DataFrame和python中數據結構互相轉換

楔子

有時候DataFrame，我們不一定要保存成文件、或者入數據庫，而是希望保存成其它的格式，比如字典、列表、json等等。當然，讀取DataFrame也不一定非要從文件、或者數據庫，根據現有的數據生成DataFrame也是可以的，那么該怎么做呢？我們來看一下

DataFrame轉成python中的數據格式

轉成json

DataFrame轉成json，可以使用df.to_json()方法

import pandas as pd

df = pd.DataFrame({"name": ["mashiro", "satori", "koishi", "nagisa"],
                   "age": [17, 17, 16, 21]})

print(df.to_json())  
# {"name":{"0":"mashiro","1":"satori","2":"koishi","3":"nagisa"},"age":{"0":17,"1":17,"2":16,"3":21}}

我們看到雖然轉化成了json，但是有些不完美，那就是它把索引也算進去了

import pandas as pd

df = pd.DataFrame({"name": ["mashiro", "satori", "koishi", "nagisa"],
                   "age": [17, 17, 16, 21]})

# 如果不想加索引的話，那么指定index=False即可
try:
    print(df.to_json(index=False))
except Exception as e:
    print(e)  # 'index=False' is only valid when 'orient' is 'split' or 'table'
# 但是它報錯了，說如果index=False，那么orient必須指定我split或者table

我們看一下這個orient是什么

首先orient可以有如下取值：split、records、index、columns、values、table

我們分別演示一下，看看orient取不同的值，結果會有什么變化

• orient='split'

import pandas as pd

df = pd.DataFrame({"name": ["mashiro", "satori", "koishi", "nagisa"],
                   "age": [17, 17, 16, 21]})

print(df.to_json(orient="split"))
"""
{
 "columns":["name","age"],
 "index":[0,1,2,3],
 "data":[["mashiro",17],["satori",17],["koishi",16],["nagisa",21]]
}
"""
print(df.to_json(orient="split", index=False))
"""
{
 "columns":["name","age"],
 "data":[["mashiro",17],["satori",17],["koishi",16],["nagisa",21]]
}
"""

我們看到會變成三個鍵值對，分別是列名、索引、數據

• orient='records'

import pandas as pd

df = pd.DataFrame({"name": ["mashiro", "satori", "koishi", "nagisa"],
                   "age": [17, 17, 16, 21]})

print(df.to_json(orient="records"))
"""
[{"name":"mashiro","age":17},
 {"name":"satori","age":17},
 {"name":"koishi","age":16},
 {"name":"nagisa","age":21}]
"""

這種格式的數據是比較常用的，相當於列名和每一行數據組合成一個字典，然后存在一個列表里面。並且我們看到生成json默認跟索引沒啥關系，所以不需要、也不可以加index=False

• orient='index'

import pandas as pd

df = pd.DataFrame({"name": ["mashiro", "satori", "koishi", "nagisa"],
                   "age": [17, 17, 16, 21]})

print(df.to_json(orient="index"))
"""
{
 "0":{"name":"mashiro","age":17},
 "1":{"name":"satori","age":17},
 "2":{"name":"koishi","age":16},
 "3":{"name":"nagisa","age":21}
}
"""

類似於records，只不過這里把字典作為value放在了外層字典里，其中key為對應的索引。當然這里同樣不可以加index=False

• orient='columns'

import pandas as pd

df = pd.DataFrame({"name": ["mashiro", "satori", "koishi", "nagisa"],
                   "age": [17, 17, 16, 21]})

print(df.to_json(orient="columns"))
"""
{"name":{"0":"mashiro","1":"satori","2":"koishi","3":"nagisa"},"age":{"0":17,"1":17,"2":16,"3":21}}
"""

我們看到這個和不指定orient得到結果是一樣的，其實不指定的話orient默認是columns

• orient=values

import pandas as pd

df = pd.DataFrame({"name": ["mashiro", "satori", "koishi", "nagisa"],
                   "age": [17, 17, 16, 21]})

print(df.to_json(orient="values"))
"""
[["mashiro",17],["satori",17],["koishi",16],["nagisa",21]]
"""
# 我們看到當orient指定為values，會只獲取數據
# 另外這個方式類似於to_numpy
print(df.to_numpy())
"""
[['mashiro' 17]
 ['satori' 17]
 ['koishi' 16]
 ['nagisa' 21]]
"""

• orient=table

import pandas as pd

df = pd.DataFrame({"name": ["mashiro", "satori", "koishi", "nagisa"],
                   "age": [17, 17, 16, 21]})

