# Similarities and differences of five pandas combinatorial functions

2022-01-30 02:33:18

## 1. explain

In the daily processing of data , Often encounter different dataframe The connection of 、 Combination and other operations , At the beginning , It's going to be a little tricky , After all, the following functions are similar , Easy to confuse .

concat

join merge combine append

Let's explain with the simplest example 、 Distinguish the role of several functions .

``````>>> import pandas as pd

>>> df0 = pd.DataFrame({"a": [1, 2, 3], "b": [4, 5, 6]})
>>> df0
a  b
0  1  4
1  2  5
2  3  6

>>> df1 = pd.DataFrame({"c": [2, 3, 4], "d": [5, 6, 7]})
>>> df1
c  d
0  2  5
1  3  6
2  4  7
Copy code ``````

## 2. concat

concat, full pinyin “concatenation”,` Allow horizontal or vertical side-by-side ` Combined data .

### 2.1 Line splicing

When merging data with the same column （2 individual df Keep the column names consistent , Line splicing ）, You can specify the axis as 0（ The default value is ） To call .

When `axis=0` It should be understood as , Splice in the direction of the column （ Splice down ）, That is to splice them in rows .

The column names must be consistent , What you get is what you want ,`df1.rename(columns={"c": "a", "d": "b"})`.

``````>>> pd.concat([df0, df1.rename(columns={"c": "a", "d": "b"})]
, axis=0)

a  b
0  1  4
1  2  5
2  3  6
0  2  5
1  3  6
2  4  7

Copy code ``````

If the column names are inconsistent , You will not get the desired row splicing result .

``````In : pd.concat([df0,df1],axis=0)
Out:
a    b    c    d
0  1.0  4.0  NaN  NaN
1  2.0  5.0  NaN  NaN
2  3.0  6.0  NaN  NaN
0  NaN  NaN  2.0  5.0
1  NaN  NaN  3.0  6.0
2  NaN  NaN  4.0  7.0
Copy code ``````

Summary ：

Line splicing , If no extra columns are generated , Note that the column names are consistent

### 2.2 Column splicing

Splicing by columns ,`axis=1` It can be understood as , Splice in the direction of the row （ Splice to the right ）, That is to splice the columns .

``````>>> pd.concat([df0, df1], axis=1)
a  b  c  d
0  1  4  2  5
1  2  5  3  6
2  3  6  4  7
Copy code ``````

By default , When horizontally combining data （ Along the column ） when , Will try to use the index . When they are not the same , Will see NaN Fill in non overlapping data , As shown below ：

``````>>> df2 = df1.copy()
>>> df2.index = [1, 2, 3]
>>> pd.concat([df0, df2], axis=1)
a    b    c    d
0  1.0  4.0  NaN  NaN
1  2.0  5.0  2.0  5.0
2  3.0  6.0  3.0  6.0
3  NaN  NaN  4.0  7.0
Copy code ``````

If you want to unify the index for splicing , You have to reset their indexes first ：

``````>>> pd.concat([df0.reset_index(drop=True),
df2.reset_index(drop=True)], axis=1)
a  b  c  d
0  1  4  2  5
1  2  5  3  6
2  3  6  4  7
Copy code ``````

Summary ：

Column splicing , If no extra rows are generated , Note that the index is consistent

## 3. join

And concat comparison ,join Dedicated to ` Use index join DataFrame Columns between objects `.

df0,df1 The index of is consistent ：

``````>>> df0.join(df1)
a  b  c  d
0  1  4  2  5
1  2  5  3  6
2  3  6  4  7
Copy code ``````

When the index is inconsistent , The connection is left on the left by default DataFrame The line of （ The default left table is the drive table ）; If the right side DataFrame There is no left side in the DataFrame Rows that match the index in , On the right side DataFrame Be deleted 、 use Null fill , As shown below ：

``````>>> df0.join(df2)
a  b    c    d
0  1  4  NaN  NaN
1  2  5  2.0  5.0
2  3  6  3.0  6.0
Copy code ``````

