README.md

GoSineSim

Cosine Similarity of two or more shallow JSON objects in Go.

Usage

gosignsim --source=$JSON_OBJ_LITERAL --pool=[$JSON_OBJ_LITERAL,...] [--threshold=float] [--output_file=/path/to/save/results] [--verbose=boolean]

Options

• source -- This is the main object that all other objects will be compared against.
• pool -- This is a list of objects that will be compared against the source object.
• threshold (optional) -- This is the minimal similarity value that an object must meet in order to make the cut.
• output_file (optional) -- If this option is selected the resulting JSON will be saved to this file.
• verbose (optional) -- If this is set to true(thy) various data will be sent to stdout. ie, if output_file was set, you would still see the resulting JSON.

GoSineSim works by comparing one data structure against one or more. Each item should be made up like this:

{
    "id": "15",
    "data": {
        "cars": 30,
        "money": 99
    }
}

[{
    "id": "44",
    "data": {
        "cars": 87,
        "money": 40
    }
}]

Data Rules

1. Each item set must have a string id field and an object literal data field. The data is a simple key: value pair where the keys are stings and the values are floats

   type Item struct {
   	Id   string
   	Data map[string]float64
   }

2. The source argument is a single item

3. The pool argument is a list of items

Example

./gosinesim -source='{"id": "15", "data": {"cars": 30, "money": 99}}' --pool='[{"id": "44", "data": {"cars": 87, "money": 40}}]'

Which would produce the result

[{"Similarity":66.32728204403626,"Id":"44","Data":{"cars":87,"money":40}}]

Benchmarks

I wrote a very simple benchmarking suite in python bench.py that simulates what it would be like to compare a random sampling of items from python's system call method.

The file creates 100,000 items which can contain between 5 and 15 entries. It then chunks the items into groups of 100, 1000, 5000... up to 100k. As you can see from the results below, a comparison of less than ~5k items can be done in realtime.

My specs are: macOS, 2.5 Core i5, 16GB 1600 DDR3, and an SSD

************************************************************

Building the pool of 100000 items

done building the pool: 3.93292379379


****************************************

Comparing 100 items

	run 1 took 0.0157909393311
	run 2 took 0.0160090923309
	run 3 took 0.0309271812439


****************************************

Comparing 1000 items

	run 1 took 0.0655910968781
	run 2 took 0.0748629570007
	run 3 took 0.0511500835419


****************************************

Comparing 5000 items

	run 1 took 0.22353386879
	run 2 took 0.204764127731
	run 3 took 0.197242021561


****************************************

Comparing 10000 items

	run 1 took 0.291663885117
	run 2 took 0.608170032501
	run 3 took 0.627039909363


****************************************

Comparing 25000 items

	run 1 took 0.51177406311
	run 2 took 0.534518003464
	run 3 took 0.545824050903


****************************************

Comparing 50000 items

	run 1 took 0.905314922333
	run 2 took 0.918902873993
	run 3 took 0.902179002762


****************************************

Comparing 75000 items

	run 1 took 1.37646508217
	run 2 took 1.3969039917
	run 3 took 1.66373586655


****************************************

Comparing 100000 items

	run 1 took 1.83165812492
	run 2 took 1.93080997467
	run 3 took 5.73800802231


============================================================

finished running: 26.3283598423