Elegant way to remove fields from nested dictionaries
Question:
I had to remove some fields from a dictionary, the keys for those fields are on a list. So I wrote this function:
def delete_keys_from_dict(dict_del, lst_keys):
"""
Delete the keys present in lst_keys from the dictionary.
Loops recursively over nested dictionaries.
"""
dict_foo = dict_del.copy() #Used as iterator to avoid the 'DictionaryHasChanged' error
for field in dict_foo.keys():
if field in lst_keys:
del dict_del[field]
if type(dict_foo[field]) == dict:
delete_keys_from_dict(dict_del[field], lst_keys)
return dict_del
This code works, but it’s not very elegant and I’m sure that there is a better solution.
Answers:
def delete_keys_from_dict(dict_del, lst_keys):
for k in lst_keys:
try:
del dict_del[k]
except KeyError:
pass
for v in dict_del.values():
if isinstance(v, dict):
delete_keys_from_dict(v, lst_keys)
return dict_del
Since you already need to loop through every element in the dict, I’d stick with a single loop and just make sure to use a set for looking up the keys to delete
def delete_keys_from_dict(dict_del, the_keys):
"""
Delete the keys present in the lst_keys from the dictionary.
Loops recursively over nested dictionaries.
"""
# make sure the_keys is a set to get O(1) lookups
if type(the_keys) is not set:
the_keys = set(the_keys)
for k,v in dict_del.items():
if k in the_keys:
del dict_del[k]
if isinstance(v, dict):
delete_keys_from_dict(v, the_keys)
return dict_del
Since the question requested an elegant way, I’ll submit my general-purpose solution to wrangling nested structures. First, install the boltons utility package with pip install boltons
, then:
from boltons.iterutils import remap
data = {'one': 'remains', 'this': 'goes', 'of': 'course'}
bad_keys = set(['this', 'is', 'a', 'list', 'of', 'keys'])
drop_keys = lambda path, key, value: key not in bad_keys
clean = remap(data, visit=drop_keys)
print(clean)
# Output:
{'one': 'remains'}
In short, the remap utility is a full-featured, yet succinct approach to handling real-world data structures which are often nested, and can even contain cycles and special containers.
This page has many more examples, including ones working with much larger objects from Github’s API.
It’s pure-Python, so it works everywhere, and is fully tested in Python 2.7 and 3.3+. Best of all, I wrote it for exactly cases like this, so if you find a case it doesn’t handle, you can bug me to fix it right here.
Using the awesome code from this post and add a small statement:
def remove_fields(self, d, list_of_keys_to_remove):
if not isinstance(d, (dict, list)):
return d
if isinstance(d, list):
return [v for v in (self.remove_fields(v, list_of_keys_to_remove) for v in d) if v]
return {k: v for k, v in ((k, self.remove_fields(v, list_of_keys_to_remove)) for k, v in d.items()) if k not in list_of_keys_to_remove}
I think the following is more elegant:
def delete_keys_from_dict(dict_del, lst_keys):
if not isinstance(dict_del, dict):
return dict_del
return {
key: value
for key, value in (
(key, delete_keys_from_dict(value, lst_keys))
for key, value in dict_del.items()
)
if key not in lst_keys
}
Example usage:
test_dict_in = {
1: {1: {0: 2, 3: 4}},
0: {2: 3},
2: {5: {0: 4}, 6: {7: 8}},
}
test_dict_out = {
1: {1: {3: 4}},
2: {5: {}, 6: {7: 8}},
}
assert delete_keys_from_dict(test_dict_in, [0]) == test_dict_out
First of, I think your code is working and not inelegant. There’s no immediate reason not to use the code you presented.
There are a few things that could be better though:
Comparing the type
Your code contains the line:
if type(dict_foo[field]) == dict:
That can be definitely improved. Generally (see also PEP8) you should use isinstance
instead of comparing types:
if isinstance(dict_foo[field], dict)
However that will also return True
if dict_foo[field]
is a subclass of dict
. If you don’t want that, you could also use is
instead of ==
. That will be marginally (and probably unnoticeable) faster.
