Full batch neural network

mdabros / SharpLearning

Machine learning for C# .Net

MIT License

384 stars 84 forks source link

Full batch neural network #157

Open SULVAL opened 2 months ago

SULVAL commented 2 months ago

Hi! I don't know if i got it right reading documentation and examples. However my question is: in order to train a neural network in full bartch mode (that is, using all the available instancdes), is it correct to impose "batchsize: number_of_all_available_instances" in the RegressionNeuralNetLearner constructor ?

Tnx a lot in advance

mdabros commented 2 months ago

Hi @SULVAL

The implementation in SharpLearning is not intended for running in "full batch mode" as such. The all the different optimizers er based on stochastic gradient descent, so is designed for mini-batch learning. However, if you want to try and use it for "full batch" learning, you are correct. The batch size should be set to the number of examples or instances in the training set. Be aware that it might require a lot of memory to run in this mode.

Best regards Mads