用Tensorflow来做随机森林预测

TensorFlow™ 是一个采用数据流图（data flow graphs），用于数值计算的开源软件库。节点（Nodes）在图中表示数学操作，图中的线（edges）则表示在节点间相互联系的多维数据数组，即张量（tensor）。它灵活的架构让你可以在多种平台上展开计算，例如台式计算机中的一个或多个CPU（或GPU），服务器，移动设备等等。

其本身入门门槛也相对较高，本次我们常见的随机森林模型演示一次完整的建模和预测过程，数据集我们采用MNIST图象数据，系统环境如下：

python：3.6.8

tensorflow-gpu：1.13.1

from __future__ import print_function

# 是否启用GPU，如果不用GPU此处留空
import os
os.environ["CUDA_VISIBLE_DEVICES"] = "0"

import tensorflow as tf
from tensorflow.python.ops import resources
from tensorflow.contrib.tensor_forest.python import tensor_forest
from tensorflow.examples.tutorials.mnist import input_data
import pandas as pd

随机森林建模：

tf.logging.set_verbosity(tf.logging.ERROR)
mnist = input_data.read_data_sets("data/", one_hot=False)
# 参数
num_steps = 500     # 训练次数
batch_size = 512    # 每批样本数
num_classes = 10    # 多少个分类0~9，10个
num_features = 784  # 特征数，每张图片是28*28点像素
num_trees = 500     # 随机森林参数，树的个数
max_nodes = 500     # 树的最大节点数

# 输入数据
X = tf.placeholder(tf.float32, shape=[None, num_features])
# Random forest, 标签必须是整数，（label编号也适用）
Y = tf.placeholder(tf.int32, shape=[None])

# Random Forest参数定义
hparams = tensor_forest.ForestHParams(num_classes=num_classes,
                                      num_features=num_features,
                                      num_trees=num_trees,
                                      max_nodes=max_nodes).fill()
# 创建Random Forest
forest_graph = tensor_forest.RandomForestGraphs(hparams)
# 生成图
train_op = forest_graph.training_graph(X, Y)
loss_op = forest_graph.training_loss(X, Y)

# 准确率计算
infer_op, _, _ = forest_graph.inference_graph(X)
correct_prediction = tf.equal(tf.argmax(infer_op, 1), tf.cast(Y, tf.int64))
accuracy_op = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))

# 初始化变量
init_vars = tf.group(tf.global_variables_initializer(),
                     resources.initialize_resources(resources.shared_resources()))

# 保存模型
saver = tf.train.Saver()

with tf.Session() as sess:

    # 运行初始化
    sess.run(init_vars)

    # 开始训练模型
    for i in range(1, num_steps + 1):
        # 数据准备
        # 获取下一批MNIST数据
        batch_x, batch_y = mnist.train.next_batch(batch_size)

        _, l = sess.run([train_op, loss_op], feed_dict={X: batch_x, Y: batch_y})

        if i % 100 == 0 or i == 1:
            acc = sess.run(accuracy_op, feed_dict={X: batch_x, Y: batch_y})
            print('Step %i, Loss: %f, Acc: %f' % (i, l, acc))
    # 测试模型
    test_x, test_y = mnist.test.images, mnist.test.labels
    value = pd.DataFrame(sess.run(infer_op, feed_dict={X: test_x}))
    value['Label'] = test_y
    print("Test Accuracy:", sess.run(accuracy_op, feed_dict={X: test_x, Y: test_y}))

    # 将模型保存到磁盘
    save_path = saver.save(sess, "model.ckpt")

此处我们可以看到value建模效果：

结果是对0~9这10个数字进行概率计算，概率最大的即我们预测结果，总体来说使用起来是非常方便的，计算结果：

此处我们保存模型为：model.ckpt，下面我们展示如何调用模型会用到。

加载我们训练好的模型：

mnist = input_data.read_data_sets("data/", one_hot=False)
# 测试
test_x, test_y = mnist.test.images, mnist.test.labels
with tf.Session() as sess:
    # 加载模型 import_meta_graph
    saver = tf.train.import_meta_graph("model.ckpt.meta")
    # 检查checkpoint
    saver.restore(sess, tf.train.latest_checkpoint("./"))
    # 获取图
    graph = tf.get_default_graph()

    X = graph.get_tensor_by_name("Placeholder:0")
    Y = graph.get_tensor_by_name("Placeholder_1:0")
    # 获取模型
    infer_op = graph.get_tensor_by_name('probabilities:0')
    value = pd.DataFrame(sess.run(infer_op, feed_dict={X: test_x}))
    value['Label'] = test_y

    correct_prediction = tf.equal(tf.argmax(infer_op, 1), tf.cast(Y, tf.int64))
    accuracy_op = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))

    print("Test Accuracy:", sess.run(accuracy_op, feed_dict={X: test_x, Y: test_y}))

