'Safe method to get value of nested dictionary

I have a nested dictionary. Is there only one way to get values out safely?

try:
    example_dict['key1']['key2']
except KeyError:
    pass

Or maybe python has a method like get() for nested dictionary ?



Solution 1:[1]

You could use get twice:

example_dict.get('key1', {}).get('key2')

This will return None if either key1 or key2 does not exist.

Note that this could still raise an AttributeError if example_dict['key1'] exists but is not a dict (or a dict-like object with a get method). The try..except code you posted would raise a TypeError instead if example_dict['key1'] is unsubscriptable.

Another difference is that the try...except short-circuits immediately after the first missing key. The chain of get calls does not.


If you wish to preserve the syntax, example_dict['key1']['key2'] but do not want it to ever raise KeyErrors, then you could use the Hasher recipe:

class Hasher(dict):
    # https://stackoverflow.com/a/3405143/190597
    def __missing__(self, key):
        value = self[key] = type(self)()
        return value

example_dict = Hasher()
print(example_dict['key1'])
# {}
print(example_dict['key1']['key2'])
# {}
print(type(example_dict['key1']['key2']))
# <class '__main__.Hasher'>

Note that this returns an empty Hasher when a key is missing.

Since Hasher is a subclass of dict you can use a Hasher in much the same way you could use a dict. All the same methods and syntax is available, Hashers just treat missing keys differently.

You can convert a regular dict into a Hasher like this:

hasher = Hasher(example_dict)

and convert a Hasher to a regular dict just as easily:

regular_dict = dict(hasher)

Another alternative is to hide the ugliness in a helper function:

def safeget(dct, *keys):
    for key in keys:
        try:
            dct = dct[key]
        except KeyError:
            return None
    return dct

So the rest of your code can stay relatively readable:

safeget(example_dict, 'key1', 'key2')

Solution 2:[2]

You could also use python reduce:

def deep_get(dictionary, *keys):
    return reduce(lambda d, key: d.get(key) if d else None, keys, dictionary)

Solution 3:[3]

By combining all of these answer here and small changes that I made, I think this function would be useful. its safe, quick, easily maintainable.

def deep_get(dictionary, keys, default=None):
    return reduce(lambda d, key: d.get(key, default) if isinstance(d, dict) else default, keys.split("."), dictionary)

Example :

from functools import reduce
def deep_get(dictionary, keys, default=None):
    return reduce(lambda d, key: d.get(key, default) if isinstance(d, dict) else default, keys.split("."), dictionary)

person = {'person':{'name':{'first':'John'}}}
print(deep_get(person, "person.name.first"))    # John

print(deep_get(person, "person.name.lastname")) # None

print(deep_get(person, "person.name.lastname", default="No lastname"))  # No lastname

Solution 4:[4]

You can .get an empty dictionary, in the first stage.

example_dict.get('key1',{}).get('key2')

Solution 5:[5]

Building up on Yoav's answer, an even safer approach:

def deep_get(dictionary, *keys):
    return reduce(lambda d, key: d.get(key, None) if isinstance(d, dict) else None, keys, dictionary)

Solution 6:[6]

A recursive solution. It's not the most efficient but I find it a bit more readable than the other examples and it doesn't rely on functools.

def deep_get(d, keys):
    if not keys or d is None:
        return d
    return deep_get(d.get(keys[0]), keys[1:])

Example

d = {'meta': {'status': 'OK', 'status_code': 200}}
deep_get(d, ['meta', 'status_code'])     # => 200
deep_get(d, ['garbage', 'status_code'])  # => None

A more polished version

def deep_get(d, keys, default=None):
    """
    Example:
        d = {'meta': {'status': 'OK', 'status_code': 200}}
        deep_get(d, ['meta', 'status_code'])          # => 200
        deep_get(d, ['garbage', 'status_code'])       # => None
        deep_get(d, ['meta', 'garbage'], default='-') # => '-'
    """
    assert type(keys) is list
    if d is None:
        return default
    if not keys:
        return d
    return deep_get(d.get(keys[0]), keys[1:], default)

Solution 7:[7]

I suggest you to try python-benedict.

It is a dict subclass that provides keypath support and much more.

