Need guidance. Do factors weights matter?
It seems to me each weight of all the factors in a ranking system should be set to anything non-zero?
Ridge regression will give weights or coefficients that do matter, indeed.. And the coefficients are never zero.
P123 will be making those coefficients available, or knowable to the user, at some point soon. So, you will be able to know what those coefficient or weights are. They never go to zero with ridge regression. The more noisy features are "shrunk," however.
If you don't like shrinkage there is linear regression. And of course, you have complete control of the weights with P123 classic.
If you are willing to explore the term "backpropigation" neural-nets have weights that are more difficult to understand and harder to quantitate. Specifically, in neural networks, 'weights' refer to the strength of connections between neurons. While more complex, these weights are crucial to the model's predictions.
BTW, a neural-net with a single layer (and a linear activation function) duplicates a linear regression where the 'weights' in a neural-net have a clear meaning. In this case, the weights in the neural network directly correspond to the coefficients in linear regression. If you want to look for more complex interactions with more layers, it does get harder to interpret. There is no free lunch.
But the trade-off for that complexity is that neural networks with more layers can capture non-linear relationships and interactions that linear regression can't,
While 'weights' might not be the right term for random forests, they do provide 'feature importance' measures, indicating how much each factor contributes to the model's predictive power."
P123 offers a wide range of choices to find the optimal "weights" for a model and with P123 classic you have complete control over the weights you use.
I meant to ask if AI models in p123 treat two input sets of the same factors differently depending on the weights or signs (lower / higher). My guess is not, but it's not clear. Thnx.
When you input features from a ranking system into an AI Factor, the node-specific weights and signs are discarded. The AI model processes the features without considering their original assigned weights or directions.
Thank you! That answers my question.