Developing models

Model development is my favorite part of the DL life cycle (and probably 90% of data scientists’ favorite part too). The goal of model development is to build a model with minimum loss over training and validation datasets to prevent overfitting, which is done by searching for the parameters that best fit the model. A typical model development phase involves four different steps, as shown in the following diagram:

Figure 9.4 – Different steps of the DL model building phase

Let’s discuss each of the four steps briefly:

1. Selecting an appropriate algorithm: In this step, an appropriate DL algorithm is selected based on the problem that you are solving.
2. Model training: Once an algorithm is selected, the DL algorithm is provided with the training data, loss function, random hyperparameters, and objective metrics to optimize using backpropagation.
3. Model tuning: The model initially has random hyperparameters that...