When it comes to LSTM models or any NN, there are no rules as to how many neurons/layers a model should have and usually you have to work your way up top perfect these parameters. But I was wondering if there is a way to let something like NEAT (Neural Evolution of Augmenting Topologies) make the evolution of the model happen, and decide for better architecture for the model with each generation.
The NEAT-Python package is: https://pypi.org/project/neat-python/
So I have two specific questions:
- Is it possible to optimize the architecture of a LSTM model using NEAT? (based on prediction accuracy that will be manually provided to NEAT through a prediction function)
- Is it at all possible to just use NEAT for a time-series problem like stock prediction? If so, how efficient would it be in comparison to LSTM?