As you dig deeper into machine learning, you’ll find yourself tuning how the model is trained and retrained to improve its performance. You do this by altering things such as the number of layers in the network, and the number of training steps. These are called hyperparameters.
In machine learning, there are parameters and hyperparameters. Parameters are the data that the model turns into a prediction, such as pixels in the images, or words if we’re doing text classification. Hyperparameters affect how well the model performs once trained.
For example, we’re setting “how_many_training_steps” to 500, but it could be that 1,000 training steps would create a more accurate model, or maybe 300 is enough. Play about with the number. Retrain the network using fewer training steps, then see whether the confidence in identifying the image performs as well. Then retrain with more steps, and compare your findings.
If you get really into this, you can plot the performance of your model against the number of trained steps to find the optimum number. Doing this is known as “hyperparameter optimization.”