What's an algorithm? What's a model?

Before we pe into data munging, let's take a moment to explain the difference between an algorithm and a model, two terms we've been using up until now without a formal definition.

Consider the simple linear regression example we saw in Chapter 1, Introduction to Machine Learning and Predictive Analytics — the linear regression equation with one predictor:

Here, x is the variable, ? the prediction, not the real value, and (a,b) the parameters of the linear regression model:

• The conceptual or theoretical model is the representation of the data that is the most adapted to the actual dataset. It is chosen at the beginning by the data scientist. In this case, the conceptual model is the linear regression model, where the prediction is a linear combination of a variable. Other conceptual models include decision trees, naive bayes, neural networks, and so on. All these models have parameters that need to be tuned to the actual data.
• The algorithm is the computational process that will calculate the optimal parameters of the conceptual model. In our simple linear regression case, the algorithm will calculate the optimal parameters a and b. Here optimal means that it gives the best predictions given the available dataset.
• Finally, the predictive model corresponds to the conceptual model associated with the optimal parameters found for the available dataset.

In reality, no one explicitly distinguishes between the conceptual model and the predictive model. Both are called the model.

In short, the algorithm is the method of learning, and the model is what results form the learning phase. The model is the conceptual model (trees, svm, linear) trained by the algorithm on your training dataset.