🆔 Command line tool for deduplicating CSV files
Python
Clone or download
Latest commit e4ed504 Feb 11, 2018
Permalink
Failed to load latest commit information.
csvdedupe check conflicting args earlier Feb 12, 2018
docs Update index.rst Mar 14, 2016
examples Update config.json.example Mar 14, 2016
tests fix tests Jul 27, 2015
.gitignore updated documentation Jan 22, 2014
.travis.yml bump version Feb 12, 2018
LICENSE.md Update LICENSE.md Sep 10, 2013
README.md update links Feb 12, 2018
requirements-test.txt fix tests Jul 27, 2015
setup.cfg handle scores for clusters Sep 13, 2016
setup.py bump version Feb 12, 2018

README.md

csvdedupe

Command line tools for using the dedupe python library for deduplicating CSV files.

Two easy commands:

csvdedupe - takes a messy input file or STDIN pipe and identifies duplicates.

csvlink - takes two CSV files and finds matches between them.

Read more about csvdedupe on OpenNews Source

Build Status

Installation and dependencies

pip install csvdedupe

Getting Started

csvdedupe

csvdedupe takes a messy input file or STDIN pipe and identifies duplicates. To get started, pick one of three deduping strategies: call csvdedupe with arguments, pipe your file using UNIX, or define a config file.

Provide an input file, field names, and output file:

csvdedupe examples/csv_example_messy_input.csv \
          --field_names "Site name" Address Zip Phone \
          --output_file output.csv

or

Pipe it, UNIX style:

cat examples/csv_example_messy_input.csv | csvdedupe --skip_training \
          --field_names "Site name" Address Zip Phone > output.csv

or

Define everything in a config file:

csvdedupe examples/csv_example_messy_input.csv \
            --config_file=config.json

Your config file may look like this:

{
  "field_names": ["Site name", "Address", "Zip", "Phone"],
  "field_definition" : [{"field" : "Site name", "type" : "String"},
                        {"field" : "Address", "type" : "String"},
                        {"field" : "Zip", "type" : "String",
                         "Has Missing" : true},
                        {"field" : "Phone", "type" : "String",
                         "Has Missing" : true}],
  "output_file": "examples/output.csv",
  "skip_training": false,
  "training_file": "training.json",
  "sample_size": 150000,
  "recall_weight": 2
}

To use csvdedupe you absolutely need:

  • input a CSV file name or piped CSV file to deduplicate

Either

  • --config_file Path to configuration file.

Or

  • --field_names List of column names for dedupe to pay attention to

You may also need:

  • --output_file OUTPUT_FILE CSV file to store deduplication results (default: None)
  • --destructive Output file will contain unique records only
  • --skip_training Skip labeling examples by user and read training from training_file only (default: False)
  • --training_file TRAINING_FILE Path to a new or existing file consisting of labeled training examples (default: training.json)
  • --sample_size SAMPLE_SIZE Number of random sample pairs to train off of (default: 150000)
  • --recall_weight RECALL_WEIGHT Threshold that will maximize a weighted average of our precision and recall (default: 2)
  • -d, --delimiter Delimiting character of the input CSV file (default: ,)
  • -h, --help show help message and exit

csvlink

csvdedupe takes two CSV files and finds matches between them.

Provide an input file, field names, and output file:

csvlink examples/restaurant-1.csv examples/restaurant-2.csv \
            --field_names name address city cuisine \
            --output_file output.csv

or

Line up different field names from each file:

csvlink examples/restaurant-1.csv examples/restaurant-2.csv \
            --field_names_1 name address city cuisine \
            --field_names_2 restaurant street city type \
            --output_file output.csv

or

Pipe the output to STDOUT:

csvlink examples/restaurant-1.csv examples/restaurant-2.csv \
            --field_names name address city cuisine \
            > output.csv

or

Define everything in a config file:

csvlink examples/restaurant-1.csv examples/restaurant-2.csv \
              --config_file=config.json

Your config file may look like this:

{
  "field_names_1": ["name", "address", "city", "cuisine"],
  "field_names_2": ["restaurant", "street", "city", "type"],
  "field_definition" : [{"field" : "name", "type" : "String"},
                        {"field" : "address", "type" : "String"},
                        {"field" : "city", "type" : "String",
                         "Has Missing" : true},
                        {"field" : "cuisine", "type" : "String",
                         "Has Missing" : true}],
  "output_file": "examples/output.csv",
  "skip_training": false,
  "training_file": "training.json",
  "sample_size": 150000,
  "recall_weight": 2
}

To use csvlink you absolutely need:

  • input two CSV file names to join together

Either

  • --config_file Path to configuration file.

