Difference between revisions of "Reusing Open Food Facts Data"
Raphael0202 (talk | contribs) |
Raphael0202 (talk | contribs) |
||
(One intermediate revision by the same user not shown) | |||
Line 198: | Line 198: | ||
==== DuckDB to query the database ==== | ==== DuckDB to query the database ==== | ||
Here again, this great tool allows to request remote parquet files with the command line. | Here again, this great tool allows to request remote parquet files with the command line. | ||
− | $ duckdb :memory: "SELECT * from 'https://huggingface.co/datasets/openfoodfacts/product-database/resolve/main/food | + | $ duckdb :memory: "SELECT * from 'https://huggingface.co/datasets/openfoodfacts/product-database/resolve/main/food.parquet' LIMIT 10;" |
The request can be a bit long (~15 seconds). | The request can be a bit long (~15 seconds). | ||
Line 208: | Line 208: | ||
==== How to exploit the Parquet database using DuckDB: Use-Cases ==== | ==== How to exploit the Parquet database using DuckDB: Use-Cases ==== | ||
− | |||
− | + | Look at our [[DuckDB Cheatsheet]] to see DuckDB queries corresponding to use-cases of the community. | |
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
=== MongoDB dump === | === MongoDB dump === |
Latest revision as of 09:21, 26 November 2024
Open Food Facts data is released as Open Data: it can be reused freely by anyone, under the Open Database License (ODBL). While this page is related to practical reuse, you must really be aware of rights and duties provided by the Open Database License (ODBL).
Where is the data?
You'll find different kind of ways to get the data.
Looking for a selection of products?
Then use the advanced search. The Open Food Facts advanced search feature allows to download selections of the data. See: https://world.openfoodfacts.org/cgi/search.pl
When your search is done, you will be able to download the selection in CSV or Excel format, just give a try!
Do notice you can download up to 10,000 results only. If you need to download more results you have to use some other methods described below.
Looking for the whole database?
The whole database can be downloaded at https://world.openfoodfacts.org/data
It's very big. Open Food Facts hosts more than 2,800,000 products (as of April 2023). So you will probably need skills to reuse the data.
You'll be able to find here different kinds of data.
The MongoDB daily export
It represents the most complete data; it's very big and you have to know how to deal with MongoDB. It's very big! More than 30GB uncompressed.
The JSONL daily export
There is a daily export of the whole database in JSONL format (sometimes called LDJSON or NDJSON) where each line is a JSON object. It represents the same data as the MongoDB export. The file is 4,8GB (2022-10), compressed with gzip. It takes more than 14GB uncompressed.
You can find it at https://static.openfoodfacts.org/data/openfoodfacts-products.jsonl.gz
The CSV daily export
It contains all the products, but with a subset of the database fields. This subset is very large and include main characteristics (EAN, name, brand...), many tags (such as categories, origins, labels, packaging...), ingredients and nutrition facts. Thus, it is generally fitted to the majority of usages. It's a 2.3GB file (as of April 2020), so it can't be opened by Libre Office or Excel with an 8GB machine.
How to reuse?
CSV daily export
csvkit tips
csvkit is a very efficient tool to manipulate huge amounts of CSV data. Here are some useful tips to manipulate Open Food Facts CSV export.
Converting whole Open Food Facts "CSV" export to regular CSV. Open Food Facts export use tabs as separator: it should be called TSV (tab separated values) instead of CSV (comma separated values). csvkit
can convert TSV file into CSV very easily:
$ csvclean -t en.openfoodfacts.org.products.csv > myCSV.csv
Selecting 2 columns. Selecting two or three columns can be useful for some usages. Extracting two columns produce a smaller CSV file which can be opened by common softwares such as Libre Office or Excel. The following command creates a CSV file (brands.csv) containing two columns from Open Food Facts (code and brands). (It generally takes more than 2 minutes, depending on your computer.)
$ csvcut -t -c code,brands en.openfoodfacts.org.products.csv > brands.csv
Selecting products based on a regular expression. csvkit can search in some specified fields, allowing to make powerful selections. The following command creates a CSV file (selection.csv) containing all products where the barcode (code) is beginning by 325798 (-r "^325798(.*)"
).
$ csvgrep -t -c code -r "^325798(.*)" en.openfoodfacts.org.products.csv > selection.csv
The following command creates a CSV file (calissons.csv) containing all products where the category (categories) is containing "calisson".
$ csvgrep -t -c categories -r "calisson" en.openfoodfacts.org.products.csv > calisson.csv
Import CSV in PostgreSQL
See this article: https://blog-postgresql.verite.pro/2018/12/21/import-openfoodfacts.html (in french, but should be understandable with Google Translator).