Installation: pip install python-benedict

from benedict import benedict

example_dict = benedict(example_dict, keypath_separator='.')

now you can access nested values using keypath:

val = example_dict['key1.key2']

# using 'get' method to avoid a possible KeyError:
val = example_dict.get('key1.key2')

or access nested values using keys list:

val = example_dict['key1', 'key2']

# using get to avoid a possible KeyError:
val = example_dict.get(['key1', 'key2'])

It is well tested and open-source on GitHub:

https://github.com/fabiocaccamo/python-benedict

Note: I am the author of this project

Solution 8:[8]

While the reduce approach is neat and short, I think a simple loop is easier to grok. I've also included a default parameter.

def deep_get(_dict, keys, default=None):
    for key in keys:
        if isinstance(_dict, dict):
            _dict = _dict.get(key, default)
        else:
            return default
    return _dict

As an exercise to understand how the reduce one-liner worked, I did the following. But ultimately the loop approach seems more intuitive to me.

def deep_get(_dict, keys, default=None):

    def _reducer(d, key):
        if isinstance(d, dict):
            return d.get(key, default)
        return default

    return reduce(_reducer, keys, _dict)

Usage

nested = {'a': {'b': {'c': 42}}}

print deep_get(nested, ['a', 'b'])
print deep_get(nested, ['a', 'b', 'z', 'z'], default='missing')

Solution 9:[9]

glom is a nice library that can into dotted queries too:

In [1]: from glom import glom

In [2]: data = {'a': {'b': {'c': 'd'}}}

In [3]: glom(data, "a.b.c")
Out[3]: 'd'

A query failure has a nice stack trace, indicating the exact failure spot:

In [4]: glom(data, "a.b.foo")
---------------------------------------------------------------------------
PathAccessError                           Traceback (most recent call last)
<ipython-input-4-2a3467493ac4> in <module>
----> 1 glom(data, "a.b.foo")

~/.cache/pypoetry/virtualenvs/neural-knapsack-dE7ihQtM-py3.8/lib/python3.8/site-packages/glom/core.py in glom(target, spec, **kwargs)
   2179 
   2180     if err:
-> 2181         raise err
   2182     return ret
   2183 

PathAccessError: error raised while processing, details below.
 Target-spec trace (most recent last):
 - Target: {'a': {'b': {'c': 'd'}}}
 - Spec: 'a.b.foo'
glom.core.PathAccessError: could not access 'foo', part 2 of Path('a', 'b', 'foo'), got error: KeyError('foo')

Safeguard with default:

In [5]: glom(data, "a.b.foo", default="spam")
Out[5]: 'spam'

The beauty of glom is in the versatile spec parameter. For example, one can easily extract all first names from the following data:

In [8]: data = {
   ...:     "people": [
   ...:         {"first_name": "Alice", "last_name": "Adams"},
   ...:         {"first_name": "Bob", "last_name": "Barker"}
   ...:     ]
   ...: }

In [9]: glom(data, ("people", ["first_name"]))
Out[9]: ['Alice', 'Bob']

Read the glom docs for more examples.

Solution 10:[10]

You can use pydash:

import pydash as _  #NOTE require `pip install pydash`

_.get(example_dict, 'key1.key2', default='Default')

https://pydash.readthedocs.io/en/latest/api.html

Solution 11:[11]

A simple class that can wrap a dict, and retrieve based on a key:

class FindKey(dict):
    def get(self, path, default=None):
        keys = path.split(".")
        val = None

        for key in keys:
            if val:
                if isinstance(val, list):
                    val = [v.get(key, default) if v else None for v in val]
                else:
                    val = val.get(key, default)
            else:
                val = dict.get(self, key, default)

            if not val:
                break

        return val

For example:

person = {'person':{'name':{'first':'John'}}}
FindDict(person).get('person.name.first') # == 'John'

If the key doesn't exist, it returns None by default. You can override that using a default= key in the FindDict wrapper -- for example`:

FindDict(person, default='').get('person.name.last') # == doesn't exist, so ''

Solution 12:[12]

I adapted GenesRus and unutbu's answer in this very simple:

class new_dict(dict):
    def deep_get(self, *args, default=None):
        _empty_dict = {}
        out = self
        for key in args:
            out = out.get(key, _empty_dict)
        return out if out else default

it works with: d = new_dict(some_data) d.deep_get("key1", "key2", "key3", ..., default=some_value)

Solution 13:[13]

for a second level key retrieving, you can do this:

key2_value = (example_dict.get('key1') or {}).get('key2')

Solution 14:[14]

