机器学习入门

使用Pandas打开csv数据

Pandas中数据是DataFrame，保存的数据类似一个Excel的sheet，或者SQL中的一个table

1
2
3
4
5
6
import pandas as pd
# save filepath to variable for easier access
melbourne_file_path = '../input/melbourne-housing-snapshot/melb_data.csv'
# read the data and store data in DataFrame titled melbourne_data
melbourne_data = pd.read_csv(melbourne_file_path)

了解数据

1
2
3
4
5
# print a summary of the data in Melbourne data
melbourne_data.describe()

# 显示所有列索引
print(melbourne_data.columns)

设定Target

1
y = melbourne_data['SalePrice']

设定Input

在所有列索引中选择需要的作为输入特征，如

1
2
3
4
5
feature_names = ['LotArea','YearBuilt','1stFlrSF','2ndFlrSF','FullBath','BedroomAbvGr','TotRmsAbvGrd']
X = melbourne_data[feature_names]
# 观察输入数据
print(X.describe())

数据划分 train_X、train_y、val_X、val_y

1
2
from sklearn.model_selection import train_test_split
train_X, train_y, val_X, val_y = train_test_split(X,y,random_state=1)

配置和拟合Model

以DecisionTreeRegressor为例

1
2
3
from sklearn.tree import DecisionTreeRegressor
melbourne_model = DecisionTreeRegressor(random_state=1)
melbourne_model.fit(train_X, train_y)

做预测

输入X为例

1
predictions = melbourne_model.predict(val_X)

模型验证

以Mean Absolute Error（MAE）为例

1
2
from sklearn.metrics import mean_absolute_error
val_mae = mean_absolute_error(melbourne_model.predict(val_X), val_y)

调参时候的操作方法

以寻找最大叶节点为例

1
2
3
4
5
6
7
8
9
10
11
12
13
def get_mae(max_leaf_nodes, train_X, val_X, train_y, val_y):
    model = DecisionTreeRegressor(max_leaf_nodes=max_leaf_nodes, random_state=0)
    model.fit(train_X, train_y)
    preds_val = model.predict(val_X)
    mae = mean_absolute_error(val_y, preds_val)
    return(mae)

# 使用不同的最大叶节点
candidate_max_leaf_nodes = [5, 25, 50, 100, 250, 500]
# Store the best value of max_leaf_nodes (it will be either 5, 25, 50, 100, 250 or 500)
maes = [get_mae(mln,train_X, val_X, train_y, val_y) for mln in candidate_max_leaf_nodes]
best_tree_size = candidate_max_leaf_nodes[maes.index(min(maes))]
print(best_tree_size)

使用全部data拟合Model

1
2
3
4
# 基于前述最佳参数建立模型，如最大叶节点100
final_model = DecisionTreeRegressor(max_leaf_nodes=100)
# 使用所有数据参与训练
final_model.fit(X,y)

使用其他算法

随机森林Random Forest 是比决策树更容易得到好结果的模型

1
2
3
4
5
6
7
8
9
10
11
12
13
from sklearn.ensemble import RandomForestRegressor
from sklearn.metric import mean_absolute_error

# Define the model. Set random_state to 1
rf_model = RandomForestRegressor(random_state=1)

# fit your model
rf_model.fit(train_X, train_y)

# Calculate the mean absolute error of your Random Forest model on the validation data
rf_val_mae = mean_absolute_error(rf_model.predict(val_X),val_y)

print("Validation MAE for Random Forest Model: {}".format(rf_val_mae))

在kaggle打比赛

首先需要加入比赛join competition
然后运行后在kaggle/working中生成submission.csv
点击submit即可

继续折腾

试验是最好的手段，尝试选用不同的features