When their task environment has a clear understanding, you can use what you know to apply the tool to determine the problem to be solved and practical algorithm. Some factors that influence your choice model as follows:
• whether the model meets the business objectives
of how much data preprocessing model needs •
• How accurate model
of how the model interpretability •
how fast the speed of operation • model: structural model how long? Model to predict how long?
• How scalability model of
an important criterion for the complexity of the model is an influence algorithm selection. In general, a more complex model has the following characteristics:
• It relies on more characteristics of learning and prediction (for example, with all ten instead of two features to predict the target)
• It depends on the more complex features engineering ( For example, using a polynomial feature, interactive features, or principal components)
• it has greater computational overhead (e.g., require a random forest tree consisting of 100, rather than a single tree)
in addition, the same machine learning algorithms can be selected based on the number of parameters and some super parameters become more complicated. For example:
• regression model may have more features, or polynomial terms and interaction terms.
• decision tree can have greater or lesser depth.
The same algorithm becomes more complex increases the probability of occurrence of over-fitting.
2020-03-19 14:29:02