Starting with Python 3.4 you may use with suppress (KeyError) to access nested json objects without worrying of Keyerror

from contextlib import suppress

with suppress(KeyError):
    a1 = json_obj['key1']['key2']['key3']
    a2 = json_obj['key4']['key5']['key6']
    a3 = json_obj['key7']['key8']['key9']

Courtesy of Techdragon. Have a look at his answer for further details: https://stackoverflow.com/a/45874251/1189659

Solution 15:[15]

After seeing this for deeply getting attributes, I made the following to safely get nested dict values using dot notation. This works for me because my dicts are deserialized MongoDB objects, so I know the key names don't contain .s. Also, in my context, I can specify a falsy fallback value (None) that I don't have in my data, so I can avoid the try/except pattern when calling the function.

from functools import reduce # Python 3
def deepgetitem(obj, item, fallback=None):
    """Steps through an item chain to get the ultimate value.

    If ultimate value or path to value does not exist, does not raise
    an exception and instead returns `fallback`.

    >>> d = {'snl_final': {'about': {'_icsd': {'icsd_id': 1}}}}
    >>> deepgetitem(d, 'snl_final.about._icsd.icsd_id')
    1
    >>> deepgetitem(d, 'snl_final.about._sandbox.sbx_id')
    >>>
    """
    def getitem(obj, name):
        try:
            return obj[name]
        except (KeyError, TypeError):
            return fallback
    return reduce(getitem, item.split('.'), obj)

Solution 16:[16]

Yet another function for the same thing, also returns a boolean to represent whether the key was found or not and handles some unexpected errors.

'''
json : json to extract value from if exists
path : details.detail.first_name
            empty path represents root

returns a tuple (boolean, object)
        boolean : True if path exists, otherwise False
        object : the object if path exists otherwise None

'''
def get_json_value_at_path(json, path=None, default=None):

    if not bool(path):
        return True, json
    if type(json) is not dict :
        raise ValueError(f'json={json}, path={path} not supported, json must be a dict')
    if type(path) is not str and type(path) is not list:
        raise ValueError(f'path format {path} not supported, path can be a list of strings like [x,y,z] or a string like x.y.z')

    if type(path) is str:
        path = path.strip('.').split('.')
    key = path[0]
    if key in json.keys():
        return get_json_value_at_path(json[key], path[1:], default)
    else:
        return False, default

example usage:

my_json = {'details' : {'first_name' : 'holla', 'last_name' : 'holla'}}
print(get_json_value_at_path(my_json, 'details.first_name', ''))
print(get_json_value_at_path(my_json, 'details.phone', ''))

(True, 'holla')

(False, '')

Solution 17:[17]

There are already lots of good answers but I have come up with a function called get similar to lodash get in JavaScript land that also supports reaching into lists by index:

def get(value, keys, default_value = None):
'''
    Useful for reaching into nested JSON like data
    Inspired by JavaScript lodash get and Clojure get-in etc.
'''
  if value is None or keys is None:
      return None
  path = keys.split('.') if isinstance(keys, str) else keys
  result = value
  def valid_index(key):
      return re.match('^([1-9][0-9]*|[0-9])$', key) and int(key) >= 0
  def is_dict_like(v):
      return hasattr(v, '__getitem__') and hasattr(v, '__contains__')
  for key in path:
      if isinstance(result, list) and valid_index(key) and int(key) < len(result):
          result = result[int(key)] if int(key) < len(result) else None
      elif is_dict_like(result) and key in result:
          result = result[key]
      else:
          result = default_value
          break
  return result

def test_get():
  assert get(None, ['foo']) == None
  assert get({'foo': 1}, None) == None
  assert get(None, None) == None
  assert get({'foo': 1}, []) == {'foo': 1}
  assert get({'foo': 1}, ['foo']) == 1
  assert get({'foo': 1}, ['bar']) == None
  assert get({'foo': 1}, ['bar'], 'the default') == 'the default'
  assert get({'foo': {'bar': 'hello'}}, ['foo', 'bar']) == 'hello'
  assert get({'foo': {'bar': 'hello'}}, 'foo.bar') == 'hello'
  assert get({'foo': [{'bar': 'hello'}]}, 'foo.0.bar') == 'hello'
  assert get({'foo': [{'bar': 'hello'}]}, 'foo.1') == None
  assert get({'foo': [{'bar': 'hello'}]}, 'foo.1.bar') == None
  assert get(['foo', 'bar'], '1') == 'bar'
  assert get(['foo', 'bar'], '2') == None

Solution 18:[18]

