The goal of this project is to uncover the latent meaning of location, by creating Geospatial vectors that take into consideration a wide array of different data points(demographic, geographic, psychographic, environmental, temporal), so the resulting embeddings can then be used in a wide array of application.

It will also be possible for you to access the feature(s) that matter(s) to your specific use case(income, population density, etc)

In industry after industry people have come to understand the importance of proximity

• Geography:

Tobler's first law of geography. "everything is related to everything else, but near things are more related than distant things."

• NLP:

You shall know a word by the company it keeps (John Rupert Firth), which is the underline principle behind word embedding (word2vec, GloVe, etc.)

• Social:

You Shall Know People by the Company They Keep: Person Name Disambiguation for Social Network Construction

• Time series:

The exponential moving average are weighted moving averages that give more weighting, or importance, to recent price data than a random sample. In this case, the distance is time.

• Computer vision:

One of the best ways to find image similarity is by using a siamese network with a triplet loss function (facenet, loc2vec, object tracking, etc), this is done by training the model on an image(anchor) and a similar/augmented image(positive) and dissimilar image (negative) and the loss function attempts to maximize the distance between the anchor and negative image.

• Structured Data:

A wide range of companies create and use embedding to understand the latent features of their subject of interest (pins, House, stores, products, industry2vec, etc) by using similarity/proximity measurement

• Addresses: OpenAddresses
• Maps: OpenStreetMap(OSM)
• Places: OpenStreetMap(OSM)
• Satelite data:
• Demographic: Census Bureau(decennial census, American Community Survey (ACS),)
• Enviromental:
• Other: osmn, liquor licenses(state agencies)