Plot loss and accuracy with Module API



Sorry for asking this basic question. When training, using the Module API, I’d like to store properties like loss, accuracy, etc during training. For instance, the Keras function fit returns a History dictionary containing these values. But I can’t find it how it is done with MXNet. I don’t assume that one needs to do complicated text parsing from the std.out (DEBUG) log.

Many thanks in advance,


You can use the Metric API to store and update your metrics:
Since you want to store multiple metrics you may want to use mxnet.metric. CompositeEvalMetric