If you want to use another library for a solution, this is work best

https://github.com/maztohir/dict-path

from dict-path import DictPath

data_dict = {
  "foo1": "bar1",
  "foo2": "bar2",
  "foo3": {
     "foo4": "bar4",
     "foo5": {
        "foo6": "bar6",
        "foo7": "bar7",
     },
  }
}

data_dict_path = DictPath(data_dict)
data_dict_path.get('key1/key2/key3')

Solution 19:[19]

Here is a solution based on the unutbu's function answer plus:

  1. python naming guidelines
  2. default value as a parameter
  3. not using try but just checking if key is on object
def safe_get(dictionary, *keys, default=None):
    for key in keys:
        if key not in dictionary:
            return default
        dictionary = dictionary[key]
    return dictionary

Solution 20:[20]

An adaptation of unutbu's answer that I found useful in my own code:

example_dict.setdefaut('key1', {}).get('key2')

It generates a dictionary entry for key1 if it does not have that key already so that you avoid the KeyError. If you want to end up a nested dictionary that includes that key pairing anyway like I did, this seems like the easiest solution.

Solution 21:[21]

Since raising an key error if one of keys is missing is a reasonable thing to do, we can even not check for it and get it as single as that:

def get_dict(d, kl):
  cur = d[kl[0]]
  return get_dict(cur, kl[1:]) if len(kl) > 1 else cur

Solution 22:[22]

Little improvement to reduce approach to make it work with list. Also using data path as string divided by dots instead of array.

def deep_get(dictionary, path):
    keys = path.split('.')
    return reduce(lambda d, key: d[int(key)] if isinstance(d, list) else d.get(key) if d else None, keys, dictionary)

Solution 23:[23]

A solution I've used that is similar to the double get but with the additional ability to avoid a TypeError using if else logic:

    value = example_dict['key1']['key2'] if example_dict.get('key1') and example_dict['key1'].get('key2') else default_value

However, the more nested the dictionary the more cumbersome this becomes.

Solution 24:[24]

For nested dictionary/JSON lookups, you can use dictor

pip install dictor

dict object

{
    "characters": {
        "Lonestar": {
            "id": 55923,
            "role": "renegade",
            "items": [
                "space winnebago",
                "leather jacket"
            ]
        },
        "Barfolomew": {
            "id": 55924,
            "role": "mawg",
            "items": [
                "peanut butter jar",
                "waggy tail"
            ]
        },
        "Dark Helmet": {
            "id": 99999,
            "role": "Good is dumb",
            "items": [
                "Shwartz",
                "helmet"
            ]
        },
        "Skroob": {
            "id": 12345,
            "role": "Spaceballs CEO",
            "items": [
                "luggage"
            ]
        }
    }
}

to get Lonestar's items, simply provide a dot-separated path, ie

import json
from dictor import dictor

with open('test.json') as data: 
    data = json.load(data)

print dictor(data, 'characters.Lonestar.items')

>> [u'space winnebago', u'leather jacket']

you can provide fallback value in case the key isnt in path

theres tons more options you can do, like ignore letter casing and using other characters besides '.' as a path separator,

https://github.com/perfecto25/dictor

Solution 25:[25]

I little changed this answer. I added checking if we're using list with numbers. So now we can use it whichever way. deep_get(allTemp, [0], {}) or deep_get(getMinimalTemp, [0, minimalTemperatureKey], 26) etc

def deep_get(_dict, keys, default=None):
    def _reducer(d, key):
        if isinstance(d, dict):
            return d.get(key, default)
        if isinstance(d, list):
            return d[key] if len(d) > 0 else default
        return default
    return reduce(_reducer, keys, _dict)

Solution 26:[26]

Recursive method (?? ??????????)

Example dict:

