Feel free to contact me on my email if you encounter any problem for setup or have any question related to this application.
It runs currently in a console mode, but displays its results as a graph on localhost (web display).
After having entered a product reference (in the same way as in www.openfoodfacts.org), it will seek for similar products in the OFF database and display them on a graph based on their nutrition-score (grades A to E) and their similarity with the product reference.
It makes it then possible to the consumer to look for more or less similar products with better nutrition properties if wanted. Proceed first with the setup described here: http://openfoodfacts.github.io/OpenFoodFacts-APIRestPython/ Once you have entered a product reference in the command line, the application fetches a subset of information for this product, and especially the nutriments (for computing the nutrition score) and the categories_tags (for fetching more or less similar products).
The application then fetches in the database all products for each entry of the categories_tags which the reference product belongs to (this is surely not optimal!). The intersection of the categories_tags for fetched products and the reference product is computed. The higher the intersection, the closer the similarity with the reference product; so for now, the "score proximity" of a product with the reference product is computed as follows:
<nb of intersected categories_tags of product X > / <nb of categories_tags of reference product>
Note: if all categories_tags of the reference product are listed in another product, then the match is 1, i.e. 100% Once the OFF database is decompacted (see "Requirements" above), perform the following actions:
- start the mongo server locally in a terminal with appropriate options (ex.: mongod --httpinterface --rest --dbpath ./db_produits/dump/off-fr/ ). Wenn the server tells the following, then the connection is ok: [initandlisten] waiting for connections on port 27017 [websvr] admin web console waiting for connections on port 28017 - before running the application, you may need to update the hard-coded localhost call: locate the Class "Querier"/method "connect" the lines: # todo: review since hard-coded! if self.verbose: print '.. connecting to server MongoClient ("127.0.0.1", 27017)' self.pongo = MongoClient("127.0.0.1", 27017) and update them accordingly with the port the mongo server diaplyed in the preceding step (in our case "27017") - by default the code displays the result on a web-page (localhost). But you may also wish to display as MatplotLib Client instead. This is actually customizable in the Graph->prepare_graph(self) with the activation of 1 of these options: - mpld3.show() - plt.show() - now run the Python file openfoodfacts.py. If the number of products is not 0, it then means that the connection to the server was successful, and you may go further - When asked to enter a product code, you may either press RETURN (in which case a default product is chosen), or enter a product code which is available in the OFF-database locally:
verbose = True
.. connecting to server MongoClient ("127.0.0.1", 27017)
.. connecting to OPENFOODFACTS database
.. getting PRODUCTS collection
68483 products are referenced
please enter a product code ['q'uit] >
- similar products are then fetched and scores are computed, which may take some time. - When you finally see the following messages (this may take some minutes!), then your graph is ready and displaayed hopefully: - 127.0.0.1 - - [15/Apr/2016 19:38:33] "GET / HTTP/1.1" 200 - - 127.0.0.1 - - [15/Apr/2016 19:38:33] "GET /d3.js HTTP/1.1" 200 - - 127.0.0.1 - - [15/Apr/2016 19:38:33] "GET /mpld3.js HTTP/1.1" 200 - - 127.0.0.1 - - [15/Apr/2016 19:38:34] code 404, message Not Found - 127.0.0.1 - - [15/Apr/2016 19:38:34] "GET /favicon.ico HTTP/1.1" 404 - - the black point along with the white square refers to the reference product (similarity of 100% and its nutrition score). All other points are the fetched more-or-less similar products. Within a square, these products are displayed according to a specific algorithm for the repartition of points in a disk-surface (this has been my own decision and is arbitrary) [file point_repartition.py] - Plot your mouse on any point to see some details of the matching product - Matplotlib Client graph displays the black point for the reference product correctly, but the corresponding white square is mirrored (probably a discrepancy between plot in mpld3 and matplotlib) - whereas plot shows correct axis data, mpld3 doesn't (bug referenced in http://stackoverflow.com/questions/35446525/setting-tick-labels-in-mpld3) - performance has to be improved in many ways: - the lower the similarity, the less points we need to retrieve (boundary instead of all products) - adding filtering criteria (country, ...) - creation of index for categories_tags (?) - use categories_hierarchy instead for optimization (?) + index - display points on a web page with labels, urls, and images in ajax-mode (see http://world.openfoodfacts.org/cgi/search.pl?action=process&tagtype_0=categories&tag_contains_0=contains&tag_0=Breakfast%20cereals&sort_by=unique_scans_n&page_size=20&axis_x=sugars&axis_y=fat&series_nutrition_grades=on&graph=1). We may take the same libraries instead of mpld3 - nutrition scores are France-based and may be different in other countries - nutrition scores should differentiate between waters and beverages (see http://fr.openfoodfacts.org/score-nutritionnel-france) - in the web-page, possible addition of fields (country, ...) to display only products of customer's country and also to reduce the number of matching products (performance) - enhance the layout/display, interface