Chapter 8: Summary of Models Flashcards
give the advantages of KNN (7)
simple intuitive no training classification and regression linear and non linear multiclass is simple as well only 2 decisions: k, distance measure
give the disadvantages of KNN (5)
slow with large data computationally complex memory cost bad at imbalance sensitive to outliers
give the advantages of regularised least squares (8)
popular inferred when people say 'regression' regression and classification efficient use of data- doesn't require too much easy to explain and understand low computational cost low memory no hyperparameter unless regularised
give the disadvantages of regularised leased squares (4)
linear only
sensitive to outliers
no probabilistic interpretation
sensitive to how classes are named
give the advantages of logistic regression (6)
simple and effective low computational cost low memory req probabilistic interpretation does not make assumptions about distribution no hyperparameters
give the disadvantages of logistic regression (3)
linear only
classification only
not good at multi-class
give the advantages of SVM (5)
competitive performance scales to high dimensions well can generalise linear and non linear not solved for local optima
give disadvantages of SVM (4)
need to choose a good kernel
need to choose hyperparameters
long training for long data
difficult to interpret
give advantages ANN (6)
non linear and complex functions can generalise no restrictions on input no assumptions about input learning stored in weights black box modelling
give disadvantages of ANN (4)
choosing architecture
long training for large data
difficult to interpret
local optimal
what does no free lunch mean
we can never know which will be better
we always need to know the data and experiment with different models