Alternative way - feel free to use a project from github: https://github.com/ArchiMageAlex/off_converter
Import CSV to SQLite
The repository foodrescue-content contains Ruby scripts that import Open Food Facts CSV data into a SQLite database with full table normalization. Only a few fields are imported so far, but this can be extended easily. Data imported so far includes:
- barcode number
- product name
- product categories
- product countries
- full categories hierarchy imported from the
categories.txt
taxonomy (see)
Import CSV to DuckDB
DuckDB is very close to SQLite, except it has higher performances: database size is 3 times lighter, and requests performs 5-10 times better.
# Discard invalid characters # duckdb doesn't like invalid UTF8. It did not want to read some parquet file as such, with the following error: # Error: near line 1: Invalid Input Error: Invalid string encoding found in Parquet file: value "........." # (occuring namely on this product: https://world.openfoodfacts.org/product/9900109008673?rev=4 ) # The issue, and its solution below, seems to be well-known: https://til.simonwillison.net/linux/iconv iconv -f utf-8 -t utf-8 -c en.openfoodfacts.org.products.csv -o en.openfoodfacts.org.products.converted.csv # Create duckdb database and import data duckdb products.db <<EOF CREATE TABLE products AS SELECT * FROM read_csv_auto('en.openfoodfacts.org.products.converted.csv', quote='', sample_size=3000000, delim='\t'); EOF # Then you can try a SQL request duckdb products.db -csv <<EOF SELECT * FROM products WHERE completeness > 0.99 -- products with a good level of completeness ORDER BY last_modified_datetime LIMIT 10; EOF
We wrote a detailled article about using DuckDB with the Open Food Facts database. Check it out!
Python
There are some articles dealing with using Python language to explore Open Food Facts data.
Step by step commands: http://www.xavierdupre.fr/app/ensae_teaching_cs/helpsphinx/notebooks/prepare_data_2017.html (also in french)
Python notebooks are great to learn Open Food Facts data, as they mix code and results together:
- Find dozens of python notebooks on Kaggle
- https://www.datasciencesociety.net/part-1-exploring-food-data/
R stat
For people who have R stat skills, there are more than 50 notebooks from Kaggle community.
Moreover, here a link to transform .bson file to a dataframe: https://github.com/gnaweric/openfoodfact_database_queries
With the use of {mongolite}, first connect to the base, then import the .bson file, then get a sample of it to make sure it is ready. Finally save it to a .rdata file for example.
Beware, each line is a product and some variable need to be unnest: tidyverser::unnest_wider()
JSONL delta exports
Every day, Open Food Facts exports all the products created during the last 24 hours. The documentation of this export can be found in the /data page.
If you don't have MongoDB and just want to use these delta exports to build an up-to-date database, you can merge each export with the help of jq
tool.
$ gunzip products_1638076899_1638162314.json.gz # will decompress the file $ wc -l products_1638076899_1638162314.json # will count the number of products in this export (in JSONL each line is a JSON object) $ jq -c '. + .' 2021-11-30.json products_1638162314_1638248379.json > 2021-12-01.json # merge the delta with previous complete data
JSONL export
JSONL is a huge file! It's not possible to play with it with common editors or common tools. There is some command line tools that allows interesting things, like DuckDB or jq.
DuckDB
DuckDB recognizes the compressed JSONL file, there is no need to uncompress it. You can start playing with JSONL export with very few commands.
At first create a DuckDB file:
$ duckdb off.db
Then you can start playing with SQL commands.
SELECT code, product_name, countries_en FROM read_ndjson('openfoodfacts-products.jsonl.gz', ignore_errors=True) WHERE countries_en LIKE "%Germany%" LIMIT 10;
The advantages of DuckDB over the other solutions are that it is straightfoward and very fast.
If you want to go further, you can check the blog article from Jeremy Arancio.
jq
- start decompress the file (be careful => 14GB after decompression):
$ gunzip openfoodfacts-products.jsonl.gz
- work on a small subset to test. E.g. for 100 products:
$ head -n 100 openfoodfacts-products.jsonl > small.jsonl
You can start playing with jq. Here are examples.
$ cat small.jsonl | jq . # print all file in JSON format
$ cat small.jsonl | jq -r .code # print all products' codes.
$ cat small.jsonl | jq -r '[.code,.product_name] | @csv' # output CSV data containing code,product_name
Then you can try on the whole database:
$ cat openfoodfacts-products.jsonl | jq -r '[.code,.product_name] | @csv' > names.csv # output CSV file (name.csv) containing all products with code,product_name
If you don't have enough disk space to uncompress the .gz file, you can use zcat directly on the compressed file. Example:
$ zcat openfoodfacts-products.jsonl.gz | jq -r '[.code,.product_name] | @csv' # output CSV data containing code,product_name
Filtering JSONL export with jq
Filtering a specific country:
$ zcat openfoodfacts-products.jsonl.gz | jq '. | select(.countries_tags[]? == "en:germany")'
The previous command produces a json output containing all the products sold in Germany. If you want a JSONL output, add -c parameter.
