Skip to content
Go to file

Latest commit


Failed to load latest commit information.
Latest commit message
Commit time

DeepDiff v 5.0.2

Downloads Python Versions License Build Status Coverage Status

DeepDiff Overview

  • DeepDiff: Deep Difference of dictionaries, iterables, strings and other objects. It will recursively look for all the changes.
  • DeepSearch: Search for objects within other objects.
  • DeepHash: Hash any object based on their content.

Tested on Python 3.5+ and PyPy3.

NOTE: Python 2 is not supported any more. DeepDiff v3.3.0 was the last version to support Python 2


Install from PyPi:

pip install deepdiff

DeepDiff prefers to use Murmur3 for hashing. However you have to manually install Murmur3 by running:

pip install 'deepdiff[murmur]'

Otherwise DeepDiff will be using SHA256 for hashing which is a cryptographic hash and is considerably slower.

If you are running into trouble installing Murmur3, please take a look at the Troubleshoot section.


>>> from deepdiff import DeepDiff  # For Deep Difference of 2 objects
>>> from deepdiff import grep, DeepSearch  # For finding if item exists in an object
>>> from deepdiff import DeepHash  # For hashing objects based on their contents

Deep Diff

DeepDiff gets the difference of 2 objects.

A few Examples

Note: This is just a brief overview of what DeepDiff can do. Please visit for full documentation.

List difference ignoring order or duplicates

>>> t1 = {1:1, 2:2, 3:3, 4:{"a":"hello", "b":[1, 2, 3]}}
>>> t2 = {1:1, 2:2, 3:3, 4:{"a":"hello", "b":[1, 3, 2, 3]}}
>>> ddiff = DeepDiff(t1, t2, ignore_order=True)
>>> print (ddiff)

Report repetitions

This flag ONLY works when ignoring order is enabled.

t1 = [1, 3, 1, 4]
t2 = [4, 4, 1]
ddiff = DeepDiff(t1, t2, ignore_order=True, report_repetition=True)

which will print you:

{'iterable_item_removed': {'root[1]': 3},
  'repetition_change': {'root[0]': {'old_repeat': 2,
                                    'old_indexes': [0, 2],
                                    'new_indexes': [2],
                                    'value': 1,
                                    'new_repeat': 1},
                        'root[3]': {'old_repeat': 1,
                                    'old_indexes': [3],
                                    'new_indexes': [0, 1],
                                    'value': 4,
                                    'new_repeat': 2}}}

Exclude certain types from comparison:

>>> l1 = logging.getLogger("test")
>>> l2 = logging.getLogger("test2")
>>> t1 = {"log": l1, 2: 1337}
>>> t2 = {"log": l2, 2: 1337}
>>> print(DeepDiff(t1, t2, exclude_types={logging.Logger}))

Exclude part of your object tree from comparison

>>> t1 = {"for life": "vegan", "ingredients": ["no meat", "no eggs", "no dairy"]}
>>> t2 = {"for life": "vegan", "ingredients": ["veggies", "tofu", "soy sauce"]}
>>> print (DeepDiff(t1, t2, exclude_paths={"root['ingredients']"}))

Exclude Regex Paths

You can also exclude using regular expressions by using exclude_regex_paths and pass a set or list of path regexes to exclude. The items in the list could be raw regex strings or compiled regex objects.

>>> t1 = [{'a': 1, 'b': 2}, {'c': 4, 'b': 5}]
>>> t2 = [{'a': 1, 'b': 3}, {'c': 4, 'b': 5}]
>>> print(DeepDiff(t1, t2, exclude_regex_paths={r"root\[\d+\]\['b'\]"}))
>>> exclude_path = re.compile(r"root\[\d+\]\['b'\]")
>>> print(DeepDiff(t1, t2, exclude_regex_paths=[exclude_path]))

Significant Digits

Digits after the decimal point. Internally it uses "{:.Xf}".format(Your Number) to compare numbers where X=significant_digits

>>> t1 = Decimal('1.52')
>>> t2 = Decimal('1.57')
>>> DeepDiff(t1, t2, significant_digits=0)
>>> DeepDiff(t1, t2, significant_digits=1)
{'values_changed': {'root': {'old_value': Decimal('1.52'), 'new_value': Decimal('1.57')}}}

Ignore Type Number - List that contains float and integer:

>>> from deepdiff import DeepDiff
>>> from pprint import pprint
>>> t1 = [1, 2, 3]
>>> t2 = [1.0, 2.0, 3.0]
>>> ddiff = DeepDiff(t1, t2)
>>> pprint(ddiff, indent=2)
{ 'type_changes': { 'root[0]': { 'new_type': <class 'float'>,
                         'new_value': 1.0,
                         'old_type': <class 'int'>,
                         'old_value': 1},
            'root[1]': { 'new_type': <class 'float'>,
                         'new_value': 2.0,
                         'old_type': <class 'int'>,
                         'old_value': 2},
            'root[2]': { 'new_type': <class 'float'>,
                         'new_value': 3.0,
                         'old_type': <class 'int'>,
                         'old_value': 3}}}
>>> ddiff = DeepDiff(t1, t2, ignore_type_in_groups=[(int, float)])


Starting with DeepDiff v 3, there are two different views into your diffed data: text view (original) and tree view (new).