If you also want to allow arbitary dict-like objects you could go a step further and test if it’s a collections.abc.MutableMapping
. That will be True
for dict
and dict
subclasses and for all mutable mappings that explicitly implement that interface without subclassing dict
, for example UserDict
:
>>> from collections import MutableMapping
>>> # from UserDict import UserDict # Python 2.x
>>> from collections import UserDict # Python 3.x - 3.6
>>> # from collections.abc import MutableMapping # Python 3.7+
>>> isinstance(UserDict(), MutableMapping)
True
>>> isinstance(UserDict(), dict)
False
Inplace modification and return value
Typically functions either modify a data structure inplace or return a new (modified) data structure. Just to mention a few examples: list.append
, dict.clear
, dict.update
all modify the data structure inplace and return None
. That makes it easier to keep track what a function does. However that’s not a hard rule and there are always valid exceptions from this rule. However personally I think a function like this doesn’t need to be an exception and I would simply remove the return dict_del
line and let it implicitly return None
, but YMMV.
Removing the keys from the dictionary
You copied the dictionary to avoid problems when you remove key-value pairs during the iteration. However, as already mentioned by another answer you could just iterate over the keys that should be removed and try to delete them:
for key in keys_to_remove:
try:
del dict[key]
except KeyError:
pass
That has the additional advantage that you don’t need to nest two loops (which could be slower, especially if the number of keys that need to be removed is very long).
If you don’t like empty except
clauses you can also use: contextlib.suppress
(requires Python 3.4+):
from contextlib import suppress
for key in keys_to_remove:
with suppress(KeyError):
del dict[key]
Variable names
There are a few variables I would rename because they are just not descriptive or even misleading:
-
delete_keys_from_dict
should probably mention the subdict-handling, maybe delete_keys_from_dict_recursive
.
-
dict_del
sounds like a deleted dict. I tend to prefer names like dictionary
or dct
because the function name already describes what is done to the dictionary.
-
lst_keys
, same there. I’d probably use just keys
there. If you want to be more specific something like keys_sequence
would make more sense because it accepts any sequence
(you just have to be able to iterate over it multiple times), not just lists.
-
dict_foo
, just no…
-
field
isn’t really appropriate either, it’s a key.
Putting it all together:
As I said before I personally would modify the dictionary in-place and not return the dictionary again. Because of that I present two solutions, one that modifies it in-place but doesn’t return anything and one that creates a new dictionary with the keys removed.
The version that modifies in-place (very much like Ned Batchelders solution):
from collections import MutableMapping
from contextlib import suppress
def delete_keys_from_dict(dictionary, keys):
for key in keys:
with suppress(KeyError):
del dictionary[key]
for value in dictionary.values():
if isinstance(value, MutableMapping):
delete_keys_from_dict(value, keys)
And the solution that returns a new object:
from collections import MutableMapping
def delete_keys_from_dict(dictionary, keys):
keys_set = set(keys) # Just an optimization for the "if key in keys" lookup.
modified_dict = {}
for key, value in dictionary.items():
if key not in keys_set:
if isinstance(value, MutableMapping):
modified_dict[key] = delete_keys_from_dict(value, keys_set)
else:
modified_dict[key] = value # or copy.deepcopy(value) if a copy is desired for non-dicts.
return modified_dict
However it only makes copies of the dictionaries, the other values are not returned as copy, you could easily wrap these in copy.deepcopy
(I put a comment in the appropriate place of the code) if you want that.
def delete_keys_from_dict(d, to_delete):
if isinstance(to_delete, str):
to_delete = [to_delete]
if isinstance(d, dict):
for single_to_delete in set(to_delete):
if single_to_delete in d:
del d[single_to_delete]
for k, v in d.items():
delete_keys_from_dict(v, to_delete)
elif isinstance(d, list):
for i in d:
delete_keys_from_dict(i, to_delete)
d = {'a': 10, 'b': [{'c': 10, 'd': 10, 'a': 10}, {'a': 10}], 'c': 1 }
delete_keys_from_dict(d, ['a', 'c']) # inplace deletion
print(d)
>>> {'b': [{'d': 10}, {}]}
This solution works for dict
and list
in a given nested dict
. The input to_delete
can be a list
of str
to be deleted or a single str
.
