alison-carrera / onn Goto Github PK

I wanted to use the algorithm. However, my feature size is 51 and your algorithm can't validate this. it throws an error if features shape is not 2.

if len(data.shape) != 2:
    raise Exception("Wrong dimension for this X data. It should have only two dimensions.")

Are you planning to modify it to work even with multiple outputs ?

Fixing it, maybe next week will be OK.

This setup seems freaking amazing. Why aren't you marketing it more? It would be very good as an online learning system.

@alison-carrera
Would it be possible to extend the hedge backpropagation to stacked LSTM structure? I am interested in doing online time series prediction, so for example at time t, the model gets length 10 sequence at times [t-9,t-8,...,t-1,t] and predicts the value at time t+1. Would you say such task is possible using ONN?

Hi, the library looks amazing. I wonder if it is possible to save the model after training and to resume training from a saved model. Thanks.

I'm trying the simple Online Neural Network (ONN). I'm try to do multiple partial fit with your test data in your README, but it's not converging.
For example when I run 100 times the partial_fit function with [0.1,0.2] -> 0 and [0.8,0.5] -> 1 I always have zero (0) as the predictions, so the training accuracy is 50% in this toy example for some reason.

Am I doing something wrong here?
Thanks!