Specify when the model needs to be trained

IRC-SPHERE / HyperStreamOnlineLearning

Plugins for Online Learning (supervised or unsupervised) in HyperStream

MIT License

3 stars 1 forks source link

Specify when the model needs to be trained #5

Open perellonieto opened 6 years ago

perellonieto commented 6 years ago

There must be some method in the tools to specify when the model needs to be trained and when to stop the training. This could be integrated into the same model, but it could be as well some other stream of options that could be given alongside the data. E.g. <stream , metadata = (train: true, learning_rate: 0.1)>.

tdiethe commented 6 years ago

I'm assuming that the learning rate should be changing over time, otherwise it should just be a parameter.

A few options here:

have a separate tool for training and prediction (maybe not ideal)
augment the data stream so that it contains this information and gets unpacked by the tool
provide a second stream that is aligned to the data stream, and contains packets with the metadata you suggest (e.g. as a dict)