nn-mapping

nn-mapping maps the given JSON objects to JSON structure which are more suitable for neural network processing. Neural networks need float values as input and output. nn-mapping maps deep structures and non numeric values n dimensional float arrays and key value pairs. Key value pairs are used as abstraction layer for the model. The mapping is done based on a JSON mapping definition.

Usage

Command line

The module installs the nn-mapping command line tool in the npm path. The following arguments are required to run the tool:

nn-mapping --mapping <mappingDefinition> <inputFile> > <outputFile>

• mappingDefinition: Path to the file that contains the JSON mapping definition.
• inputFile: The JSON input. stdin is used if this arguments is not given.
• outputFile: The mapped JSON is written to stdout.

API

The module exports the following functions:

• mapItem(mapping, object): Maps the given object and returns the mapped object.
• mapArray(mapping, objects): Maps an array of objects and returns an array of the mapped objects.
• mapStringStream(mapping, stream): Parses the given JSON string stream and maps each object of the array.

Mapping definition

The mapping definition is JSON object. It starts with an object mapping.

Object mapping

• map: An array of property mappings.
• loop An array of array mappings.

Property mapping

• group: The output group the value will be assigned to.
• property: The property that will be used as source for the mapping.
• mapping: The value mapping.

Value mapping

• neuron: The name of the neuron the value will be assigned to.
• equals: The mapping will be only processed if value is equals to the value of equals.
• value: A hard coded value. Useful conditional mappings.

Array mapping

• group: The output group for the array
• property: The property that contains the array.
• map: An array of property mappings.
• loop An array of array mappings.

Example: Calculator

The examples folder contains a simple calcualtor example, which generates datasets sequences of numbers and +-* operations. The calculator.sh script prepares two training datasets with different sequence lengths and a test dataset. It also does the mapping. To run the examples switch to the example folder and run:

./calculator.sh

Keras Gaia is able to process the dataset. It also contains prepared example projects for the nn-mapping calculator example.