See the demo at https://facebookincubator.github.io/overture2osmjs
The source of the demo is in demo/index.js
You can find library distributables in dist/. It should be on npm soon!
The main function is overtureToOSMData(overtureFeature, processAddress).
overtureFeature is a Overture Place-type feature, which will be converted to OSM tags. processAddress is a boolean, if set to True it will also do a best-effort conversion of the address using the Nominatim API. Converting an address will need network access and takes about half a second, while without it the processing is almost instant.
The function is asynchronous, so the output is wrapped in a promise. It returns an object where the properties are OpenStreetMap tags. See https://wiki.openstreetmap.org/wiki/Tags for more info.
You can test the library using the CLI utility provided. It needs a recent version of node installed.
You also need to run npm install to install its dependencies.
node o2o-cli.js <filename> [--output <console|file|both|xml>] [-a/--address]
Process an Overture .geojson file. By default it will convert it to all possible formats, without address processing. The -a switch will turn on address processing, which is set to only process one feature per second because of Nominatim API limits.
This section outlines how to use the machine learning model to validate and predict address data based on a neural network.
This system leverages a neural network to predict whether an address is valid based on various features like freeform, locality, region, and country. The model uses a trained neural network to classify the input and provide a confidence score for its prediction.
- validatedAddress.json: This file contains the pre-trained neural network model. It is loaded into the script and used to process address data.
- vocab.json: This file includes the vocabulary of possible values for the address components (
freeform,locality,region,country). It is used to encode the input data into a format that the model can understand.
The model is implemented using brain.js, a simple neural network library for JavaScript. The neural network is trained to classify address data based on a variety of features (e.g., locality, region, country) and predicts the likelihood of an address being valid or not.
The input address is encoded into an array of binary values. Each feature of the address (e.g., freeform, locality, region, country) is represented as a vector where each element corresponds to a unique feature in the vocabulary file. If a feature is present in the input, it is marked as 1; otherwise, it is marked as 0.
The system checks if all required address fields (freeform, locality, region, country) are present. If any fields are missing, it will prompt the user to provide them.
Based on the user's input, the system can suggest possible values for missing fields. For example:
- If the user provides a
locality(e.g., "Springfield"), the system can suggest potential regions and countries associated with that locality.
Once all required fields are provided, the input is encoded and passed to the neural network. The model then outputs a prediction with a confidence score indicating the likelihood of the address being valid.
- Run the Script:
To run the model and prompt the user for address input, execute the script:
node testAddressModel.js- Enter Address:
The script will ask you to input an address in JSON format, such as:
{
"freeform": "1234 Elm St",
"locality": "Springfield",
"region": "",
"country": ""
}- Missing Fields and Suggestions:
If any fields are missing, the script will prompt you to provide them. It will also provide dynamic suggestions for fields like region and country based on the locality entered.
- Model Prediction Result:
Once all fields are provided, the model will process the input and output a prediction:
Model prediction result: { "validity": "true", "confidence": 0.98 }validity: Indicates whether the address is valid (true/false).confidence: A confidence score (between 0 and 1) indicating how likely the address is valid based on the model’s training.
{
"freeform": "1234 Elm St",
"locality": "Springfield",
"region": "",
"country": ""
}Please enter a value for region: Illinois
Suggested values for region: Illinois, Missouri, Ohio
Please enter a value for country: USA
Suggested values for country: USA, Canada, UK
Model prediction result: { "validity": "true", "confidence": 0.98 }The validateAddress function checks if the required fields are provided. It returns a list of missing fields that need to be filled:
function validateAddress(address) {
const requiredFields = ['freeform', 'locality', 'region', 'country'];
const missingFields = requiredFields.filter(field => !address[field]);
return missingFields;
}The suggestField function provides dynamic suggestions for certain fields based on the provided input. For example, if the locality is "Springfield", it suggests possible regions and countries:
async function suggestField(field, input) {
if (field === "region" && input.locality) {
if (input.locality.toLowerCase().includes("springfield")) {
return ["Illinois", "Missouri", "Ohio"];
}
return ["California", "Texas", "New York"]; // Default suggestions
}
if (field === "country" && input.locality) {
return ["USA", "Canada", "UK"];
}
return []; // Return empty array for other fields
}- Model Updates: The model can be retrained with more diverse address data to improve predictions.
- API Integration: The system could be extended to fetch real-time suggestions from external APIs for regions and countries.
- Extended Validation: Additional fields and more complex address structures can be incorporated for advanced use cases.
brain.js: For implementing and running the neural network.readline: For handling user input through the command line.
To install the necessary dependencies, run:
npm install brain.js readline