Text View

Text view is the original and currently the default view of DeepDiff.

It is called text view because the results contain texts that represent the path to the data:

Example of using the text view.

>>> from deepdiff import DeepDiff
>>> t1 = {1:1, 3:3, 4:4}
>>> t2 = {1:1, 3:3, 5:5, 6:6}
>>> ddiff = DeepDiff(t1, t2)
>>> print(ddiff)
{'dictionary_item_added': {'root[5]', 'root[6]'}, 'dictionary_item_removed': {'root[4]'}}

So for example ddiff['dictionary_item_removed'] is a set if strings thus this is called the text view.

The following examples are using the *default text view.*
The Tree View is introduced in DeepDiff v3
and provides traversing capabilities through your diffed data and more!
Read more about the Tree View at the [tree view section](#tree-view) of this page.

Tree View

Starting the version v3 You can choose the view into the deepdiff results. The tree view provides you with tree objects that you can traverse through to find the parents of the objects that are diffed and the actual objects that are being diffed.

Value of an item has changed (Tree View)

>>> from deepdiff import DeepDiff
>>> from pprint import pprint
>>> t1 = {1:1, 2:2, 3:3}
>>> t2 = {1:1, 2:4, 3:3}
>>> ddiff_verbose0 = DeepDiff(t1, t2, verbose_level=0, view='tree')
>>> ddiff_verbose0
{'values_changed': {<root[2]>}}
>>> ddiff_verbose1 = DeepDiff(t1, t2, verbose_level=1, view='tree')
>>> ddiff_verbose1
{'values_changed': {<root[2] t1:2, t2:4>}}
>>> set_of_values_changed = ddiff_verbose1['values_changed']
>>> # since set_of_values_changed includes only one item in a set
>>> # in order to get that one item we can:
>>> (changed,) = set_of_values_changed
>>> changed  # Another way to get this is to do: changed=list(set_of_values_changed)[0]
<root[2] t1:2, t2:4>
>>> changed.t1
>>> changed.t2
>>> # You can traverse through the tree, get to the parents!
>>> changed.up
<root t1:{1: 1, 2: 2,...}, t2:{1: 1, 2: 4,...}>


In order to convert the DeepDiff object into a normal Python dictionary, use the to_dict() method. Note that to_dict will use the text view even if you did the diff in tree view.


>>> t1 = {1: 1, 2: 2, 3: 3, 4: {"a": "hello", "b": [1, 2, 3]}}
>>> t2 = {1: 1, 2: 2, 3: 3, 4: {"a": "hello", "b": "world\n\n\nEnd"}}
>>> ddiff = DeepDiff(t1, t2, view='tree')
>>> ddiff.to_dict()
{'type_changes': {"root[4]['b']": {'old_type': <class 'list'>, 'new_type': <class 'str'>, 'old_value': [1, 2, 3], 'new_value': 'world\n\n\nEnd'}}}

In order to do safe json serialization, use the to_json() method.


>>> t1 = {1: 1, 2: 2, 3: 3, 4: {"a": "hello", "b": [1, 2, 3]}}
>>> t2 = {1: 1, 2: 2, 3: 3, 4: {"a": "hello", "b": "world\n\n\nEnd"}}
>>> ddiff = DeepDiff(t1, t2, view='tree')
>>> ddiff.to_json()
'{"type_changes": {"root[4][\'b\']": {"old_type": "list", "new_type": "str", "old_value": [1, 2, 3], "new_value": "world\\n\\n\\nEnd"}}}'

Deep Search

DeepDiff comes with a utility to find the path to the item you are looking for. It is called DeepSearch and it has a similar interface to DeepDiff.

Let's say you have a huge nested object and want to see if any item with the word somewhere exists in it. Just grep through your objects as you would in shell!

from deepdiff import grep
obj = {"long": "somewhere", "string": 2, 0: 0, "somewhere": "around"}
ds = obj | grep("somewhere")

Which will print:

{'matched_paths': {"root['somewhere']"},
 'matched_values': {"root['long']"}}

And you can pass all the same kwargs as DeepSearch to grep too:

>>> obj | grep(item, verbose_level=2)
{'matched_paths': {"root['somewhere']": 'around'}, 'matched_values': {"root['long']": 'somewhere'}}

Deep Hash

(New in v4-0-0)

DeepHash is designed to give you hash of ANY python object based on its contents even if the object is not considered hashable! DeepHash is supposed to be deterministic in order to make sure 2 objects that contain the same data, produce the same hash.

Let's say you have a dictionary object.

