What are the requirements for machine learning on the prediction of the RNG?

Recently there was a following problem. There is a random number generator (pseudo-random of course). The number distributed in the range of 1 -1000, inclusive. Next was assembled dataset in which the order of 45000+-2000, the data of numbers generated randomly. Is it possible to train the network so that it predicted the next number? Or need more data, not quantity of numbers and the input variables depends on the number of the output?
April 19th 20 at 12:08
2 answers
April 19th 20 at 12:10
The neural network - hardly. Its basic profile - work with a smooth continuous values. And if we are talking about prediction here is rather close to the term - approximation. Or extrapolation.

And if your RNG - strong - it is considered the case hopeless. He spices was created to nobody never guessed of projections.

RNG is something discrete. It is more suitable combinatorics and HA. For example just to verify the ownership of your dataset some class variables. On replays. The shape of the distribution. If it is not linear.
April 19th 20 at 12:12
In your case the problem is reduced to the prediction quantity. First try to resolve it with the help of "classical" machine learning methods (e.g. linear regression). The link to the blitz test. And there neobhodimot in the ins will disappear.)

Find more questions by tags Random numbersNeural networks