- DeepDiff: Deep Difference of dictionaries, iterables, strings, and ANY other object.
- DeepSearch: Search for objects within other objects.
- DeepHash: Hash any object based on their content.
- Delta: Store the difference of objects and apply them to other objects.
- Extract: Extract an item from a nested Python object using its path.
- commandline: Use DeepDiff from commandline.
Tested on Python 3.8+ and PyPy3.
Please check the ChangeLog file for the detailed information.
DeepDiff 8-0-1
- Bugfix. Numpy should be optional.
DeepDiff 8-0-0
With the introduction of threshold_to_diff_deeper
, the values returned are different than in previous versions of DeepDiff. You can still get the older values by setting threshold_to_diff_deeper=0
. However to signify that enough has changed in this release that the users need to update the parameters passed to DeepDiff, we will be doing a major version update.
use_enum_value=True
makes it so when diffing enum, we use the enum's value. It makes it so comparing an enum to a string or any other value is not reported as a type change.threshold_to_diff_deeper=float
is a number between 0 and 1. When comparing dictionaries that have a small intersection of keys, we will report the dictionary as anew_value
instead of reporting individual keys changed. If you set it to zero, you get the same results as DeepDiff 7.0.1 and earlier, which means this feature is disabled. The new default is 0.33 which means if less that one third of keys between dictionaries intersect, report it as a new object.- Deprecated
ordered-set
and switched toorderly-set
. Theordered-set
package was not being maintained anymore and starting Python 3.6, there were better options for sets that ordered. I forked one of the new implementations, modified it, and published it asorderly-set
. - Added
use_log_scale:bool
andlog_scale_similarity_threshold:float
. They can be used to ignore small changes in numbers by comparing their differences in logarithmic space. This is different than ignoring the difference based on significant digits. - json serialization of reversed lists.
- Fix for iterable moved items when
iterable_compare_func
is used. - Pandas and Polars support.
DeepDiff 7-0-1
- Fixes the translation between Difflib opcodes and Delta flat rows.
DeepDiff 7-0-0
- DeepDiff 7 comes with an improved delta object. Delta to flat dictionaries have undergone a major change. We have also introduced Delta serialize to flat rows.
- Subtracting delta objects have dramatically improved at the cost of holding more metadata about the original objects.
- When
verbose=2
, and the "path" of an item has changed in a report between t1 and t2, we include it asnew_path
. path(use_t2=True)
returns the correct path to t2 in any reported change in thetree view
- Python 3.7 support is dropped and Python 3.12 is officially supported.
DeepDiff 6-7-1
- Support for subtracting delta objects when iterable_compare_func is used.
- Better handling of force adding a delta to an object.
- Fix for
Can't compare dicts with both single and double quotes in keys
- Updated docs for Inconsistent Behavior with math_epsilon and ignore_order = True
DeepDiff 6-7-0
- Delta can be subtracted from other objects now.
- verify_symmetry is deprecated. Use bidirectional instead.
- always_include_values flag in Delta can be enabled to include values in the delta for every change.
- Fix for Delta.add breaks with esoteric dict keys.
- You can load a delta from the list of flat dictionaries.
DeepDiff 6-6-1
- Fix for DeepDiff raises decimal exception when using significant digits
- Introducing group_by_sort_key
- Adding group_by 2D. For example
group_by=['last_name', 'zip_code']
pip install deepdiff
If you want to use DeepDiff from commandline:
pip install "deepdiff[cli]"
If you want to improve the performance of DeepDiff with certain functionalities such as improved json serialization:
pip install "deepdiff[optimize]"
Install optional packages:
- yaml
- tomli (python 3.10 and older) and tomli-w for writing
- clevercsv for more rubust CSV parsing
- orjson for speed and memory optimized parsing
- pydantic
https://zepworks.com/deepdiff/current/
👋 Hi there,
Thank you for using DeepDiff! As an engineer, I understand the frustration of wrestling with unruly data in pipelines. That's why I developed a new tool - Qluster to empower non-engineers to control and resolve data issues at scale autonomously and stop bugging the engineers! 🛠️
If you are going through this pain now, I would love to give you early access to Qluster and get your feedback.
Please take a look at the CHANGELOG file.
📣 Please fill out our fast 5-question survey so that we can learn how & why you use DeepDiff, and what improvements we should make. Thank you! 👯
- Please make your PR against the dev branch
- 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 --runslow
to see the coverage report. Note that the --runslow
flag will run some slow tests too. In most cases you only want to run the fast tests which so you wont add the --runslow
flag.
Or to see a more user friendly version, please run: pytest --cov=deepdiff --cov-report term-missing --runslow
.
Thank you!
Please take a look at the AUTHORS file.