我们用get_tensor_by_name加载模型，由于我们在建模时未指定变量名称，所以此处调用用默认的名称。

如此一个简单的调用完成了，下面是所有代码：

from __future__ import print_function

# 是否启用GPU，如果不用GPU此处留空
import os
os.environ["CUDA_VISIBLE_DEVICES"] = "0"

import tensorflow as tf
from tensorflow.python.ops import resources
from tensorflow.contrib.tensor_forest.python import tensor_forest
from tensorflow.examples.tutorials.mnist import input_data
import pandas as pd

# Random Forest（随机森林）算法是通过训练多个决策树，生成模型，然后综合利用多个决策树进行分类。


if __name__ == "__main__":

    is_Train = True

    if is_Train:
        tf.logging.set_verbosity(tf.logging.ERROR)
        mnist = input_data.read_data_sets("data/", one_hot=False)
        # 参数
        num_steps = 500     # 训练次数
        batch_size = 512    # 每批样本数
        num_classes = 10    # 多少个分类0~9，10个
        num_features = 784  # 特征数，每张图片是28*28点像素
        num_trees = 500     # 随机森林参数，树的个数
        max_nodes = 500     # 树的最大节点数

        # 输入数据
        X = tf.placeholder(tf.float32, shape=[None, num_features])
        # Random forest, 标签必须是整数，（label编号也适用）
        Y = tf.placeholder(tf.int32, shape=[None])

        # Random Forest参数定义
        hparams = tensor_forest.ForestHParams(num_classes=num_classes,
                                              num_features=num_features,
                                              num_trees=num_trees,
                                              max_nodes=max_nodes).fill()
        # 创建Random Forest
        forest_graph = tensor_forest.RandomForestGraphs(hparams)
        # 生成图
        train_op = forest_graph.training_graph(X, Y)
        loss_op = forest_graph.training_loss(X, Y)

        # 准确率计算
        infer_op, _, _ = forest_graph.inference_graph(X)
        correct_prediction = tf.equal(tf.argmax(infer_op, 1), tf.cast(Y, tf.int64))
        accuracy_op = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))

        # 初始化变量
        init_vars = tf.group(tf.global_variables_initializer(),
                             resources.initialize_resources(resources.shared_resources()))

        # 保存模型
        saver = tf.train.Saver()

        with tf.Session() as sess:

            # 运行初始化
            sess.run(init_vars)

            # 开始训练模型
            for i in range(1, num_steps + 1):
                # 数据准备
                # 获取下一批MNIST数据
                batch_x, batch_y = mnist.train.next_batch(batch_size)

                _, l = sess.run([train_op, loss_op], feed_dict={X: batch_x, Y: batch_y})

                if i % 100 == 0 or i == 1:
                    acc = sess.run(accuracy_op, feed_dict={X: batch_x, Y: batch_y})
                    print('Step %i, Loss: %f, Acc: %f' % (i, l, acc))
            # 测试模型
            test_x, test_y = mnist.test.images, mnist.test.labels
            value = pd.DataFrame(sess.run(infer_op, feed_dict={X: test_x}))
            value['Label'] = test_y
            print("Test Accuracy:", sess.run(accuracy_op, feed_dict={X: test_x, Y: test_y}))

            # 将模型保存到磁盘
            save_path = saver.save(sess, "model.ckpt")
    else:
        mnist = input_data.read_data_sets("data/", one_hot=False)
        # 测试
        test_x, test_y = mnist.test.images, mnist.test.labels
        with tf.Session() as sess:
            # 加载模型 import_meta_graph
            saver = tf.train.import_meta_graph("model.ckpt.meta")
            # 检查checkpoint
            saver.restore(sess, tf.train.latest_checkpoint("./"))
            # 获取图
            graph = tf.get_default_graph()

            X = graph.get_tensor_by_name("Placeholder:0")
            Y = graph.get_tensor_by_name("Placeholder_1:0")
            # 获取模型
            infer_op = graph.get_tensor_by_name('probabilities:0')
            value = pd.DataFrame(sess.run(infer_op, feed_dict={X: test_x}))
            value['Label'] = test_y

            correct_prediction = tf.equal(tf.argmax(infer_op, 1), tf.cast(Y, tf.int64))
            accuracy_op = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))

            print("Test Accuracy:", sess.run(accuracy_op, feed_dict={X: test_x, Y: test_y}))

参考资料：

1.http://www.tensorfly.cn/

2.https://tensorflow.google.cn/