$ zcat openfoodfacts-products.jsonl.gz | jq -c '. | select(.countries_tags[]? == "en:germany")'
You can add multiple filters and export the result to a CSV file. For example, here is a command that 1. selects products having the Nutri-Score computed and belonging to the TOP 90% most scanned products in 2020, and 2. exports barcode (code
) and number of scans (scans_n
) as a CSV file.
$ zcat openfoodfacts-products.jsonl.gz | jq -r '. | select(.misc_tags[]? == "en:nutriscore-computed" and .popularity_tags[]? == "top-90-percent-scans-2020") | [.code,.scans_n] | @csv' > displayed.ns.in.top90.2020.world.csv
Filtering barcodes which are different from a code containing 1 to 13 digits:
$ zcat openfoodfacts-products.jsonl.gz | jq -r '. | select(.code|test("^[0-9]{1,13}$") | not) | .code' > ean_gt_13.csv
Some part of the data are arrays, you must aggregate them using join
for CSV export. For example, to export each product and its states in CSV:
$ zcat openfoodfacts-products.jsonl.gz | jq -r '[.code,(.states_tags|join(","))] | @csv'
Selecting products with quality issues and exporting the barcode and the issues in CSV:
$ zcat openfoodfacts-products.jsonl.gz | jq -r '. | select(.data_quality_errors_tags[]? != "")' | jq -r '[.code,(.data_quality_errors_tags|join(","))] | @csv'
These operations can be quite long (more than 10 minutes depending on your computer and your selection).
Parquet file hosted on Hugging Face (beta)
The Parquet format is currently in beta, meaning it shouldn't be used in production yet.
A simplified version of the JSONL export (the whole database) is also available in the Parquet format. In the conversion process, we filtered out columns that contain duplicate information, are intended for internal debugging, or are not relevant to users.
The Parquet format has proved to be handy:
- Data are organized by column, rather than by row, which saves storage space and speeds up analytics queries, i.e. you can select just the columns you care about, optimizing query performances, even on entry-level computers,
- Highly efficient data compression and decompression, making it good for storing and sharing big data of any kind,
- Supports complex data types and advanced nested data structures.
The dataset is available on the Hugging Face platform, a collaborative Machine Learning ecosystem where developers and researchers can share models and datasets.
In-browser queries
Just go the dataset's page -- https://huggingface.co/datasets/openfoodfacts/product-database -- and click on the "SQL" yellow button.
You'll see a SQL interface in your browser, where you can perform queries.
DuckDB to query the database
Here again, this great tool allows to request remote parquet files with the command line.
$ duckdb :memory: "SELECT * from 'https://huggingface.co/datasets/openfoodfacts/product-database/resolve/main/food.parquet' LIMIT 10;"
The request can be a bit long (~15 seconds).
From the command line, through the local filesystem
If you want faster results, just download the parquet file from Hugging Face. You'll then be able to query the file with DuckDB, with better request times.
You can easily install DuckDB on your Command Line Interface (CLI) by reading our blog post.
$ duckdb :memory: "SELECT * from 'food.parquet' LIMIT 10;"
How to exploit the Parquet database using DuckDB: Use-Cases
Look at our DuckDB Cheatsheet to see DuckDB queries corresponding to use-cases of the community.
MongoDB dump
The MongoDB dump needs to be reused with MongoDB. It allows building a full replication of the Open Food Facts database and use MongoDB for selecting, filtering and exporting data. Using MongoDB allows faster manipulations compared to the other methods.
First, you need a running MongoDB installation. Open Food Facts is using MongoDB 4.4. It has been reported that prior version should not work for Open Food Facts dump.
You can see here a quick tutorial on how to install MongoDB on Debian 10 or Debian 11.
