1

scoop install anaconda3

1

conda create -n learn-scikit

1

conda activate learn-scikit

1

conda deactive

1

conda remove -n learn-scikit --all

1

conda env list

1
2
3
4
5
6
7
8

{
	// Make changes here to the cmd.exe profile.
	"guid": "{0caa0dad-35be-5f56-a8ff-afceee452369}",
	"name": "Anaconda",
	"icon": "%USERPROFILE%\\scoop\\apps\\anaconda3\\current\\Menu\\anaconda-navigator.ico",
	"commandline": "%windir%\\System32\\WindowsPowerShell\\v1.0\\powershell.exe -ExecutionPolicy ByPass -NoExit -Command \"& 'C:\\Users\\LittleGhost\\scoop\\apps\\anaconda3\\2020.02\\shell\\condabin\\conda-hook.ps1'\"",
	"hidden": false
}

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49

from sklearn.model_selection import train_test_split
from sklearn.tree import DecisionTreeClassifier
from sklearn.metrics import accuracy_score
import pandas as pd

data = pd.read_csv('in.csv')
# 读取数据
ipdata = data.copy()
# 拷贝数据

ipdata.dropna(axis=0, how='any', inplace=True)
# 当某一行有空值时，将该行数据删除
# 当然，可以选择其他处理空值的方法，例如使用平均值或者中值等数据进行填充

y = ipdata['type']
# 可以将数据集中的最后一列看作y轴上的变量，其余值当作X轴上的变量，因此此处设置数据中`type`这一列为y
# 等下进行X的设置

single_unique_cols = [col for col in ipdata.columns if ipdata[col].nunique() == 1]
# 找出只有一种值的那一列

ipdata.drop(single_unique_cols, axis=1, inplace=True)
# 丢掉只有一种值的那几列，因为这几列作为特征没有意义，因此丢掉
ipdata.drop('type', axis=1, inplace=True)
# 丢掉y轴上的变量，此时X轴上的变量即为ipdate

X_train, X_test, y_train, y_test = train_test_split(ipdata.values, y, test_size=0.3, random_state=17)
# 使用ipdata.values, y作为X和y
# test_size=0.3表示为：将数据集中30%作为测试集，其余为训练集
# random_state=17表示为：打乱数据集的程度


# 以下以决策树为例

tree = DecisionTreeClassifier(max_depth=11, random_state=17)
# 导入决策树模型
# max_depth=11表示为：决策树最大深度为11
# random_state=17表示为：分枝中的随机模式的参数，默认值None

tree.fit(X_train, y_train)
# 拟合训练数据

tree_pred = tree.predict(X_test)
# 使用已训练的模型对测试集进行预测

accuracy = accuracy_score(y_test, tree_pred)
# 将由模型出来的预测结果与测试机作比较，获得准确度

print('accuracy', accuracy)

1
2
3

train = pd.read_csv(train_csv_filepath, parse_dates=True)
test = pd.read_csv(test_csv_filepath, parse_dates=True)
# 其中parse_dates参数的作用是将csv中的时间字符串转换成日期格式，如果不涉及时间时可以不设置此参数

1
2
3
4
5
6
7

data = pd.read_csv(csv_filepath, parse_dates=True)

...
# 数据预处理

X_train, X_test, y_train, y_test = train_test_split(data.values, y, test_size=0.3, random_state=17)
# 区分数据集

1
2
3
4

from sklearn.metrics import confusion_matrix, precision_recall_curve, roc_curve, auc, log_loss
logreg = LogisticRegression()
logreg.fit(X_train, y_train)
y_pred = logreg.predict(X_test)

1
2
3
4
5
6

from sklearn.ensemble import RandomForestClassifier
rf_clf = RandomForestClassifier()
rf_clf.fit(X_train, y_train)
pred_rf = rf_clf.predict(X_test)
acc_rf = accuracy_score(y_test, pred_rf)
print(acc_rf)

1

print(data.head())

1

print(data.shape)

1

print(data.columns)

1

print(data.info())

1
2

non_num_cols = [col for col in data.columns if data[col].dtype == 'O']
non_num_data = data[non_num_cols]

1

num_cols = list(set(data.columns) - set(non_num_cols))

1

print([(col, non_num_data[col].nunique()) for col in non_num_cols])

