Model Compilation

The compile method

After the model architecture is set, the model can be compiled using the compile method. This sets up the network for optimization. It creates an internal function to perform backpropagation efficiently.

Arguments for the compile method

The compile method has two arguments:

1. Optimizer

The optimizer helps specify the learning rate. The learning rate is important since it will help the model quickly find a good set of weights.

📝 Note: There are a few optimizers that automatically tune the learning rate. We do not need to know the details of each optimizer. However, we need to choose an optimizer that serves as a good example for many problems.

Adam (adaptive moment estimation) is a good choice. It adjusts the learning rate during gradient descent and ensures reasonable values for the weights during the weight optimization process.

2. Loss function

The loss function specifies the loss during model training.

Regression

Mean squared error is the most common choice for regression problems.