公衆號：尤而小屋
作者：Peter
編輯：Peter
大家好，我是Peter~
本文記錄的是如何使用Python、pandas、numpy、scikit-learn來實現隨機打亂、抽取和切割數據。主要的方法包含：
- sample
- shuffle
- np.random.permutation
- train_test_split
[圖片上傳失敗...(image-1161e4-1650638148355)]
導入數據
In [1]:
import pandas as pd
import numpy as np
import random # 隨機模塊
import plotly_express as px # 可視化庫
import plotly.graph_objects as go
內置數據
採用的是plotly庫中內置的一份消費數據集：
In [2]:
df = px.data.tips()
df.head()
[圖片上傳失敗...(image-876b5e-1650638148356)]
基本信息
In [3]:
df.shape
Out[3]:
(244, 7)
In [4]:
columns = df.columns
columns
Out[4]:
Index(['total_bill', 'tip', 'sex', 'smoker', 'day', 'time', 'size'], dtype='object')
sample實現
行方向
In [5]:
隨機抽取一行記錄：
df.sample() # 隨機抽取一行記錄
[圖片上傳失敗...(image-b2166a-1650638148356)]
隨機抽取多行數據：
[圖片上傳失敗...(image-eede73-1650638148356)]
通過參數frac實現按照比例隨機抽樣：
df.sample(frac=0.05)
[圖片上傳失敗...(image-ac8557-1650638148356)]
列方向
主要是選擇不同數量或者比例的屬性；整體的行數量是不變的
In [8]:
df.sample(3, axis=1) # 在列屬性上抽取
[圖片上傳失敗...(image-736f06-1650638148356)]
shuffle實現
scikit-Learn的shuffle
In [9]:
from sklearn.utils import shuffle
In [10]:
shuffle(df) # 打亂數據
[圖片上傳失敗...(image-533ea7-1650638148356)]
random模塊的shuffle
In [11]:
length = list(range(len(df))) # 原始的長度作爲索引
length[:5]
Out[11]:
[0, 1, 2, 3, 4]
In [12]:
random.shuffle(length) # 打亂索引
In [13]:
length[:5]
Out[13]:
[136, 35, 207, 127, 29] # 打亂後的結果
In [14]:
df.iloc[length] # 通過打亂後的索引獲取數據
[圖片上傳失敗...(image-2823ae-1650638148356)]
numpy實現
In [15]:
# 先打亂每個索引
np.random.permutation(len(df))
Out[15]:
array([223, 98, 238, 17, 101, 26, 122, 212, 27, 79, 210, 147, 176,
82, 164, 142, 141, 219, 6, 63, 185, 112, 158, 188, 242, 207,
45, 55, 178, 150, 217, 32, 16, 160, 157, 234, 95, 174, 93,
52, 57, 220, 216, 230, 35, 86, 125, 114, 100, 73, 83, 88,
34, 7, 40, 115, 97, 165, 84, 18, 197, 151, 135, 121, 72,
173, 228, 143, 227, 9, 183, 56, 23, 237, 136, 106, 133, 189,
139, 0, 208, 74, 166, 4, 68, 12, 71, 85, 172, 138, 149,
144, 232, 186, 99, 130, 41, 201, 204, 10, 167, 195, 66, 159,
213, 87, 103, 117, 31, 211, 190, 24, 243, 127, 48, 218, 233,
113, 81, 235, 229, 206, 96, 46, 222, 50, 156, 180, 214, 124,
240, 140, 89, 225, 2, 120, 58, 169, 193, 39, 102, 104, 148,
184, 170, 152, 153, 146, 179, 137, 129, 64, 3, 65, 128, 90,
110, 14, 226, 181, 131, 203, 221, 80, 51, 94, 231, 44, 108,
43, 145, 47, 75, 162, 163, 69, 126, 200, 1, 123, 37, 205,
111, 25, 91, 11, 42, 67, 118, 196, 161, 28, 116, 105, 33,
38, 78, 76, 224, 20, 202, 171, 177, 107, 8, 209, 239, 77,
241, 154, 5, 198, 92, 61, 182, 36, 70, 22, 54, 187, 175,
119, 215, 49, 134, 21, 60, 62, 168, 59, 155, 194, 109, 132,
19, 199, 29, 191, 13, 30, 192, 236, 15, 53])
In [16]:
# 通過打亂後的索引來選擇數據
df.iloc[np.random.permutation(len(df))]
[圖片上傳失敗...(image-75de24-1650638148356)]
train_test_split實現
from sklearn.model_selection import train_test_split
data = []
for i in train_test_split(df, test_size=0.2):
data.append(i)
In [18]:
第一份數據是80%的：
data[0] # 80%的數據
[圖片上傳失敗...(image-500fd4-1650638148356)]
剩餘的20%的數據：
[圖片上傳失敗...(image-7993c6-1650638148356)]