来源：15 Python Snippets to Optimize your Data Science Pipeline

翻译：RankFan

15种Python片段去优化你的数据科学管道

为什么片段对于数据科学是重要的

在我的日常中，我经常处理许多同样的状况，主要是从加载 csv 文件到数据可视化。因此，为了流水线这个过程，我有兴趣去储存一些 code 片段, 在不同的情形下，加载csv文件到数据可视化是非常有帮助的。

在这篇短文中，我将分享15个Python片段去简化你不同的数据分析管道。

1. 通过 GLob 和 List 加载多个文件

import glob
import pandas as pd
csv_files = glob.glob("path/to/folder/with/csvs/*.csv")
dfs = [pd.read_csv(filename) for filename in csv_flies]

2. 得到列中的唯一值

import pandas as pd
df = pd.read_csv("path/to/csv/file.csv")
df = ["Item_Identifier"].unique()

array['FDA15', 'DRC01', 'FDN15', ..., 'NCF55', 'NCW30', 'NCW05'],dtype = object]

3. 并排展示Pandas Dataframe

from IPython.display import display_html
from itertools import chain, cycle

def display_side_by_side(*arg, title = cycle([''])):
    html_str = ""
    for df, title in zip(args, chain(title, cycle(['</br>']))):
        html_str += '< the style = " text-align : center "> < td style = "vertical-align : top">'
        html_str += "<br>"
        html_str += f'<h2>{title}</h2>'
        html_str += df.to_html().replace('table', tabel style="display:inline")
        html_str += '</td></th>'
        
    display_html(html_str, raw = True)
    
df1 = pd.csv_read("file_csv")
df2 = pd.csv_read("file2")

dispaly_side_by_side(df1.head(), df2.head(), titles=[Sales, Advertising])
        

4. 移除Pandas DataFrame中的缺失值

df = pd.DataFrame(dict(a = [1, 2, 3, None]))
df
df.dropna(inplace = True)
df

5. 显示缺失值的个数

def FindNanCol(df):
    for col in df:
        print(f"Column : {col}")
        num_Nans = df[col].isnull().sum()
        print(f"Number of Nans : {num_Nans}")
        
df = pd.DataFrame(dict(a = [1, 2, 3, None], b = [None, None, 5, 6]))
FindNanCol(df)

6. 使用.apply 函数和 lambda 函数 转变列

df = pd.DataFrame(dict(a = [10, 20 ,30, 40, 50]))
square = lambda x: x**2
df["a"] = df["a"].apply(square)
df

7. 将两个DataFrame列转化为字典

df = pd.DataFrame(dict(a = ["a", "b", "c"], b = [1, 2, 3]))
df_dictionary = dict(zip(df["a"], df["b"]))
df_dictionary

8. 绘制列的网格分布

import numpy as np
import matplotlib.pyplot as plt
import seaborns as sns
import pandas as pd
sns.set()

df = pd.DataFrame(dict(a = np.random.randint(0, 100, 100), b = np.arange(0, 100, 1)))
plt.figure(figsize = (15,7))
plt.subplot(1, 2, 1)
df["b"][df["a"]>50].hist(color='green', label="bigger than 50")
plt.legend()

plt.subplot(1, 2, 1)
df["b"][df["a"]<50].hist(color='orange', label="small than 50")
plt.legend()
plt.show

9. 在pandas中对不同的列进行t检验

from scipy.stats import ttest_rel

data = np.arange(0, 1000, 1)
data_plus_noise = np.arange(0, 1000, 1) + np.random.normal(0, 1, 1000)
df = pd.DataFrame(dict(data = data, data_plus_noise = data_plus_noise))
print(ttest_rel(df["data"], df["data_plus_noise"]))

10. 合并数据

df1 = pd.DataFrame(dict(a = [1, 2, 3], b=[10, 20, 30], col_to_merge= ["a", "b", "c"]))
df2 = pd.DataFrame(dict(d = [10, 20, 30], col_to_merge=["a", "b", "c"]))
df_merged = df1.merge(df2, on='col_to_merge')

11. 用sklearn进行标准化

from sklearn.preprocessing import MinMaxScaler
scaler = MinMaxScaler()
scores = scaler.fit_transform(df["a"].values.reshape(-1, 1))

12. 丢弃特定列的缺失值

df.dropna(subset = ["col_to_remove_NaNs_from"], inplace = True)

13. 有条件的选择dataframe的子集

df = pd.Dataframe(dict(result = ["pass", "Fail", "pass", "Fail", "Distinction", "Distinction"]))
pass_index = (df["result"] == "pass") | (df["result"] == "Distinction")
df_pass = df['pass_index']
df_pass

14. 饼图

import matplotlib.pyplot as plt
df = pd.DataFrame(dict(a = [10, 20, 50, 10, 10], b=["A", "B", "C", "D", "E"]))
labels = df["b"]
sizes = df["a"]
plt.pie(sizes, labels = labels, autopct = '%1.1f%%', shadow = True, startangle=140)
plt.axis('equal')
plt.show

15. 将百分数字符串转化为数值

def change_to_numerical(x):
    try:
        x = int(x.strip("%")[:2])
    except:
        x = int(x.strip("%")[:1])
    return x

df = pd.DataFrame(dict(a =["A", "B" ,"C"], col_with_percentage = ["10%", "20%", "70%"]))
df["col_with_percentage"] = df["col_with_percentage"].apply(change_to_numerical)
df

结论

我认为代码片段是非常有用，重新写是浪费时间的，因此，有一个完整的工具包可以对数据分析进行流水线处理，这是非常有帮助的。