This is a pre-production deployment of Warehouse. Changes made here affect the production instance of PyPI (
Help us improve Python packaging - Donate today!

workflow support for reproducible deduplication and merging

Project Description


workflow support for reproducible deduplication and merging

Say you have a bunch of strings, and some of them are different but refer to the same thing. Maybe it’s just some long list, maybe it’s the contents of two key columns from data sets that you’d like to merge.

David Copperfield
Lance Burton
Dave Copperfield
Levar Burton

Here’s what you can do with mergic:

Give mergic all your identifiers, one per line. If they are in a file called originals.txt:

mergic calc originals.txt

You will see output about the possible groupings mergic can produce based on its default distance function. (It’s easy to use a custom distance function—see below.)

num groups, max group, num pairs, cutoff
         4,         1,         0, -0.909090909091
         3,         2,         1, 0.0909090909091
         2,         2,         2, 0.25
         1,         4,         6, 0.714285714286

There are 4 groupings that mergic can make with the given distance function, ranging from a group for every name to a single group for all four names.

The num groups is the number of groups that mergic can make with the given cutoff.

The max group is the number of items in the largest group for that grouping.

The num pairs column indicates the number of within-group links that correspond to the grouping. In some sense this represents how much work it is to check all the linkages that mergic is making. Thinking in groups is much better than thinking in individual pairwise comparisons, but the metric is useful for summarizing how much linking is happening.

The cutoff determines which pairs of items are put in the same group. If the distance between two items is equal to or less than the cutoff, those items will be grouped together.

Select a cutoff to produce the grouping you would like to see. If you would like to use the cutoff 0.3 and put the results in a file called grouping.json:

mergic make originals.txt 0.3 > grouping.json

Now grouping.json contains a grouping of your original data, specified in JSON. It’s designed to be human-readable, and human-editable, so you can check it and improve it as needed. The largest groups will be at the top, and similar items will be near one another.

    "Lance Burton": [
        "Lance Burton",
        "Levar Burton"
    "David Copperfield": [
        "Dave Copperfield",
        "David Copperfield"

There are two proposed groups, with proposed names “Lance Burton” and “David Copperfield”. You probably want to copy the produced file and edit the copy.

cp grouping.json grouping_fixed.json
# edit `grouping_fixed.json`

You can edit the proposed group names—the keys of the object—if you like. The original values from the data are in the array values of the object, and you can move them around to correct the grouping. In this case you could also re-run mergic with a cutoff of 0.1 to get a correct grouping:

    "David Copperfield": [
        "Dave Copperfield",
        "David Copperfield"
    "Lance Burton": [
        "Lance Burton"
    "Levar Burton": [
        "Levar Burton"

Now that grouping_fixed.json is as perfect as it can be, you can move forward.

You can also compare your two JSON grouping files and see what you changed:

mergic diff grouping.json grouping_fixed.json > diff.json

Now the file diff.json contains just what’s different between grouping.json and grouping_fixed.json. The mergic diff command is analogous to regular diff for text files, but it is aware of the JSON partition format so it can capture changes intelligently.

You can apply changes to a mergic-produced file to recover your edited version.

mergic apply grouping.json diff.json > grouping_new.json

Now grouping_new.json is equivalent to grouping_fixed.json, as you can verify:

mergic diff grouping_fixed.json grouping_new.json
# {}  // (no changes)

In this way you have a complete and verifiable record of your work, at the level of whole files and also at the level of changes made by hand.

The JSON grouping format is very convenient for humans, but for tabular data a merge table is more useful. A merge table has one column with the original values from your data and one column with the new keys. These are named original and mergic in the output:

mergic table grouping_fixed.json > merge_table.csv

The file merge_table.csv looks like this:

Lance Burton,Lance Burton
Levar Burton,Levar Burton
Dave Copperfield,David Copperfield
David Copperfield,David Copperfield

This merge table can now be used with any tabular data system. For merges, first merge it on to both tables and then merge by the mergic key. For deduplication, merge it on to the table(s) of interest and then use the mergic column as you would have used the original data.


pip install mergic

Using a Custom Distance Function

The mergic package provides a command-line script called mergic that uses Python’s built-in difflib.SequenceMatcher.ratio() for calculating string distances, but a major strength of mergic is that it enables easy customization of the distance function via the mergic.Blender class. Making a custom mergic script is as easy as:

#!/usr/bin/env python
import mergic

# Any custom distance you want to try! e.g.,
def my_distance(a, b):
    return abs(len(a) - len(b))


Now can be used just like the standard mergic script!

There is also a function that determines how to make a key for a group. It takes a list and returns a string. By default mergic.Blender will use the longest of a group’s values (the first longest, in stable sorted order). You can also supply a custom function for generating keys.

Distances You Might Like

Levenshtein string edit distance: The classic! It has many implementations; one of them is python-Levenshtein.

# pip install python-Levenshtein
import Levenshtein
Levenshtein.distance("fuzzy", "wuzzy")
# 1

SeatGeek’s fuzzywuzzy: As described in their blog post, some people have found these variants to work well in practice. Responses from fuzzywuzzy are phrased as integer percent similarities; one way to make a distance is to subtract from 100.

# pip install fuzzywuzzy
from fuzzywuzzy import fuzz
100 - fuzz.ratio("Levensthein", "Leviathan")
# 50

There are a ton of distances, even just within the two packages mentioned!

You can also make your own!

Release History

Release History

This version
History Node


History Node


History Node


History Node

History Node


History Node


History Node


History Node


Download Files

Download Files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

File Name & Checksum SHA256 Checksum Help Version File Type Upload Date
mergic-0.0.7.tar.gz (8.7 kB) Copy SHA256 Checksum SHA256 Source Jul 5, 2015

Supported By

WebFaction WebFaction Technical Writing Elastic Elastic Search Pingdom Pingdom Monitoring Dyn Dyn DNS Sentry Sentry Error Logging CloudAMQP CloudAMQP RabbitMQ Heroku Heroku PaaS Kabu Creative Kabu Creative UX & Design Fastly Fastly CDN DigiCert DigiCert EV Certificate Rackspace Rackspace Cloud Servers DreamHost DreamHost Log Hosting