Monitoring models
To create an ML system, it is not enough to find an appropriate ML method and run the model training code. Since the training depends on the data and model hyperparameters, it is important to monitor the model while training to ensure proper convergence to the optimal parameters. Furthermore, after the training, we need to monitor the model to make sure that the performance of the model has not dropped below the acceptable threshold.
In the next section, we’ll give an introduction to the model monitoring during the training process.
Monitoring during training
During the training process, it’s common to monitor the changes in the loss function to ensure that the model parameters are converging to optimal values. Furthermore, we can monitor performance on the training and test set on each training iteration, or on every N iteration. This way, we will know whether the model is still improving and we should keep the training running, or whether...
