Tree Neural Network #10
Description
New kind of algorithm with the idea to combine Forest algorithms with Neuralnetworks. It will have the following characteristics:
- Each layer will be a bootstraop forest without essembling
- The number of outputs of each forest will be the input of the next forest
- The final forest should predict the tarjet with bagging
The main concern about it is how to implement the backpropagation without the concept of activation function derivative. This should be the first issue to think about