foo = [{'feature_name': 'Sample Creator > Contract Details > Elements of the page',
  'scenarios': [{'scenario_name': 'SC, CD, Elements of the page',
                 'scenario_status': 'failed',
                 'scenario_tags': None,
                 'steps': [{'duration': 0,
                            'name': 'I open application Stage and login by '
                                    'SPT_LOGIN and password SPT_PWD',
                            'status': 'untested'},
                           {'duration': 0,
                            'name': 'I open Sample Creator query page',
                            'status': 'untested'},
                           {'duration': 7.78166389465332,
                            'name': 'I open application Stage and login by '
                                    'SPT_LOGIN and password SPT_PWD',
                            'status': 'passed'},
                           {'duration': 3.985326051712036,
                            'name': 'I open Sample Creator query page',
                            'status': 'passed'},
                           {'duration': 2.9063704013824463,
                            'name': 'Enter value: '
                                    'X-2008-CON-007,X-2011-CON-016 in '
                                    'textarea: project_text_area sleep: 1',
                            'status': 'passed'},
                           {'duration': 4.4447715282440186,
                            'name': 'I press on GET DATA',
                            'status': 'passed'},
                           {'duration': 1.1209557056427002,
                            'name': 'Verify the top table on Contract Details',
                            'status': 'passed'},
                           {'duration': 3.8173601627349854,
                            'name': 'I export contract_details table by offset '
                                    'x:100, y:150',
                            'status': 'passed'},
                           {'duration': 1.032956600189209,
                            'name': 'Check data of '
                                    'sc__cd_elements_of_the_page_1 and skip '
                                    'cols None',
                            'status': 'passed'},
                           {'duration': 0.04593634605407715,
                            'name': "Verify 'Number of Substances' column "
                                    'values',
                            'status': 'passed'},
                           {'duration': 0.10199904441833496,
                            'name': 'Substance Sample Details bottom table '
                                    'columns',
                            'status': 'passed'},
                           {'duration': 0.0009999275207519531,
                            'name': 'Verify the Substance Sample Details '
                                    'bottom table',
                            'status': 'passed'},
                           {'duration': 3.8558616638183594,
                            'name': 'I export substance_sample_details table '
                                    'by offset x:100, y:150',
                            'status': 'passed'},
                           {'duration': 1.0329277515411377,
                            'name': 'Check data of '
                                    'sc__cd_elements_of_the_page_2 and skip '
                                    'cols None',
                            'status': 'passed'},
                           {'duration': 0.2879970073699951,
                            'name': 'Click on AG-13369',
                            'status': 'passed'},
                           {'duration': 3.800830364227295,
                            'name': 'I export substance_sample_details table '
                                    'by offset x:100, y:150',
                            'status': 'passed'},
                           {'duration': 1.0169551372528076,
                            'name': 'Check data of '
                                    'sc__cd_elements_of_the_page_3 and skip '
                                    'cols None',
                            'status': 'passed'},
                           {'duration': 1.7484464645385742,
                            'name': 'Select all cells, table: 2',
                            'status': 'passed'},
                           {'duration': 3.812828779220581,
                            'name': 'I export substance_sample_details table '
                                    'by offset x:100, y:150',
                            'status': 'passed'},
                           {'duration': 1.0029594898223877,
                            'name': 'Check data of '
                                    'sc__cd_elements_of_the_page_2 and skip '
                                    'cols None',
                            'status': 'passed'},
                           {'duration': 1.6729373931884766,
                            'name': 'Set window size x:800, y:600',
                            'status': 'passed'},
                           {'duration': 30.145705699920654,
                            'name': 'All scrollers are placed on top 6 and far '
                                    'left 8',
                            'status': 'failed'}]}]},
  {'feature_name': 'Sample Creator > Substance Sample History > Elements of the '
                  'page',
  'scenarios': [{'scenario_name': 'SC, SSH, Elements of the page',
                 'scenario_status': 'passed',
                 'scenario_tags': None,
                 'steps': [{'duration': 0,
                            'name': 'I open application Stage and login by '
                                    'SPT_LOGIN and password SPT_PWD',
                            'status': 'untested'},
                           {'duration': 0,
                            'name': 'I open Sample Creator query page',
                            'status': 'untested'},
                           {'duration': 7.305850505828857,
                            'name': 'I open application Stage and login by '
                                    'SPT_LOGIN and password SPT_PWD',
                            'status': 'passed'},
                           {'duration': 3.500955104827881,
                            'name': 'I open Sample Creator query page',
                            'status': 'passed'},
                           {'duration': 3.0419492721557617,
                            'name': 'Enter value: NOA401800 SYN-NOA '
                                    'A,S4A482070C SYN-ISN-OLD '
                                    'O,S04A482167T,S04A482190Y,CSAA796564,CSCD106701 '
                                    'in textarea: id_text_area sleep: 1',
                            'status': 'passed'},
                           {'duration': 49.567158460617065,
                            'name': 'I press on GET DATA',
                            'status': 'passed'},
                           {'duration': 0.13904356956481934,
                            'name': 'Open substance_sample_history',
                            'status': 'passed'},
                           {'duration': 1.1039845943450928,
                            'name': 'Columns displayed',
                            'status': 'passed'},
                           {'duration': 3.881945848464966,
                            'name': 'I export export_parent_table table by '
                                    'offset x:100, y:150',
                            'status': 'passed'},
                           {'duration': 1.0334820747375488,
                            'name': 'Check data of '
                                    'sc__ssh_elements_of_the_page_1 and skip '
                                    'cols None',
                            'status': 'passed'},
                           {'duration': 0.0319981575012207,
                            'name': "Title is 'Additional Details for Marked "
                                    "Rows'",
                            'status': 'passed'},
                           {'duration': 0.08897256851196289,
                            'name': 'Columns displayed (the same as in top '
                                    'table)',
                            'status': 'passed'},
                           {'duration': 25.192569971084595,
                            'name': 'Verify the content of the bottom table',
                            'status': 'passed'},
                           {'duration': 4.308935880661011,
                            'name': 'I export '
                                    'additional_details_for_marked_rows table '
                                    'by offset x:100, y:150',
                            'status': 'passed'},
                           {'duration': 1.0089836120605469,
                            'name': 'Check data of '
                                    'sc__ssh_elements_of_the_page_1 and skip '
                                    'cols None',
                            'status': 'passed'}]}]}]

