Message Boards Message Boards


Training a model to detect a very specific pattern on data?

Posted 5 months ago
0 Replies
0 Total Likes

So I am new to machine learning and I want to take on a small project to get my hands dirty.

I want to train a model to detect a very specific pattern on a given data. Given a set of numbers, I want it to be able to tell if the numbers follow a very specific pattern (so kind of classification model - 1 if the data follows the pattern and 0 if not).

Let's say its a really arbitrary set of rules, a series of 10 numbers follow the pattern if all the following conditions are true:

  • There is at least one possible way to divide them into 2 groups of 5 numbers where the sum of one group is equal to the sum of the other group.

  • The first 3 numbers are arranged in ascending order

  • The last 3 numbers are arranged in descending order

I can obviously generate a lot of samples which follow these rules, and tell the model that they are classified as 'True'. But what do I do to teach it the 'False' classification? There isn't really a way to classify a 'False', there is only a way to classify a 'True'. The only way to classify a 'False' is if it's not classified as 'True'.

I can generate samples which doesn't follow these rules, but won't it just confuse the training algorithm? Won't it try to "figure out" a pattern for those samples as well? While there isn't such.

I want a model which "understands" that it only has to check if a sample belongs to one group and if it's not, it belongs in the other group, which might not have a common pattern.

What is the best way to train such model? Is there a different kind of model which is more suitable for this situation?

Reply to this discussion
Community posts can be styled and formatted using the Markdown syntax.
Reply Preview
or Discard

Group Abstract Group Abstract