A parametric model assumes a functional form for the data and learns a fixed set of parameters. Linear regression, logistic regression, and neural networks are examples. The parameter count is chosen in advance rather than growing with the dataset. section 3.3.1
A typical parametric-model workflow separates learning from prediction:
- During training, a loss function measures individual prediction errors, a cost function aggregates them, and a learning routine adjusts weights to reduce the cost. section 3.3.1
- After training, the weights are frozen. During inference, the model applies them to previously unseen data without further learning. section 3.3.1
- Generalization depends on learning from representative samples rather than merely fitting the available training points. section 3.3.1
Predicting hERG blocking
A model can map chemical inputs to a yes-or-no prediction about hERG blocking. Training estimates weights from labeled compounds; inference uses the frozen model to score new compounds. section 3.3.1