Plese note, that if you remove the only key in a dict
, you will get an empty dict
.
this works with dict
s containing Iterable
s (list
, …) that may contain dict
. Python 3. For Python 2 unicode
should also be excluded from the iteration. Also there may be some iterables that don’t work that I’m not aware of. (i.e. will lead to inifinite recursion)
from collections.abc import Iterable
def deep_omit(d, keys):
if isinstance(d, dict):
for k in keys:
d.pop(k, None)
for v in d.values():
deep_omit(v, keys)
elif isinstance(d, Iterable) and not isinstance(d, str):
for e in d:
deep_omit(e, keys)
return d
I came here to search for a solution to remove keys from deeply nested Python3 dicts and all solutions seem to be somewhat complex.
Here’s a oneliner for removing keys from nested or flat dicts:
nested_dict = {
"foo": {
"bar": {
"foobar": {},
"shmoobar": {}
}
}
}
>>> {'foo': {'bar': {'foobar': {}, 'shmoobar': {}}}}
nested_dict.get("foo", {}).get("bar", {}).pop("shmoobar", None)
>>> {'foo': {'bar': {'foobar': {}}}}
I used .get() to not get KeyError and I also provide empty dict as default value up to the end of the chain. I do pop() for the last element and I provide None as the default there to avoid KeyError.
If you have nested keys as well and based on @John La Rooy’s answer here is an elegant solution:
from boltons.iterutils import remap
def sof_solution():
data = {"user": {"name": "test", "pwd": "******"}, "accounts": ["1", "2"]}
sensitive = {"user.pwd", "accounts"}
clean = remap(
data,
visit=lambda path, key, value: drop_keys(path, key, value, sensitive)
)
print(clean)
def drop_keys(path, key, value, sensitive):
if len(path) > 0:
nested_key = f"{'.'.join(path)}.{key}"
return nested_key not in sensitive
return key not in sensitive
sof_solution() # prints {'user': {'name': 'test'}}
Since nobody posted an interactive version that could be useful for someone:
def delete_key_from_dict(adict, key):
stack = [adict]
while stack:
elem = stack.pop()
if isinstance(elem, dict):
if key in elem:
del elem[key]
for k in elem:
stack.append(elem[k])
This version is probably what you would push to production. The recursive version is elegant and easy to write but it scales badly (by default Python uses a maximum recursion depth of 1000).
I had to remove some fields from a dictionary, the keys for those fields are on a list. So I wrote this function:
def delete_keys_from_dict(dict_del, lst_keys):
"""
Delete the keys present in lst_keys from the dictionary.
Loops recursively over nested dictionaries.
"""
dict_foo = dict_del.copy() #Used as iterator to avoid the 'DictionaryHasChanged' error
for field in dict_foo.keys():
if field in lst_keys:
del dict_del[field]
if type(dict_foo[field]) == dict:
delete_keys_from_dict(dict_del[field], lst_keys)
return dict_del
This code works, but it’s not very elegant and I’m sure that there is a better solution.
def delete_keys_from_dict(dict_del, lst_keys):
for k in lst_keys:
try:
del dict_del[k]
except KeyError:
pass
for v in dict_del.values():
if isinstance(v, dict):
delete_keys_from_dict(v, lst_keys)
return dict_del
Since you already need to loop through every element in the dict, I’d stick with a single loop and just make sure to use a set for looking up the keys to delete
def delete_keys_from_dict(dict_del, the_keys):
"""
Delete the keys present in the lst_keys from the dictionary.
Loops recursively over nested dictionaries.
"""
# make sure the_keys is a set to get O(1) lookups
if type(the_keys) is not set:
the_keys = set(the_keys)
for k,v in dict_del.items():
if k in the_keys:
del dict_del[k]
if isinstance(v, dict):
delete_keys_from_dict(v, the_keys)
return dict_del
Since the question requested an elegant way, I’ll submit my general-purpose solution to wrangling nested structures. First, install the boltons utility package with pip install boltons
, then:
from boltons.iterutils import remap
data = {'one': 'remains', 'this': 'goes', 'of': 'course'}
bad_keys = set(['this', 'is', 'a', 'list', 'of', 'keys'])
drop_keys = lambda path, key, value: key not in bad_keys
clean = remap(data, visit=drop_keys)
print(clean)
# Output:
{'one': 'remains'}
In short, the remap utility is a full-featured, yet succinct approach to handling real-world data structures which are often nested, and can even contain cycles and special containers.