Code:

def get_keys(_dict: dict, prefix: list):
    prefix += list(_dict.keys())
    return prefix


def _loop_elements(elems:list, prefix=None, limit=None):
    prefix = prefix or []
    limit = limit or 9
    try:
        if len(elems) != 0 and isinstance(elems, list):
            for _ in elems:
                if isinstance(_, dict):
                    get_keys(_, prefix)
                    for item in _.values():
                        _loop_elements(item, prefix, limit)
        return prefix[:limit]
    except TypeError:
        return


>>>goo = _loop_elements(foo,limit=9)
>>>goo
['feature_name', 'scenarios', 'scenario_name', 'scenario_status', 'scenario_tags', 'steps', 'duration', 'name', 'status']

Solution 27:[27]

def safeget(_dct, *_keys):
    if not isinstance(_dct, dict): raise TypeError("Is not instance of dict")
    def foo(dct, *keys):
        if len(keys) == 0: return dct
        elif not isinstance(_dct, dict): return None
        else: return foo(dct.get(keys[0], None), *keys[1:])
    return foo(_dct, *_keys)

assert safeget(dict()) == dict()
assert safeget(dict(), "test") == None
assert safeget(dict([["a", 1],["b", 2]]),"a", "d") == None
assert safeget(dict([["a", 1],["b", 2]]),"a") == 1
assert safeget({"a":{"b":{"c": 2}},"d":1}, "a", "b")["c"] == 2

Solution 28:[28]

I have written a package deepextract that does exactly what you want: https://github.com/ya332/deepextract You can do

from deepextract import deepextract
# Demo: deepextract.extract_key(obj, key)
deeply_nested_dict = {
    "items": {
        "item": {
            "id": {
                "type": {
                    "donut": {
                        "name": {
                            "batters": {
                                "my_target_key": "my_target_value"
                            }
                        }
                    }
                }
            }
        }
    }
}
print(deepextract.extract_key(deeply_nested_dict, "my_target_key") == "my_target_value")

returns

True

Solution 29:[29]

My implementation that descends into sub-dicts, ignores None values, but fails with a TypeError if anything else is discovered

def deep_get(d: dict, *keys, default=None):
    """ Safely get a nested value from a dict

    Example:
        config = {'device': None}
        deep_get(config, 'device', 'settings', 'light')
        # -> None
        
    Example:
        config = {'device': True}
        deep_get(config, 'device', 'settings', 'light')
        # -> TypeError

    Example:
        config = {'device': {'settings': {'light': 'bright'}}}
        deep_get(config, 'device', 'settings', 'light')
        # -> 'light'

    Note that it returns `default` is a key is missing or when it's None.
    It will raise a TypeError if a value is anything else but a dict or None.
    
    Args:
        d: The dict to descend into
        keys: A sequence of keys to follow
        default: Custom default value
    """
    # Descend while we can
    try:
        for k in keys:
            d = d[k]
    # If at any step a key is missing, return default
    except KeyError:
        return default
    # If at any step the value is not a dict...
    except TypeError:
        # ... if it's a None, return default. Assume it would be a dict.
        if d is None:
            return default
        # ... if it's something else, raise
        else:
            raise
    # If the value was found, return it
    else:
        return d