[PYTHON] How to format a table using Pandas apply, pivot and swaplevel
Overview
Let's create a table like this with Pandas.
Input data
import pandas as pd
tokyo = pd.DataFrame(data={'year':['2013','2014','2015'],
'cnt':[10,15,20],
'price':[100,100,90],},
columns=['year','cnt','price'])
osaka = pd.DataFrame(data={'year':['2013','2014','2015'],
'cnt':[5,6,7],
'price':[100,100,100],},
columns=['year','cnt','price'])
yokohama = pd.DataFrame(data={'year':['2015'],
'cnt':[1],
'price':[50],},
columns=['year','cnt','price'])
result
tokyo
-----------------------
year cnt price
0 2013 10 100
1 2014 15 100
2 2015 20 90
osaka
-----------------------
year cnt price
0 2013 5 100
1 2014 6 100
2 2015 7 100
yokohama
-----------------------
year cnt price
0 2015 1 50
The sample data assumes sales of a certain product in Tokyo, Osaka, and Yokohama for three years. cnt is the number and price is the unit price. Yokohama has data for 2015 only.
Concatenate input data (concat)
Give each DataFrame a city name and join them vertically.
#Add city name to each DataFrame
tokyo['city'] = 'tokyo'
osaka['city'] = 'osaka'
yokohama['city'] = 'yokohama'
#Combine vertically
df = pd.concat([tokyo,osaka,yokohama], ignore_index=True)
result
df
-------------------------------------
year cnt price city
0 2013 10 100 tokyo
1 2014 15 100 tokyo
2 2015 20 90 tokyo
3 2013 5 100 osaka
4 2014 6 100 osaka
5 2015 7 100 osaka
6 2015 1 50 yokohama
Create a new column using the information of an existing column (apply)
With the apply method, you can perform uniform processing for all columns and all rows. In this example, axis = 1 is specified, so the lambda expression is called line by line.
** Calculate sales by multiplying the number (cnt) and unit price (price) **
#axis=In 1, apply function to each line. axis axis='columns'But the same effect.
#x.cnt * x.The price part is the number*unit price
df["sales"] = df.apply(lambda x:x.cnt * x.price,axis=1)
result
df
------------------------------------------------
year cnt price city sales
0 2013 10 100 tokyo 1000
1 2014 15 100 tokyo 1500
2 2015 20 90 tokyo 1800
3 2013 5 100 osaka 500
4 2014 6 100 osaka 600
5 2015 7 100 osaka 700
6 2015 1 50 yokohama 50
** Determine the Kanto / Kansai area from the city name **
#'west' if x.city == 'osaka' else 'east'Is a ternary operation.
df["area"] = df.apply(lambda x:'west' if x.city == 'osaka' else 'east',axis=1)
result
df
------------------------------------------------
year cnt price city sales area
0 2013 10 100 tokyo 1000 east
1 2014 15 100 tokyo 1500 east
2 2015 20 90 tokyo 1800 east
3 2013 5 100 osaka 500 west
4 2014 6 100 osaka 600 west
5 2015 7 100 osaka 700 west
6 2015 1 50 yokohama 50 east
Create a table of cities and years (pivot_table)
List the sales for each year in each city. Pandas automatically edits with pivot_table, but the arguments are complicated and you need to get used to it. .. This time, I referred to this page.
#Create a table of cnt and sales with area and city on the horizontal axis and year on the vertical axis.
#fill_value fills the missing part of the data with 0.
df2 = pd.pivot_table(df,
index=['area','city'],
columns=['year'],
values=['cnt','sales'],
fill_value=0)
result
df2
------------------------------------------------
cnt sales
year 2013 2014 2015 2013 2014 2015
area city
east tokyo 10 15 20 1000 1500 1800
yokohama 0 0 1 0 0 50
west osaka 5 6 7 500 600 700
The above is the output result of Python, but if you make it a little easier to understand, it looks like this.
| cnt | sales <Column 1st layer(No name)
|================|===================
year | 2013 2014 2015 | 2013 2014 2015 <Column 2nd layer(The name is year)
==============|================|===================
east|tokyo | 10 15 20 | 1000 1500 1800
|yokohama | 0 0 1 | 0 0 50
west|osaka | 5 6 7 | 500 600 700
==============|================|===================
area|city <<The Index also has two levels. The names are area and city.
Change the order of columns in Pivot results (swap level)
I'm pretty close to the output, but I want the columns to be in the order of "year"> "quantity, sales", so [swaplevel](http://pandas.pydata.org/pandas-docs/stable/generated/pandas. Edit with DataFrame.swaplevel.html). (If you know how to do it with pivot alone, please let me know.)
** Swap 2 rows in column **
#The first argument, 0, points to the 0th line of colum.
#The second argument, 1 points to the first line of colum.
df2=df2.swaplevel(0,1, axis=1)
Result (year has moved to the beginning)
df2
------------------------------------------------
year 2013 2014 2015 2013 2014 2015
cnt cnt cnt sales sales sales
area city
east tokyo 10 15 20 1000 1500 1800
yokohama 0 0 1 0 0 50
west osaka 5 6 7 500 600 700
** Sort by year **
df3=df2[['2013','2014','2015']]
result
df3
------------------------------------------------
year 2013 2014 2015
cnt sales cnt sales cnt sales
area city
east tokyo 10 1000 15 1500 20 1800
yokohama 0 0 0 0 1 50
west osaka 5 500 6 600 7 700
Output to excel
Now that you have created the output you want, output it to Excel.
#pip install openpyxl
writer = pd.ExcelWriter('output.xlsx')
df3.to_excel(writer,'Sheet1')
writer.save()
The rest is done by drawing ruled lines and coloring. It seems that it can be done with openpyxl, but this time I did it manually with Excel.
Recommended Posts