Json prediction and recommendation engine
JSON prediction and recommendation engine
Fussy is an inference engine for Node. You can use it to guess missing values in a JSON object.
It tried to find the most probable values for missing attributes by scanning a database of previously seen objects, computing a weighted average based on similarity.
Demo examples (included in the project):
BSD (see LICENCE.txt file).
Fussy is an experimental project, and has a number of pitfalls:
The algorithm works as the following:
For a given JSON object with some missing fields, it tries to determine the most probable value of these fields, by looking at all the past values, and computing the average.
However, it does a weighted average: a different "trust" is given to each past object, depending on how close and relevant they are from the object to repair.
For this, what Fussy do is iteraring over all the stored objects (in a map-reduce fashion), and computing a distance score, based on the similarity between values: strings, numbers..
This distance will be used to weight the value of the missing field when computing the average.
Go to your Node (and NPM-managed) project, and run:
$ npm install fussy --save
To download the sources, build the coffee-script and link into your system:
$ git clone firstname.lastname@example.org:jbilcke/node-fussy.git $ cd node-fussy $ npm run build $ npm link
First you have to get an instance of the class Fussy
var Fussy = require'fussy';var fussy = ;
Then you can insert documents. You have a few ways of doing this.
You can use use the
insert method, which takes a JSON object, or an array of
fussyinsert 'food': 'rice' 'taste': 'good';insert'food': 'salad' 'taste': 'good''food': 'grass' 'taste': 'bad';
After spending some time using Fussy on various datasets, I found it handy to write a small importer for CSV files. So here we go!
import function takes an input CSV file, and a list of columns as parameter:
var data = fussyimport'thermal.csv' 'day' 'temperature' ;
This second parameter can be used to define types:
You can also define a dictionary of values, when using Strings:
fussyimport'thermal.csv''day''mon': 'Monday''tue': 'Tuesday''wed': 'Wednesday''thu': 'Thursday''fri': 'Friday''sat': 'Saturday''sun': 'Sunday''temperature''Number';
Sometimes you need to do some operations on a dataset before using it.
For instance, maybe you only want to keep a subset of the dataset, or do random sampling, so you need access to the array before importing it.
Fussy provides a function to create a dataset (array of JSONs), available
fussy.toolbox.dataset takes an input CSV file and a list of columns as
It works like the import function:
var data = fussytoolboxdataset'thermal.csv' 'day' 'temperature' ;
You can then manipulate this array, before importing it. For instance:
var data = fussytoolboxshufflefussytoolboxdataset'data.csv' 'schema.json';
Will load the dataset and shuffle it.
var query = fussyqueryselect: 'column'where:foo: ''bar: '';
When you call the query object, what you get is a result set, or "view" on the data. This view has the following methods:
best function returns the best value for a given field. It actually
just takes the first element of the
Depending on the distribution and the category of problem you are trying to solve, this might not be the best choice for you.
mix compute the weighted average value for a numeric field.
For instance, say there are 3 possible values for a "temperature" field: 10, 20, 40..
all function will returns an array of value->weight, the
function will directly returns you the weighted average, eg. 23.33.
all function returns the distribution of values: an array sorted by weight,
of all possible choices for the requested fields.
This is actually an array of
(value, weight) tuples.
Use this function if you want access to raw data, and need to make multiple, weighted decisions. (eg. for investment, risk management use cases).