How to make the flow of character data in a neural network?

Good day!
The input neurons of the Ann(GRNN) must submit names of cities.
How to do it, because you can't just tell us the id of the city or its name.
Thanks in advance.
September 26th 19 at 11:52
1 answer
September 26th 19 at 11:54
And why can't I send ID? If a lot of cities, a little hard to believe that the neural will work well.
ask weight for id=1 and then the same weight to the id=101 ??) - Leone_Bednar commented on September 26th 19 at 11:57
Small discrete values, of the order of 10 can be represented in the form of code 001000, where the unity is in place with the corresponding number. If you consider the hundreds of cities, then maybe it's worth to try geographic coordinates? - Alvah.Kuhic22 commented on September 26th 19 at 12:00
if you could talk a little bit about the task, any advice would be easier. - Alvah.Kuhic22 commented on September 26th 19 at 12:03
filing date, city, temperature, humidity, etc. on the forecast weather conditions after a specified period of time. - Leone_Bednar commented on September 26th 19 at 12:06
Again perceptron are considered to be panaceas. It is better to study meteorology and use the right techniques, rather than obscure the neural network.

In short: we need a trivial function of temperature from the date, humidity and pressure. However, meteorologists use data from hundreds of drones and Autonomous sensors, satellites and thousands of man-hours. But the result still leaves much to be desired. Yes, it uses neural networks, but it is rather a toy. The fact that neural networks are well characterized and are able to look for connections. Meteorology is a thick stack of formulas and strong datamining, there is no need to remember or look for the hidden connections - they are not hidden.

Example: if in Arkhangelsk damp and very cold, and the pressure is very high, whereas in St. Petersburg heat, 70% humidity and very low pressure... Then soon in St. Petersburg there will be rain and sharply colder. Why? Air under pressure will rush from Arkhangelsk to St. Petersburg, and a cold moisture will rush to Leningrad cooling air. Damp Petersburg air skondensirovat and fall in precipitation.

You can try brute force to teach the neural network to give long-term forecasts. Susesi for a person is quite easy and you can prepare a pack of formula to arrive at the simpler neural perceptron post ant need a lot, a lot of data and huge capacity. - Clare63 commented on September 26th 19 at 12:09
also I want a toy!!!1!! What to do with the letter names? - Leone_Bednar commented on September 26th 19 at 12:12
In this case, the names of cities 100% not needed. Moreover, the IDs is also not needed. Still you have to transfer data in all cities for 1 time. Data will only correspond to a subset of the inputs. - Alvah.Kuhic22 commented on September 26th 19 at 12:15
well, I thought to reserve for the neuron in the city, but then you have 720+ inputs - Leone_Bednar commented on September 26th 19 at 12:18
okay, then name a better taminent numbers: their neural network is better chews. And even better - coordinates: in any case You will need to enter a distance, coordinates all do and the network may become a little smarter. So something right.

And Yes, I was warned that it will be long. 720+ inputs? There will be thousands and thousands of neurons and millions of connections. Training will take months, years even on a Core i7 @ 5 GHz 8 threads 128 GB RAM (at the other hardware simply will not work), forecasting for a couple of days will require weeks. Optimization is so important pull off the accuracy and performance.

I speak from experience, drop the case. In a pinch, you can look for other tasks neural networks - restoration of damaged/blind video, photo, solving optimization problems, narrow datamining. In General, anything where a small perceptron able somehow to show themselves. - Clare63 commented on September 26th 19 at 12:21
you can try to simplify the task. To start, build a network predskazuemo weather the next day in the same city for 5-7 points on the map. Then, if the weather in the city will identify, say, 10 inputs, the network obtained with 7*10=70 inputs and 10 outputs. - Alvah.Kuhic22 commented on September 26th 19 at 12:24

Find more questions by tags Artificial intelligenceNeural networks