TensorFlow How to predict corresponding numbers?

Question

Python

What should be the code for predicting an approximate corresponding value for a given number?
For example: 1 corresponds to 1, 2 -> 4, 3 -> 9, 4 -> ?
Or 1->1, 1->2, 2->99, 2->101, 3->1000, 4->?

Answer 1

Well, obviously - something like this:
- collect a training sample that, in your opinion, will well describe the function that needs to be predicted. And at the same time - quite large (well, for example, 10 times more than it is planned to have weights for the trained network).
- divide the sample into training and test parts (randomly). For example, in the ratio 0.7-0.3 (70% for the training part, 30% for the test part)
- choose the error function that suits your case. MSE, for example (mean squared error).
- repeat until an acceptable error level is reached on the test sample:
- select the ANN configuration (number/types of layers, their activation functions)
-- for a pre-selected number of epochs - train the ANN on the training sample (read - minimize the value of the error function. By the way, the question is which optimizer to choose).
-- evaluate the error on the test set.
And here is the code - with this already on fl/upwork/etc :-)
ps haven't fiddled with tensorflow specifically, but it seems it shouldn't be a problem to figure out any of the steps above once you have the algorithm, isn't it :-) ?

Answer 2

with numbers instead of questions

Answer 3

How to predict the corresponding numbers?
to collect a training sample :)
there is a thin line between overtraining and misunderstanding of the neuro-shell :) the
neuro-mesh cannot reliably predict the future when it comes to lotteries :)
because it doesn’t have the understanding of the whole brain,
but there are certainly methods of forecasting,
but for not understanding them for a long 9 years