Import Open Food Facts MongoDB dump into MongoDB
# Download wget https://static.openfoodfacts.org/data/openfoodfacts-mongodbdump.gz # Import (no need to uncompress) # --nsFrom=off.products # source of the database / collection # --nsTo=mydatabase.mycollection # local database / collection # --drop # Before restoring the collections from the dumped backup, drops the collections from the target database. # # --drop does not drop collections that are not in the backup. mongorestore -vvvvv --gzip --archive="./openfoodfacts-mongodbdump.gz" --nsFrom=off.products --nsTo=mydatabase.mycollection --drop
(Deprecated) Import Open Food Facts MongoDB dump into MongoDB
# Download and decompress the dump wget https://static.openfoodfacts.org/data/openfoodfacts-mongodbdump.tar.gz tar -xzf openfoodfacts-mongodbdump.tar.gz # Restore all the database. mongorestore recreates indexes recorded by mongodump. mongorestore --drop ./dump # => 2254885 document(s) restored successfully. 0 document(s) failed to restore. #On newer versions of mongorestore, you can use `--gzip --archive=openfoodfacts-mongodbdump.tar.gz`
Play with the database
# Display 5 first products in JSON format, using pagination # https://www.codementor.io/@arpitbhayani/fast-and-efficient-pagination-in-mongodb-9095flbqr mongo off --eval 'db.products.find().limit(5).pretty().shellPrint()' --quiet # Combined with JQ (JSON tool) to provide colors # JQ has to installed separatly. See https://stedolan.github.io/jq/ mongo off --eval 'db.products.find().limit(5).pretty().shellPrint()' --quiet | jq . # Combined with JQ (JSON tool) to provide colors and compact output (each JSON object on a single line (aka JSONL format)) mongo off --eval 'db.products.find().limit(5).pretty().shellPrint()' --quiet | jq . -c # Get products from Germany; return fields "code" and "counties_tags"; limit to 2 products mongo off --eval 'db.products.find({countries_tags: "en:germany"}, {code: 1, countries_tags: 1}).limit(2).pretty().shellPrint()' --quiet # get the data from one field without _id mongo off --eval 'db.products.find({countries_tags: "en:germany"}, {_id: 0, countries_tags: 1}).limit(2).pretty().shellPrint()' --quiet
Export the database
# Exports # See: https://www.mongodb.com/docs/database-tools/mongoexport/ # 1. The "aggregate" way mongo off --eval 'db.products.aggregate([{$match: {product_name: "Coke"}},{$out: "result"}])' mongoexport --db off --collection result --fields code,product_name --type=csv --out result.csv # 2. the -q,--query option way # Export 5 first german products mongoexport -d off -c products --type=csv --fields code,countries_tags -q '{"countries_tags": "en:germany"}}' --out report.csv --limit 5 # Export to STDIN in CSV format; notice option --quiet mongoexport -d off -c products --type=csv --fields code,countries_tags -q '{"countries_tags": "en:germany"}' --limit 5 --quiet # How long to export all German products? time mongoexport -d off -c products --type=csv --fields code,countries_tags -q '{"countries_tags": "en:germany"}' --out report.csv # real 0m10.135s # Specify the fields in a file containing the line-separated list of fields to export (--fieldFile option) # Official csv export fields are coming from @export_fields variable in /lib/ProductOpener/Config_off.pm mongoexport -d off -c products --type=csv --fieldFile official_csv_export_fields.txt -q '{"countries_tags": "en:germany"}' --limit 5 --quiet
List all fields used in the database
# Open Food Facts database contains hundreds of fields. # An easy way to list them all is to use "variety" Schema Analyzer: # https://github.com/variety/variety # 1. Install "variety" git clone https://github.com/variety/variety.git # 2. Use it cd ./variety # Analyzing can be very long (hours). You can restrict the analysis to a small number time mongo off --eval "var collection = 'products', limit = 1000" variety.js > off_schema_1000.txt # (17 s) time mongo off --eval "var collection = 'products', limit = 10000" variety.js > off_schema_10000.txt # (3 minutes) time mongo off --eval "var collection = 'products', limit = 100000" variety.js > off_schema_100000.txt # (75 minutes) time mongo off --eval "var collection = 'products'" variety.js > off_schema_all.txt # (more than two days)
CSV export via SQL (beta)
We are testing a new kind of tool to provide the data: every day an SQL database is fed by the regular daily CSV export, and published online thanks to Datasette tool.
The tool, called Mirabelle, can be found here: http://mirabelle.openfoodfacts.org/
The whole CSV export can be found here: http://mirabelle.openfoodfacts.org/products/all
- The tool supports simple queries with a form, and also facet navigation.
- For those who know SQL language, it allows rich and complex queries.
What's different with Open Food Facts advanced search?
- It's possible to export selections with more than 10,000 products (eg. big queries by countries).
- It's possible to build queries by date.
- It allows richer queries with OR, AND, NOT, REGEXP, etc.
- It is possible to restrict the number of fields displayed and exported.
- It is possible to order results by any field.
Example
1 -- Build your query (or ask someone to build it for you)
Eg. all German products that have been scanned at least one time.
-- Products from Germany that have been scanned at least one time select code, product_name from [all] where countries_en like "%germany%" and unique_scans_n is not null order by unique_scans_n desc -- the limit here displays 20 results; the link "CSV without limit" below allow you to download all the data without limit limit 20
2 -- Click on the link "CSV without limit"
Maybe you have to wait several seconds. It will download a product.csv file.
Tips
- Several fields -- such as
countries_en
,categories_en
, etc. -- contain multiple values. To query a particular value you have to use the operatorlike
and use percents like this:like %italy%
.