>>> from deepdiff import DeepHash
>>> obj = {1: 2, 'a': 'b'}

If you try to hash it:

>>> hash(obj)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: unhashable type: 'dict'

But with DeepHash:

>>> from deepdiff import DeepHash
>>> obj = {1: 2, 'a': 'b'}
>>> DeepHash(obj)
{4355639248: 2468916477072481777512283587789292749, 4355639280: -35787773492556653776377555218122431491, 4358636128: -88390647972316138151822486391929534118, 4358009664: 8833996863197925870419376694314494743, 4357467952: 34150898645750099477987229399128149852}

So what is exactly the hash of obj in this case? DeepHash is calculating the hash of the obj and any other object that obj contains. The output of DeepHash is a dictionary of object IDs to their hashes. In order to get the hash of obj itself, you need to use the object (or the id of object) to get its hash:

>>> hashes = DeepHash(obj)
>>> hashes[obj]

Which you can write as:

>>> hashes = DeepHash(obj)[obj]

At first it might seem weird why DeepHash(obj)[obj] but remember that DeepHash(obj) is a dictionary of hashes of all other objects that obj contains too.

Using DeepDiff in unit tests

result is the output of the function that is being tests. expected is the expected output of the function.

self.assertEqual(DeepDiff(expected, result), {})

or if you are using Pytest:

assert not DeepDiff(expected, result)

In other words, assert that there is no diff between the expected and the result.

Difference with Json Patch

Unlike Json Patch which is designed only for Json objects, DeepDiff is designed specifically for almost all Python types. In addition to that, DeepDiff checks for type changes and attribute value changes that Json Patch does not cover since there are no such things in Json. Last but not least, DeepDiff gives you the exact path of the item(s) that were changed in Python syntax.

Example in Json Patch for replacing:

{ "op": "replace", "path": "/a/b/c", "value": 42 }

Example in DeepDiff for the same operation:

>>> item1 = {'a':{'b':{'c':'foo'}}}
>>> item2 = {'a':{'b':{'c':42}}}
>>> DeepDiff(item1, item2)
{'type_changes': {"root['a']['b']['c']": {'old_type': <type 'str'>, 'new_value': 42, 'old_value': 'foo', 'new_type': <type 'int'>}}}

Pycon 2016

I was honored to give a talk about how DeepDiff does what it does at Pycon 2016. Please check out the video and let me know what you think:

Diff It To Dig It Video And here is more info:




Failed to build mmh3 when installing DeepDiff

DeepDiff prefers to use Murmur3 for hashing. However you have to manually install murmur3 by running: pip install mmh3

On MacOS Mojave some user experience difficulty when installing Murmur3.

The problem can be solved by running:

xcode-select --install

And then running

pip install mmh3


Please take a look at the changelog file.


We use bump2version to bump and tag releases.

git checkout master && git pull
bumpversion {patch|minor|major}
git push && git push --tags


  1. Please make your PR against the dev branch
  2. Please make sure that your PR has tests. Since DeepDiff is used in many sensitive data driven projects, we strive to maintain around 100% test coverage on the code.

Please run pytest --cov=deepdiff to see the coverage report.

Or to see a more user friendly version, please run: pytest --cov=deepdiff --cov-report term-missing.

Thank you!


Authors listed in the order of the contributions:

  • Sep Dehpour (Seperman)
  • Victor Hahn Castell for the tree view and major contributions:
  • nfvs for Travis-CI setup script.
  • brbsix for initial Py3 porting.
  • WangFenjin for unicode support.
  • timoilya for comparing list of sets when ignoring order.
  • Bernhard10 for significant digits comparison.
  • b-jazz for PEP257 cleanup, Standardize on full names, fixing line endings.
  • finnhughes for fixing slots
  • moloney for Unicode vs. Bytes default
  • serv-inc for adding help(deepdiff)
  • movermeyer for updating docs
  • maxrothman for search in inherited class attributes
  • maxrothman for search for types/objects
  • MartyHub for exclude regex paths
  • sreecodeslayer for DeepSearch match_string
  • Brian Maissy brianmaissy for weakref fix, enum tests
  • Bartosz Borowik boba-2 for Exclude types fix when ignoring order
  • Brian Maissy brianmaissy for fixing classes which inherit from classes with slots didn't have all of their slots compared
  • Juan Soler Soleronline for adding ignore_type_number
  • mthaddon for adding timedelta diffing support
  • Necrophagos for Hashing of the number 1 vs. True
  • gaal-dev for adding exclude_obj_callback
  • Ivan Piskunov van-ess0 for deprecation warning enhancement.
  • Michał Karaś MKaras93 for the pretty view
  • Christian Kothe chkothe for the basic support for diffing numpy arrays
  • Timothy for truncate_datetime
  • d0b3rm4n for bugfix to not apply format to non numbers.
  • MyrikLD for Bug Fix NoneType in ignore type groups
You can’t perform that action at this time.