Forecasting model on epochs of numerical events

I’m I’m trying to create a forecasting model on pyTorch, capable of predicting the selection of certain values ​​on future events. The data is collected in Epochs, each epoch is a set of Events, each event is made up of a set of numbers. I managed to find the method to predict the possible values ​​that will occur in the next event, but I can’t understand how to instruct pyTorch.I know I should create a model, and I know I should choose a method for training the model, but I don’t know how to do it. The project concerns a new industrial product, which could save the lives of many people.If anyone wants to help me, I would be grateful.

Hi Klode!

Can you share with us any of the details of what you are trying to do?

It sounds like you have a time series of “Events” and want to predict the
next “Event.” This is certainly the kind of thing you might be able to do
with a neural network.

It’s hard to give useful advice without some understanding of your actual
use case. What, conceptually, is an “Event?” What do the numbers that
make up an Event mean? In what sense are Events (naturally) grouped
together into “Epochs?” And so on …

Give us some context and maybe ask some questions about what you’ve
tried or thought about trying, and someone on the forum could well have
some concrete suggestions.

Good luck!

K. Frank

Thanks K. Frank,

We have a completely closed system. This system periodically generates a series of values, called blocks, which consist of 12 hexadecimal values.

Considering an EPOCH as a group of EVENTS, and considering an EVENT as a Block of Data, the size of an EPOCH is approximately 120 EVENTS.

We cannot predict what the next EVENT, or Data Block, is, because it is almost impossible, but we have discovered that, for each EPOCH, we can limit the number of possible values present in the next Block, i.e. the EVENT following the EPOCH taken into account examination.

An example: suppose we have an EPOCH “E” of 120 EVENTS. In an ideal way, we know the FUTURE EVENT “F”, following this EPOCH. But we know that the values of this FUTURE EVENT are not present in some of the EVENTS already defined in EPOCH “E”. We know that for this given EPOCH “E”, by collecting the values of certain EVENTS taken only once, we are able to reduce the set of possible values in the FUTURE EVENT “F”, thus going from 255 values to just 40-50 values.

The script in Python must use torch, such that:

  • During the TRAINING phase, it analyzes the values of the EVENTS of the known EPOCH “E0”, “E1”, E2", “En”, to which we associate a SELECTION MAP “M”, of the same size as the EPOCH “E0” , “E1”, “E2”, “En”. That is, each EPOCH is made up of 120 EVENTS, therefore each EPOCH will have the SELECTION MAP “M0”, “M1”, “M2”, “Mn”, with 120 values that they can be 0 or 1. Each index of the values of “M0” corresponds to the index of EVENT in “E0”. If the index 0 of M0 is 0, it means that the EVENT of index 0 must be DISCARDED, therefore it contains at least a value of the FUTURE EVENT “F0”. On the contrary, if the index 0 of M0 is 1, it means that the EVENT with index 1 must be SELECTED, therefore it does not contain any element of the FUTURE EVENT “F0”. Therefore, always during the TRAINING phase, the Script must contextually analyze both the EVENT E0 and the MAP M0, in order to understand (and here we don’t yet know how, but I hope in the power of pyTorch!) what connection there is between the EVENTS and the MAP SELECTIONS, analyzing sequentiality of values, or their position, or any other algorithm to me unknown;
  • In the PRODUCTION phase, the Script analyzes ONLY one EPOCH, without associating a SELECTION MAP; based on TRAINING it should predict what will be the correct selection of EVENTS to DISCARD and SELECT, so that we can reduce the set of possible values in the FUTURE EVENT, still unknown. So, in the PRODUCTION phase, it will generate an Array of 120 values.

So pyTorch should find the algorithm that connects the EPOCHES to the SELECTION MAPS, so that it can predict which EVENTS can be discarded.

If you need some other details, don’t hesitate to reply.

Is there anybody who can help me?