Or

  • --field_names_1 List of column names in first file for dedupe to pay attention to
  • --field_names_2 List of column names in second file for dedupe to pay attention to

You may also need:

  • --output_file OUTPUT_FILE CSV file to store deduplication results (default: None)
  • --inner_join Only return matches between datasets
  • --skip_training Skip labeling examples by user and read training from training_file only (default: False)
  • --training_file TRAINING_FILE Path to a new or existing file consisting of labeled training examples (default: training.json)
  • --sample_size SAMPLE_SIZE Number of random sample pairs to train off of (default: 150000)
  • --recall_weight RECALL_WEIGHT Threshold that will maximize a weighted average of our precision and recall (default: 2)
  • -d, --delimiter Delimiting character of the input CSV file (default: ,)
  • -h, --help show help message and exit

Training

The secret sauce of csvdedupe is human input. In order to figure out the best rules to deduplicate a set of data, you must give it a set of labeled examples to learn from.

The more labeled examples you give it, the better the deduplication results will be. At minimum, you should try to provide 10 positive matches and 10 negative matches.

The results of your training will be saved in a JSON file ( training.json, unless specified otherwise with the --training-file option) for future runs of csvdedupe.

Here's an example labeling operation:

Phone :  2850617
Address :  3801 s. wabash
Zip :
Site name :  ada s. mckinley st. thomas cdc

Phone :  2850617
Address :  3801 s wabash ave
Zip :
Site name :  ada s. mckinley community services - mckinley - st. thomas

Do these records refer to the same thing?
(y)es / (n)o / (u)nsure / (f)inished

Output

csvdedupe attempts to identify all the rows in the csv that refer to the same thing. Each group of such records are called a cluster. csvdedupe returns your input file with an additional column called Cluster ID, that either is the numeric id (zero-indexed) of a cluster of grouped records or an x if csvdedupe believes the record doesn't belong to any cluster.

csvlink operates in much the same way as csvdedupe, but will flatten both CSVs in to one output file similar to a SQL OUTER JOIN statement. You can use the --inner_join flag to exclude rows that don't match across the two input files, much like an INNER JOIN.

Preprocessing

csvdedupe attempts to convert all strings to ASCII, ignores case, new lines, and padding whitespace. This is all probably uncontroversial except the conversion to ASCII. Basically, we had to choose between two ways of handling extended characters.

distance("Tomas", "Tomás')  = distance("Tomas", "Tomas")

or

distance("Tomas, "Tomás") = distance("Tomas", "Tomzs")

We chose the first option. While it is possible to do something more sophisticated, this option seems to work pretty well for Latin alphabet languages.

Testing

Unit tests of core csvdedupe functions

pip install -r requirements-test.txt
nosetests

Community

Recipes

Combining and deduplicating files from different sources.

Lets say we have a few sources of early childhood programs in Chicago and we'd like to get a canonical list. Let's do it with csvdedupe, csvkit, and some other common command line tools.

Alignment and stacking

Our first task will be to align the files and have the same data in the same columns for stacking.

First, let's look at the headers of the files.

File 1

> head -1 CPS_Early_Childhood_Portal_Scrape.csv
"Site name","Address","Phone","Program Name","Length of Day"

File 2

> head -1 IDHS_child_care_provider_list.csv
"Site name","Address","Zip Code","Phone","Fax","IDHS Provider ID"

So, we'll have to add "Zip Code", "Fax", and "IDHS Provider ID" to CPS_Early_Childhood_Portal_Scrape.csv, and we'll have to add "Program Name", "Length of Day" to IDHS_child_care_provider_list.csv.

> cd examples
> sed '1 s/$/,"Zip Code","Fax","IDHS Provider ID"/' CPS_Early_Childhood_Portal_Scrape.csv > input_1a.csv
> sed '2,$s/$/,,,/' input_1a.csv > input_1b.csv
> sed '1 s/$/,"Program Name","Length of Day"/' IDHS_child_care_provider_list.csv > input_2a.csv
> sed '2,$s/$/,,/' input_2a.csv > input_2b.csv

Now, we reorder the columns in the second file to align to the first.

> csvcut -c "Site name","Address","Phone","Program Name","Length of Day","Zip Code","Fax","IDHS Provider ID" \
         input_2b.csv > input_2c.csv

And we are finally ready to stack.

> csvstack -g CPS_Early_Childhood_Portal_Scrape.csv,IDHS_child_care_provider_list.csv \
           -n source \
           input_1b.csv input_2c.csv > input.csv

Dedupe it!

And now we can dedupe

> cat input.csv | csvdedupe --field_names "Site name" Address "Zip Code" Phone > output.csv

Let's sort the output by duplicate IDs, and we are ready to open it in your favorite spreadsheet program.

> csvsort -c "Cluster ID" output.csv > sorted.csv

Errors and Bugs

If something is not behaving intuitively, it is a bug, and should be reported. Report it here.

Patches and Pull Requests

We welcome your ideas! You can make suggestions in the form of github issues (bug reports, feature requests, general questions), or you can submit a code contribution via a pull request.

How to contribute code:

  • Fork the project.
  • Make your feature addition or bug fix.
  • Send us a pull request with a description of your work! Don't worry if it isn't perfect - think of a PR as a start of a conversation, rather than a finished product.

Copyright and Attribution

Copyright (c) 2016 DataMade. Released under the MIT License.