Location had been a sacred topic in sales for a long time. Nowadays one and all know that if salesmen do not have the right location, they cannot sell in good health. With companies moving into the global market and being online 100% of their time, location had become even more important than before. Closing a sale, working on a sales prospect, comparing competitors – these are only a fraction of tasks companies are interested in doing. Interactive data visualization had become bread and butter for CEO and board members. To obtain all of the above, a cluster of a pair of coordinates needs to be placed on a map as map pins.
The current standard for visualization on a map that humans employ is WGS84 which defines how latitude and longitude will be represented on a map. Machines can handle that standard well; humans in the majority of cases, not at all.
The postal Addresses represent how humans observe the world as a location: tokenization of words, with each word representing part of the decision for a location. The problems start when the two worlds meet. And since humans had the advantage of being around longer than machines, all address standards around the world are done with the understanding that humans will use it.
In this session, I will demonstrate a real case study of how frustrating it can be to work with machines when an incomplete or mixed, immense amount of data is given by humans and how a geocoding map API similar to Google or Here can produce garbage. The API’s perceive it as Garbage In, Garbage Out – GIGO.
Then we will demonstrate how to glean Deep Learning with a Recurrent Neutral Network that solves the problem much better. I will bring into play Python calls with TensorFlow for the demonstration.