This page has many more examples, including ones working with much larger objects from Github’s API.
It’s pure-Python, so it works everywhere, and is fully tested in Python 2.7 and 3.3+. Best of all, I wrote it for exactly cases like this, so if you find a case it doesn’t handle, you can bug me to fix it right here.
Using the awesome code from this post and add a small statement:
def remove_fields(self, d, list_of_keys_to_remove):
if not isinstance(d, (dict, list)):
return d
if isinstance(d, list):
return [v for v in (self.remove_fields(v, list_of_keys_to_remove) for v in d) if v]
return {k: v for k, v in ((k, self.remove_fields(v, list_of_keys_to_remove)) for k, v in d.items()) if k not in list_of_keys_to_remove}
I think the following is more elegant:
def delete_keys_from_dict(dict_del, lst_keys):
if not isinstance(dict_del, dict):
return dict_del
return {
key: value
for key, value in (
(key, delete_keys_from_dict(value, lst_keys))
for key, value in dict_del.items()
)
if key not in lst_keys
}
Example usage:
test_dict_in = {
1: {1: {0: 2, 3: 4}},
0: {2: 3},
2: {5: {0: 4}, 6: {7: 8}},
}
test_dict_out = {
1: {1: {3: 4}},
2: {5: {}, 6: {7: 8}},
}
assert delete_keys_from_dict(test_dict_in, [0]) == test_dict_out
First of, I think your code is working and not inelegant. There’s no immediate reason not to use the code you presented.
There are a few things that could be better though:
Comparing the type
Your code contains the line:
if type(dict_foo[field]) == dict:
That can be definitely improved. Generally (see also PEP8) you should use isinstance
instead of comparing types:
if isinstance(dict_foo[field], dict)
However that will also return True
if dict_foo[field]
is a subclass of dict
. If you don’t want that, you could also use is
instead of ==
. That will be marginally (and probably unnoticeable) faster.
If you also want to allow arbitary dict-like objects you could go a step further and test if it’s a collections.abc.MutableMapping
. That will be True
for dict
and dict
subclasses and for all mutable mappings that explicitly implement that interface without subclassing dict
, for example UserDict
:
>>> from collections import MutableMapping
>>> # from UserDict import UserDict # Python 2.x
>>> from collections import UserDict # Python 3.x - 3.6
>>> # from collections.abc import MutableMapping # Python 3.7+
>>> isinstance(UserDict(), MutableMapping)
True
>>> isinstance(UserDict(), dict)
False
Inplace modification and return value
Typically functions either modify a data structure inplace or return a new (modified) data structure. Just to mention a few examples: list.append
, dict.clear
, dict.update
all modify the data structure inplace and return None
. That makes it easier to keep track what a function does. However that’s not a hard rule and there are always valid exceptions from this rule. However personally I think a function like this doesn’t need to be an exception and I would simply remove the return dict_del
line and let it implicitly return None
, but YMMV.
Removing the keys from the dictionary
You copied the dictionary to avoid problems when you remove key-value pairs during the iteration. However, as already mentioned by another answer you could just iterate over the keys that should be removed and try to delete them:
for key in keys_to_remove:
try:
del dict[key]
except KeyError:
pass
That has the additional advantage that you don’t need to nest two loops (which could be slower, especially if the number of keys that need to be removed is very long).
If you don’t like empty except
clauses you can also use: contextlib.suppress
(requires Python 3.4+):
from contextlib import suppress
for key in keys_to_remove:
with suppress(KeyError):
del dict[key]
Variable names
There are a few variables I would rename because they are just not descriptive or even misleading:
-
delete_keys_from_dict
should probably mention the subdict-handling, maybedelete_keys_from_dict_recursive
. -
dict_del
sounds like a deleted dict. I tend to prefer names likedictionary
ordct
because the function name already describes what is done to the dictionary. -
lst_keys
, same there. I’d probably use justkeys
there. If you want to be more specific something likekeys_sequence
would make more sense because it accepts anysequence
(you just have to be able to iterate over it multiple times), not just lists. -
dict_foo
, just no… -
field
isn’t really appropriate either, it’s a key.
Putting it all together:
As I said before I personally would modify the dictionary in-place and not return the dictionary again. Because of that I present two solutions, one that modifies it in-place but doesn’t return anything and one that creates a new dictionary with the keys removed.
