【機器學習】用 tensorflow 實現隨機森林 RandomForest in tensorflow (mnist 示例)

隨機森林

隨機森林的基本知識，許多博客都已經講的很清晰。 但是用 tensorflow 實現的代碼卻很少。
博主屬於機器學習未入門級別，純屬分享一段可以運行的隨機森林代碼。若有錯誤，麻煩指出。

本代碼實現用 mnist 示例，通過 tensorflow 實現隨機森林。

import tensorflow as tf
import pandas as pd
import numpy as np
from tensorflow.contrib.tensor_forest.python import tensor_forest
from tensorflow.python.ops import resources

dataset = pd.read_csv('./data/train.csv').values

####################
# 讀入數據進行數據處理
# 最終 X.shape 爲 [?,784]
# y.shape 爲 [?]，y 必須爲數字標籤，非 onehot標籤。 
# 
# X 需要歸一化處理
####################
# paremeters
num_steps = 5000    # 迭代次數
batch_size = 1000   # 每個處理批次的大小
num_classes = 10    # 類別的數目
num_features = 784  # 特徵的數目
num_trees = 50      # 森林裏樹的個數
max_nodes = 800     # 每棵數的最大節點數目

# 定義 X 和 y
X = tf.placeholder(tf.float32, shape = [None, 784])
y = tf.placeholder(tf.int32, shape = [None])
  
# 定義隨機森林的內部參數
hparams = tensor_forest.ForestHParams(num_classes = num_classes, num_features = num_features, num_trees = num_trees, max_nodes = max_nodes).fill()

# 創建隨機森林運算圖
forest_graph = tensor_forest.RandomForestGraphs(hparams)

# 定義訓練運算圖和 loss
train_op = forest_graph.training_graph(X, y)
loss_op = forest_graph.training_loss(X, y)
 
# 計算準確率
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))

# 定義圖的初始化內容
init_var = tf.group(tf.global_variables_initializer(), resources.initialize_resources(resources.shared_resources()))

# 初始化
sess = tf.Session()
sess.run(init_var)


# 定義批次爲隨機 size 個訓練數據
def next_batch(X_data, y_data, size):
    rand = np.arange(y_data.shape[0])
    np.random.shuffle(rand)
    return X_data[rand[0 : size]], y_data[rand[0 : size]]


# 訓練過程
for i in range(1, num_steps + 1):
    batch = next_batch(X_train, y_train, batch_size)
    _, cost = sess.run([train_op, loss_op], feed_dict = {X: batch[0], y:batch[1]})
    
    if i % 100 == 0 or i == 1:
        train_acc = sess.run(accuracy_op, feed_dict={X: batch[0], y: batch[1]})
        val_acc = sess.run(accuracy_op, feed_dict={X: X_val, y: y_val})
        print('Step %i, Loss: %f, Train_Acc: %s, Test_acc: %s' % (i, cost, train_acc, val_acc))
 
print("Test Accuracy:", sess.run(accuracy_op, feed_dict={X:X_test y:y_test}))