Common operations of numpy and sklearn

Some operations that some individuals usually encounter during data processing are often forgotten, and they are placed here for direct access to
sklearn official documents later.

numpy

Pearson coefficient: Compare the correlation of features with the target. (Used to measure the linear correlation between two variables X and Y, the value range is between -1 and 1)

import numpy as np
pccs = np.corrcoef(x, y)

Numpy array adds a column (row)

import numpy as np
x = np.array([1,2],[3,4])
# 行
row = np.array([5,6])
x = np.row_stack((x,row))
# 列
column = np.array([3,5,7])
y = np.column_stack((x,column))

cosine similarity

from sklearn.metrics.pairwise import cosine_similarity

tmp = []
tmp.append(df.iloc[:,-1])
tmp.append(df.iloc[:,4])# wdir
tmp.append(df.iloc[:,-6])#ndir
print("wdir:%.3f  ndir:%.3f"%(cosine_similarity(tmp)[0][1],cosine_similarity(tmp)[0][2]))

Get the top N largest indices in the list

## 获取numn数组中的N个最大值的索引？ 使用numpy中的argsort函数
import numpy as np

arr = np.array([1, 3, 2, 4, 5])
arr.argsort()[-3:][::-1]

# 返回结果
# Out[3]: array([4, 3, 1])

Random forest for feature importance ranking

# 引入库
from sklearn.ensemble import RandomForestClassifier
import numpy as np

forest = RandomForestClassifier(n_estimators=50)
forest.fit(x_train, y_train)   # x_train, y_train自定义
importances = forest.feature_importances_
indices = np.argsort(importances)[::-1] # 下标排序
print(indices) #打印结果

Np.argmax is cleverly applied after multi-classification softmax output

res = np.argmax(onehot, axis=-1)

learned

split dataset

# 随机选择数据(二八分)
seed = 1234
from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(x, y, test_size=0.2,
                                                    random_state=seed, shuffle=True, stratify=y)# shuffle默认为True

PCA

from sklearn.decomposition import PCA

t_feature_num = 3
pca = PCA(n_components=t_feature_num)
df= pca.fit_transform(df)

shuffle the rows in df

from sklearn.utils import shuffle
df = shuffle(df)

Normalized

from sklearn.preprocessing import MinMaxScaler,StandardScaler
# 实例化一个转换器类
transfer = MinMaxScaler()
# 调用fit_transform
data_new = transfer.fit_transform(df)
# 将结果反归一化，还原
# data_old = transfer.inverse_transform(data_new)

Acc、Pre、Recall、F1、Roc

from sklearn.metrics import accuracy_score, roc_auc_score
from sklearn.metrics import precision_score,recall_score
from sklearn.metrics import f1_score
print("acc:",accuracy_score(y_test, y_predict))
print("pre:",precision_score(y_test, y_predict))#多分类加 average='micro'
print("recall:",recall_score(y_test, y_predict))
print("f1:",f1_score(y_test, y_predict))
print("roc:",roc_auc_score(y_test, y_predict))

Use of MSE, MAE, RMSE, R2

# 衡量线性回归的MSE 、 RMSE、 MAE、r2
from math import sqrt
from sklearn.metrics import mean_absolute_error
from sklearn.metrics import mean_squared_error
from sklearn.metrics import r2_score
print("mean_absolute_error:", mean_absolute_error(y_test, y_predict))
print("mean_squared_error:", mean_squared_error(y_test, y_predict))
print("rmse:", sqrt(mean_squared_error(y_test, y_predict)))
print("r2 score:", r2_score(y_test, y_predict))

View all model evaluation metrics in sklearn
(Can be used for parameter scoring in RandomizedSearchCV or GridSearchCV)

import sklearn
sklearn.metrics.SCORERS.keys()