The version that modifies in-place (very much like Ned Batchelders solution):
from collections import MutableMapping
from contextlib import suppress
def delete_keys_from_dict(dictionary, keys):
for key in keys:
with suppress(KeyError):
del dictionary[key]
for value in dictionary.values():
if isinstance(value, MutableMapping):
delete_keys_from_dict(value, keys)
And the solution that returns a new object:
from collections import MutableMapping
def delete_keys_from_dict(dictionary, keys):
keys_set = set(keys) # Just an optimization for the "if key in keys" lookup.
modified_dict = {}
for key, value in dictionary.items():
if key not in keys_set:
if isinstance(value, MutableMapping):
modified_dict[key] = delete_keys_from_dict(value, keys_set)
else:
modified_dict[key] = value # or copy.deepcopy(value) if a copy is desired for non-dicts.
return modified_dict
However it only makes copies of the dictionaries, the other values are not returned as copy, you could easily wrap these in copy.deepcopy
(I put a comment in the appropriate place of the code) if you want that.
def delete_keys_from_dict(d, to_delete):
if isinstance(to_delete, str):
to_delete = [to_delete]
if isinstance(d, dict):
for single_to_delete in set(to_delete):
if single_to_delete in d:
del d[single_to_delete]
for k, v in d.items():
delete_keys_from_dict(v, to_delete)
elif isinstance(d, list):
for i in d:
delete_keys_from_dict(i, to_delete)
d = {'a': 10, 'b': [{'c': 10, 'd': 10, 'a': 10}, {'a': 10}], 'c': 1 }
delete_keys_from_dict(d, ['a', 'c']) # inplace deletion
print(d)
>>> {'b': [{'d': 10}, {}]}
This solution works for dict
and list
in a given nested dict
. The input to_delete
can be a list
of str
to be deleted or a single str
.
Plese note, that if you remove the only key in a dict
, you will get an empty dict
.
this works with dict
s containing Iterable
s (list
, …) that may contain dict
. Python 3. For Python 2 unicode
should also be excluded from the iteration. Also there may be some iterables that don’t work that I’m not aware of. (i.e. will lead to inifinite recursion)
from collections.abc import Iterable
def deep_omit(d, keys):
if isinstance(d, dict):
for k in keys:
d.pop(k, None)
for v in d.values():
deep_omit(v, keys)
elif isinstance(d, Iterable) and not isinstance(d, str):
for e in d:
deep_omit(e, keys)
return d
I came here to search for a solution to remove keys from deeply nested Python3 dicts and all solutions seem to be somewhat complex.
Here’s a oneliner for removing keys from nested or flat dicts:
nested_dict = {
"foo": {
"bar": {
"foobar": {},
"shmoobar": {}
}
}
}
>>> {'foo': {'bar': {'foobar': {}, 'shmoobar': {}}}}
nested_dict.get("foo", {}).get("bar", {}).pop("shmoobar", None)
>>> {'foo': {'bar': {'foobar': {}}}}
I used .get() to not get KeyError and I also provide empty dict as default value up to the end of the chain. I do pop() for the last element and I provide None as the default there to avoid KeyError.
If you have nested keys as well and based on @John La Rooy’s answer here is an elegant solution:
from boltons.iterutils import remap
def sof_solution():
data = {"user": {"name": "test", "pwd": "******"}, "accounts": ["1", "2"]}
sensitive = {"user.pwd", "accounts"}
clean = remap(
data,
visit=lambda path, key, value: drop_keys(path, key, value, sensitive)
)
print(clean)
def drop_keys(path, key, value, sensitive):
if len(path) > 0:
nested_key = f"{'.'.join(path)}.{key}"
return nested_key not in sensitive
return key not in sensitive
sof_solution() # prints {'user': {'name': 'test'}}
Since nobody posted an interactive version that could be useful for someone:
def delete_key_from_dict(adict, key):
stack = [adict]
while stack:
elem = stack.pop()
if isinstance(elem, dict):
if key in elem:
del elem[key]
for k in elem:
stack.append(elem[k])
This version is probably what you would push to production. The recursive version is elegant and easy to write but it scales badly (by default Python uses a maximum recursion depth of 1000).