Pandas DataFrame to_dict() method converts Pandas DataFrame into a Dictionary. It provides the “orient” argument that determines the type of dictionary values.

When working with Web APIs, you often encounter JSON data, a string representation of lists and dictionaries.

Some machine learning algorithms might expect data in the form of dictionaries or lists of dictionaries instead of a DataFrame. That’s where we need the conversion.

Types of “Orient”

There are the following seven types of orientations:

Name	Value
“dict” (default)	It will create a nested dictionary, with outer keys being column names and inner dictionaries having index labels as keys and cell values.
“list”	It will create a dictionary in which keys are column names and values are lists of the values of those columns.
“Series”	It will create a dictionary in which keys are column names and values are Pandas Series objects containing column data.
“split”	It will create a dictionary with three key-value pairs.
“tight”	It has information about the index and column names if explicitly set.
“records”	It will create a list of dictionaries where each dictionary represents the row of DataFrame.
“index”	It will create a dictionary where keys are index labels and values are dictionaries representing each row.

orient=’dict’ (default)

If you want a nested-column-centric dictionary, you can use this orientation.

import pandas as pd

data = {'Name': ['Krunal', 'Ankit', 'Rushabh'],
        'Age': [32, 30, 33],
        'City': ['New York', 'London', 'Paris']
        }

df = pd.DataFrame(data)

df_to_dict = df.to_dict()

print(df_to_dict) # Default Dictionary

Output

{
 'Name': {0: 'Krunal', 1: 'Ankit', 2: 'Rushabh'}, 
 'Age': {0: 32, 1: 30, 2: 33}, 
 'City': {0: 'New York', 1: 'London', 2: 'Paris'}
}

The output is a nested dictionary, where each column name becomes a “key,” and the dictionary becomes a “value.“

orient=’list’

If you need a simple dictionary of columns and their values as lists from a DataFrame, you can use this approach.

import pandas as pd

data = {'Name': ['Krunal', 'Ankit', 'Rushabh'],
        'Age': [32, 30, 33],
        'City': ['New York', 'London', 'Paris']
        }

df = pd.DataFrame(data)

dict_with_list = df.to_dict(orient='list')

print(dict_with_list)

Output

{
  'Name': ['Krunal', 'Ankit', 'Rushabh'], 
  'Age': [32, 30, 33], 
  'City': ['New York', 'London', 'Paris']
}

orient=’records’

Another famous orientation is “records,” which is suitable for JSON serialization or iteration over rows. If you need a list of row-based dictionaries, you can use this option.

import pandas as pd

data = {'Name': ['Krunal', 'Ankit', 'Rushabh'],
        'Age': [32, 30, 33],
        'City': ['New York', 'London', 'Paris']
        }

df = pd.DataFrame(data)

dict_with_json = df.to_dict(orient='records')

print(dict_with_json)

Output

[
  {'Name': 'Krunal', 'Age': 32, 'City': 'New York'}, 
  {'Name': 'Ankit', 'Age': 30, 'City': 'London'}, 
  {'Name': 'Rushabh', 'Age': 33, 'City': 'Paris'}
]

orient=’index’

If you have to access data by row index, you can use this approach.

import pandas as pd

data = {'Name': ['Krunal', 'Ankit', 'Rushabh'],
        'Age': [32, 30, 33],
        'City': ['New York', 'London', 'Paris']
        }

df = pd.DataFrame(data)

dict_with_index = df.to_dict(orient='index')

print(dict_with_index)

Output

{
  0: {'Name': 'Krunal', 'Age': 32, 'City': 'New York'}, 
  1: {'Name': 'Ankit', 'Age': 30, 'City': 'London'}, 
  2: {'Name': 'Rushabh', 'Age': 33, 'City': 'Paris'}
}

orient=’series’

You can use this option if you want to retain the Series functionality (like indexing, vectorized operations, etc.) while working with a dictionary structure.

import pandas as pd

data = {'Name': ['Krunal', 'Ankit', 'Rushabh'],
        'Age': [32, 30, 33],
        'City': ['New York', 'London', 'Paris']
        }

df = pd.DataFrame(data)

df_to_dict = df.to_dict(orient='series')

print(df_to_dict)

Output

orient=’split’

The “split” argument will create a dictionary with three key-value pairs:

1. index: It is a list of row index labels.
2. columns: It is a list of the column names.
3. data: It is a list of lists.

If you want to transfer structured data, you can use this orientation.

import pandas as pd

data = {'Name': ['Krunal', 'Ankit', 'Rushabh'],
        'Age': [32, 30, 33],
        'City': ['New York', 'London', 'Paris']
        }

df = pd.DataFrame(data)

dict_with_split = df.to_dict(orient='split')

print(dict_with_split)

Output

{
  'index': [0, 1, 2], 
  'columns': ['Name', 'Age', 'City'], 
  'data': [['Krunal', 32, 'New York'], ['Ankit', 30, 'London'], ['Rushabh', 33, 'Paris']]
}

orient=’tight’

This argument is helpful when you need to preserve metadata about DataFrame’s structure.

import pandas as pd

data = {'Name': ['Krunal', 'Ankit', 'Rushabh'],
        'Age': [32, 30, 33],
        'City': ['New York', 'London', 'Paris']
        }

df = pd.DataFrame(data)

dict_with_tight = df.to_dict(orient='tight')

print(dict_with_tight)

Output

{
  'index': [0, 1, 2], 
  'columns': ['Name', 'Age', 'City'], 
  'data': [['Krunal', 32, 'New York'], ['Ankit', 30, 'London'], ['Rushabh', 33, 'Paris']],   'index_names': [None], 'column_names': [None]
}

That’s all!