs):
            total_loss = 0
            for i in range(len(X_train)):
                _, loss = sess1.run([train_op, cost], feed_dict={X:[X_train[i]], Y:[Y_train[i]]})
                total_loss += loss
            print('Epoch: %04d, total loss=%.9f' % (epoch + 1, total_loss))
            # 保存model
            if (epoch + 1) % train_step == 0:
                save_path = saver.save(sess1, './model/model.ckpt', global_step=epoch + 1)
        print('Training complete!')
        pred = sess1.run(y_pred, feed_dict={X: X_test})
        # np.argmax的axis＝1表示第2轴最大值的索引（这里表示列与列对比，最大值的索引）
        correct = np.equal(np.argmax(pred, axis=1), np.argmax(Y_test, axis=1))
        accuracy = np.mean(correct.astype(np.float32))
        print("Accuracy on validation set: %.9f" % accuracy)

# 恢复model参数
with tf.Session() as sess2:
    # 从'checkpoint'文件中读出最新存档的路径
    print('restore lastest model, compute Accuracy!')
    ckpt = tf.train.get_checkpoint_state('./model')
    if ckpt and ckpt.model_checkpoint_path:
        saver.restore(sess2, ckpt.model_checkpoint_path)
    pred = sess2.run(y_pred, feed_dict={X: X_test})
    # np.argmax的axis＝1表示第2轴最大值的索引（这里表示列与列对比，最大值的索引）
    correct = np.equal(np.argmax(pred, axis=1), np.argmax(Y_test, axis=1))
    accuracy = np.mean(correct.astype(np.float32))
    print("Accuracy on validation set: %.9f" % accuracy)

TensorFlow自带的可视化工具TensorBoard

在当前目录的命令行下键入：tensorboard --logdir=logfile

根据命令行的提示，在浏览器里输入相应的网址。