You can also set how Parameter to change the drive table , That is to say SQL Several association connections in .

``````# "right" uses df2’s index
>>> df0.join(df2, how="right")
a    b  c  d
1  2.0  5.0  2  5
2  3.0  6.0  3  6
3  NaN  NaN  4  7# "outer" uses the union
>>> df0.join(df2, how="outer")
a    b    c    d
0  1.0  4.0  NaN  NaN
1  2.0  5.0  2.0  5.0
2  3.0  6.0  3.0  6.0
3  NaN  NaN  4.0  7.0# "inner" uses the intersection
>>> df0.join(df2, how="inner")
a  b  c  d
1  2  5  2  5
2  3  6  3  6
Copy code ``````

Summary ：

join Is an index based connection , Only column connections , And sql The association is similar to

## 4. merge

And join comparison ,merge More general , You can merge columns and indexes .

stay a Column ：

``````>>> df0.merge(df1.rename(columns={"c": "a"}),
on="a", how="inner")
a  b  d
0  2  5  5
1  3  6  6
Copy code ``````

If you want to keep the associated columns at the same time , You can write like this ：

``````>>> df0.merge(df1, left_on="a", right_on="c")
a  b  c  d
0  2  5  2  5
1  3  6  3  6
Copy code ``````

When two DataFrame Objects have the same columns , Instead of merging ,`suffixes` Parameter sets the suffix for renaming these columns ; By default , Left 、 The suffixes of the right data frame are “_x” and “_y”, You can also customize it .

``````>>> df0.merge(df1.rename(columns={"c": "a", "d": "b"}),
on="a",
how="outer",
suffixes=("_l", "_r"))

a  b_l  b_r
0  1  4.0  NaN
1  2  5.0  5.0
2  3  6.0  6.0
3  4  NaN  7.0
Copy code ``````

## 5. combine

combine Functions also act on 2 individual DataFrame Between objects , Combine by column , But it is very different from the above functions .

combine The special thing about the function is that it requires a ` Function parameter `. This function takes two series, Every series Corresponds to each DataFrame Merge columns in , And return a series The final value of the operation as an element of the same column .

It's a little twisted , Take an example ：

``````>>> def taking_larger_square(s1, s2):
...     return s1 * s1 if s1.sum() > s2.sum() else s2 * s2

>>> df0.combine(df1.rename(columns={"c": "a", "d": "b"}),
taking_larger_square)
a   b
0   4  25
1   9  36
2  16  49

Copy code ``````

take_larger_square Function pair df0 and df1 Medium a And df0 and df1 Medium b Column to operate . In two columns a And two columns b Between ,taking_larger_square Take the square of the value in the larger column . under these circumstances ,df1 Of a and b The column will be used as the square , Produce the final value , This happens to be df1 Of a、b Is greater than df0 Of a、b, If df1 One of the big 、 One small , Then take the largest one as the square .

``````In : df0,df1
Out:
(   a  b
0  1  2
1  2  3
2  3  4,
c  d
0  4  1
1  5  2
2  6  3)

In : df0.combine(df1.rename(columns={"c": "a", "d": "b"}),
taking_larger_square)
Out:
a   b
0  16   4
1  25   9
2  36  16

Copy code ``````

Summary ：

use combine Are combined , Mainly aimed at 2 individual DataFrame Object series Do function processing , Take one as the result .

## 6. append

append The function is dedicated to ` Append rows to existing DataFrame object `, Create a new object .

``````>>> df0.append(df1.rename(columns={"c": "a", "d": "b"}))

a  b
0  1  4
1  2  5
2  3  6
0  2  5
1  3  6
2  4  7
Copy code ``````

This sum concat( ,axis=0) The effect is the same .

append It is unique in that it can also add dict object , This gives us the flexibility to append different types of data . Be careful , Must be ignore_index Set to True, because dict Objects have no DataFrame Available index information .

``````>>> df0.append({"a": 1, "b": 2}, ignore_index=True)

a  b
0  1  4
1  2  5
2  3  6
3  1  2
Copy code ``````

## 7. summary

• concat： Combine data by row and column

• join： Use index , Merge data by row

• merge： Merge data by column , More like database connection operation

• combine： Merge data by column , With inter column （ Same column ） Element operation

• append： With DataFrame or dict Append data line by line in the form of object

Welcome to follow individual public number ：`Distinct Count ` 