如何控制特征的数量【机器学习】？

Question

我正在编写此机器学习代码（分类）以在两个类之间进行分类。我开始为我的所有图像捕获一个功能。

例如： （注：1 & 0 用于标记） A类=[(4295046.0, 1), (4998220.0, 1), (4565017.0, 1), (4078291.0, 1), (4350411.0, 1), (4434050.0, 1), (4201831.0, 1), (4203570.0, 1 ), (4197025.0, 1), (4110781.0, 1), (4080568.0, 1), (4276499.0, 1), (4363551.0, 1), (4241573.0, 1), (4455070.0, 1), (5682823.0, 1), (5572122.0, 1), (5382890.0, 1), (5217487.0, 1), (4714908.0, 1), (4697137.0, 1), (4057898.0, 1), (4143981.0, 1), (3899129.0, 1), (383058 , 1), (3557377.0, 1), (3125518.0, 1), (3197039.0, 1), (3109404.0, 1), (3024219.0, 1), (3066759.0, 1), (2726363.0, 1), (3507626.0, 1 ), .....等]

B类=[(7179088.0, 0), (7144249.0, 0), (6806806.0, 0), (5080876.0, 0), (5170390.0, 0), (5694876.0, 0), (6210510.0, 0), ( 5376014.0, 0), (6472171.0, 0), (7112956.0, 0), (7356507.0, 0), (9180030.0, 0), (9183460.0, 0), (9212517.0, 0), (9055663.0, 0.9, 0), (90530) 0), (9103067.0, 0), (8889903.0, 0), (8328604.0, 0), (8475442.0, 0), (8499221.0, 0), (8752169.0, 0), (8779133.0, 0), (8756789.0, 0) , (8990732.0, 0), (9027381.0, 0), (9090035.0, 0), (9343846.0, 0), (9518609.0, 0), (9435149.0, 0), (9365842.0, 0), (9395256.0, 0), ( 4381880.0, 0), (4749338.0, 0), (5296143.0, 0), (5478942.0, 0), (5610865.0, 0), (5514997.0, 0), (5381010.0, 0), (5090416.0, 0), (46635) 0), (4804526.0, 0), (4743107.0, 0), (4898914.0, 0), (5018503.0, 0), (5778240.0, 0), (5741893.0, 0), (4632926.0, 0), (5208486.0, 0) , (5633403.0, 0), (5699410.0, 0), (5748260.0, 0), (5869260.0, 0), ....等]

/data is A and B combined

x = [[each[0]] for each in data]
y = [[each[1]] for each in data]
print (len(x), len(y))

x_train, x_test, y_train, y_test = train_test_split(x, y, test_size=0.2, 
random_state=42)
print (len(x_train), len(x_test))
print (len(y_train), len(y_test))

from sklearn.ensemble import RandomForestClassifier

clf = RandomForestClassifier(n_estimators=100, max_depth=2, random_state=0)
clf.fit(x_train, y_train)

问题：

要更改什么以添加其他功能？添加功能时 A 和 B 的外观如何，我是否更改此行

clf = RandomForestClassifier(n_estimators=100, max_depth=2, random_state=0)

何时使用两个功能？

我的猜测：

class A=[(4295046.0,secons features, 1), (4998220.0,secons features, 1), (4565017.0,secons features, 1), (4078291.0,secons features, 1), (4350411.0,secons features, 1 ), (4434050.0, 1),......] 那正确吗？有没有更好的办法？

Answer 1

此模型不需要明确的特征数量。
如果类始终是数据中每个元组中的最后一个元素，您可以这样做：

x = [[each[:-1]] for each in data]
y = [[each[-1]] for each in data]

然后从那里继续。

Answer 2

随机森林的想法是你有很多简单的模型，你可以平均下来。这意味着无论你有多少特征，你的树都不应该太深。如果你有很多特征，并且使用了很多树，你可以尝试增加深度，但一般来说，对于随机森林，树应该是浅的。尝试一下！

举个例子：

https://medium.com/all-things-ai/in-depth-parameter-tuning-for-random-forest-d67bb7e920d

在这个实验中，有 +900 个数据点和 9 个特征。他们测试了 1 到 32 之间的 max_depth 值，从结果来看，我认为 5 左右是最好的。但这可能会有所不同，具体取决于相关数据集和功能。