# 以數據庫二維表的形式返回
print(df.to_json(orient="table"))
"""
{
    "schema": {
        "fields": [{"name": "index", "type": "integer"},
                   {"name": "name", "type": "string"},
                   {"name": "age", "type": "integer"}],
        "primaryKey": ["index"],
        "pandas_version": "0.20.0"
    },
    "data": [{"index": 0, "name": "mashiro", "age": 17},
             {"index": 1, "name": "satori", "age": 17},
             {"index": 2, "name": "koishi", "age": 16},
             {"index": 3, "name": "nagisa", "age": 21}]
}
"""
print(df.to_json(orient="table", index=False))
"""
{
    "schema": {
        "fields": [{"name": "name", "type": "string"},
                   {"name": "age", "type": "integer"}],
        "pandas_version": "0.20.0"
    },
    "data": [{"name": "mashiro", "age": 17},
             {"name": "satori", "age": 17},
             {"name": "koishi", "age": 16},
             {"name": "nagisa", "age": 21}]
}
"""

轉成dict

DataFrame也可以轉成字典，轉換成字典里面也有一個orient參數，里面有一部分和to_json是類似的。因為json這個數據結構本身就借鑒了python中的字典，是的你沒有看錯，json這種數據結構參考了python中的字典。

to_dict中的orient可以有如下取值：dict、list、series、split、records、index，默認是dict

• orient='dict'

from pprint import pprint
import pandas as pd

df = pd.DataFrame({"name": ["mashiro", "satori", "koishi", "nagisa"],
                   "age": [17, 17, 16, 21]})

pprint(df.to_dict(orient="dict"))
"""
{'age': {0: 17, 1: 17, 2: 16, 3: 21},
 'name': {0: 'mashiro', 1: 'satori', 2: 'koishi', 3: 'nagisa'}}
"""

• orient='list'

from pprint import pprint
import pandas as pd

df = pd.DataFrame({"name": ["mashiro", "satori", "koishi", "nagisa"],
                   "age": [17, 17, 16, 21]})

pprint(df.to_dict(orient="list"))
"""
{'age': [17, 17, 16, 21], 'name': ['mashiro', 'satori', 'koishi', 'nagisa']}
"""

• orient='series'

from pprint import pprint
import pandas as pd

df = pd.DataFrame({"name": ["mashiro", "satori", "koishi", "nagisa"],
                   "age": [17, 17, 16, 21]})

# 這種結構真的不常用，就是一個key對應一個series
pprint(df.to_dict(orient="series"))
"""
{'age': 
0    17
1    17
2    16
3    21
Name: age, dtype: int64,

'name': 0    mashiro
1     satori
2     koishi
3     nagisa
Name: name, dtype: object}
"""

• orient='split'

from pprint import pprint
import pandas as pd

df = pd.DataFrame({"name": ["mashiro", "satori", "koishi", "nagisa"],
                   "age": [17, 17, 16, 21]})

pprint(df.to_dict(orient="split"))
"""
{'columns': ['name', 'age'],
 'data': [['mashiro', 17], ['satori', 17], ['koishi', 16], ['nagisa', 21]],
 'index': [0, 1, 2, 3]}
"""

• orient='records'

from pprint import pprint
import pandas as pd

df = pd.DataFrame({"name": ["mashiro", "satori", "koishi", "nagisa"],
                   "age": [17, 17, 16, 21]})

pprint(df.to_dict(orient="records"))
"""
[{'age': 17, 'name': 'mashiro'},
 {'age': 17, 'name': 'satori'},
 {'age': 16, 'name': 'koishi'},
 {'age': 21, 'name': 'nagisa'}]
"""

• orient='index'

from pprint import pprint
import pandas as pd

df = pd.DataFrame({"name": ["mashiro", "satori", "koishi", "nagisa"],
                   "age": [17, 17, 16, 21]})

pprint(df.to_dict(orient="index"))
"""
{0: {'age': 17, 'name': 'mashiro'},
 1: {'age': 17, 'name': 'satori'},
 2: {'age': 16, 'name': 'koishi'},
 3: {'age': 21, 'name': 'nagisa'}}
"""

python中的數據格式轉成DataFrame

字典轉成DataFrame

import pandas as pd

data = {0: {'age': 17, 'name': 'mashiro'},
        1: {'age': 17, 'name': 'satori'},
        2: {'age': 16, 'name': 'koishi'},
        3: {'age': 21, 'name': 'nagisa'}}

df = pd.DataFrame.from_dict(data)
# 顯然不是我們期待的格式
print(df)
"""
            0       1       2       3
age        17      17      16      21
name  mashiro  satori  koishi  nagisa
"""

df = pd.DataFrame.from_dict(data, orient="index")
print(df)
"""
   age     name
0   17  mashiro
1   17   satori
2   16   koishi
3   21   nagisa
"""

所以df.to_dict和pd.DataFrame.from_json實現的是相反的功能，但是from_dict中的orient參數只有兩種選擇，要么是index，要么是columns，默認是columns

from_records

from_records是專門針對外層是列表的數據

import pandas as pd

data = [{'age': 17, 'name': 'mashiro'},
        {'age': 17, 'name': 'satori'},
        {'age': 16, 'name': 'koishi'},
        {'age': 21, 'name': 'nagisa'}]

df = pd.DataFrame.from_records(data)
print(df)
"""
   age     name
0   17  mashiro
1   17   satori
2   16   koishi
3   21   nagisa
"""

其實這種數據就是to_dict(orient="records")生成的