I understand Recurrence Neural Networks are used frequently in natural language processing or similar processes. For example, predicting the next word in the sentence: “I took my dog for a ___” (walk). As an analogy, I’m trying to instead use RNNs for a version of generating what can be seen as a “coherent” sentence, given multiple other sentences.
What I’m really trying to do is, given multiple sets of points, and how each set of points is partitioned, I’m trying to find how I would partition another set of points in order to match the pattern of the previous sets.
RNNs seemed like a good method to do this, purely because of their high data interdependence (nearby data decisions affecting other data), which I want so that the decision of what cluster a specific point belongs to affects future decisions within the same set. Does an RNN seem like the correct method for this, or does something else come to mind?
One difficulty I have already come across is that the “sequence size” parameter kind of changes with what point you are currently deciding. So for example, if you are making a decision on the third point in the set, your “sequence size” should be 2, since you are deciding based on current weights & what you classified the other two points as. On the other hand, if you are deciding the 10th point, your sequence size should be 9. This is different from RNNs ive seen, where the sequence size is constant.
Thank you so much for reading or if you decide to respond!