Page History

...

The new API described below reduces the number of lines of code to be written by the user. In cases where the user implements logging and checkpointing, the number of lines is reduced from ~40 to ~6

Below is an example for the Fit API implementing similar functionality to the one using the existing training loop in Appendix A.

import mx.gluon.estimator as est
net = get_model() ## get the network
loss = gluon.loss.CrossEntropy()
e = est(net, lossfn = loss)
## training 
trainers = [gluon.Trainer('sgd',{'learning_rate':0.001})]
e.fit(train_data, val_data, epochs, trainers, context)

...