Hands-On Deep Learning with Apache Spark

8.1.3 Using visualization to tune a network

Now, let's look at how we can interpret the visual results presented in the DL4J UI and use them to tune a neural network. Let's start from the Overview page. The Model Score vs. Iteration chart, which presents the loss function for the current minibatch, should go down over time (as shown in the example in Figure 8.2). Regardless of whether the observed score should increase consistently, the learning rate is likely set too high. In this case, it should be reduced until the scores become more stable. Observing increasing scores could also be indicative of other issues, such as incorrect data normalization. On the other hand, if the score is flat or decreases very slowly, this means that the learning ...