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16

def summarize_cat(col_name):
    sorted_values = sorted(non_num_data[col_name].value_counts().iteritems(), key = lambda x:x[1], reverse=True)
    remaining_per = 100
    for (value, count) in sorted_values:
        per = count / len(non_num_data) * 100
        if per >= 1:
            print(f'{value} : {per:.2f}%')
        else :
            print(f'Others : {remaining_per:.2f}%')
            break
        remaining_per = remaining_per - per

for col in non_num_cols:
	print(f"Summary of {col} column : ")
	summarize_cat(col)
	print('\n')

1

print(data[num_cols].describe())

1

print([col for col in num_cols if data[col].isnull().any()])

1
2
3

print("range and no. of unique values in numeric columns")
for col in num_cols:
    print(f'{col}\tRange : {max(data[col]) - min(data[col])}, No. of unique values : {data[col].nunique()}')

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46

# range and no. of unique values in numeric columns
# 37      Range : 83331.97094, No. of unique values : 1201
# 42      Range : 6749637, No. of unique values : 978
# 11      Range : 3453236.71542, No. of unique values : 1129
# 30      Range : 3437.0, No. of unique values : 993
# 43      Range : 4947, No. of unique values : 151
# 40      Range : 8853, No. of unique values : 232
# 8       Range : 1087.0764488286068, No. of unique values : 1118
# 24      Range : 1281.3886792452831, No. of unique values : 1015
# 35      Range : 45.45751905441284, No. of unique values : 662
# 18      Range : 1452, No. of unique values : 530
# type    Range : 1, No. of unique values : 2
# 4       Range : 33.33312187194824, No. of unique values : 1490
# 28      Range : 369042, No. of unique values : 961
# 38      Range : 3442.1111111111113, No. of unique values : 988
# 0       Range : 1093.2091245376077, No. of unique values : 1112
# 41      Range : 13725520.2, No. of unique values : 1014
# 9       Range : 0, No. of unique values : 1
# 32      Range : 8847, No. of unique values : 244
# 15      Range : 3134, No. of unique values : 100
# 7       Range : 16658.35719, No. of unique values : 1342
# 22      Range : 297.29929924011225, No. of unique values : 1489
# 1       Range : 12, No. of unique values : 3
# 20      Range : 42.96079697779247, No. of unique values : 1490
# 27      Range : 504810.25, No. of unique values : 1045
# 36      Range : 499.9957299232483, No. of unique values : 1455
# 14      Range : 9237, No. of unique values : 212
# 10      Range : 8853, No. of unique values : 273
# 12      Range : 7038191, No. of unique values : 1100
# 26      Range : 1460, No. of unique values : 536
# 17      Range : 12, No. of unique values : 3
# 2       Range : 8845, No. of unique values : 268
# 39      Range : 2, No. of unique values : 2
# 29      Range : 6151, No. of unique values : 144
# 16      Range : 1275.4185428798348, No. of unique values : 1024
# 13      Range : 3609.0353598594666, No. of unique values : 1488
# 3       Range : 3452641.65455, No. of unique values : 1131
# 25      Range : 0, No. of unique values : 1
# 34      Range : 166.66551174720132, No. of unique values : 1457
# 6       Range : 499.88302969932556, No. of unique values : 1490
# 19      Range : 502931.58268, No. of unique values : 1051
# 33      Range : 13720819.8, No. of unique values : 1014
# 21      Range : 0.3386540412902832, No. of unique values : 351
# 5       Range : 0.0003230571746826172, No. of unique values : 92
# 23      Range : 6732.37035, No. of unique values : 1313
# 31      Range : 20, No. of unique values : 6

1
2

cols_for_hist = [col for col in num_cols if data[col].nunique() <= 50]
print(cols_for_hist, len(cols_for_hist))

1

# ['type', '39', '17', '31', '25', '1', '9'] 7

1
2

cols_for_desc = [col for col in num_cols if data[col].nunique() > 50]
print(cols_for_desc)

1

$ sudo apt install graphviz

1
2
3
4
5
6
7
8
9

from sklearn.tree import export_graphviz
import pydotplus
# pip install pydotplus

def tree_graph_to_png(tree, feature_names, png_file_to_save):
    # def tree_graph_to_png(tree, png_file_to_save):
    tree_str = export_graphviz(tree, feature_names=feature_names, filled=True, out_file=None)
    graph = pydotplus.graph_from_dot_data(tree_str)
    graph.write_png(png_file_to_save)

1

tree_graph_to_png(tree=tree, feature_names=ipdata.columns, png_file_to_save='topic3_decision_tree3.png')