Course Practice Problems Flashcards
Describe the Bayes Classifier
The Bayes classifier is a classification procedure that reaches the true minimum misclassification rate. It can be thought of as the underlying model that generates the true categories of the observations.
Will the Bayes Classifier result in 0 misclassifications?
It can, but its highly unlikely
Under what assumptions is LDA the Bayes Classifier?
If each group (or subpopulation) is assumed to be mul-tivariate normally distributed, and all groups have a common covariance matrix
Under what assumptions is QDA the Bayes Classifier?
If each group is assumed to be multivariate normally distributed with uniquecovariance matrices
Describe the difference between clustering & classification.
Clustering is attempting to separate observations into groups according to the predictors (X) — there is noknown response (Y) that we are actively modelling, it is an exploratory procedure.
Classification is the process of fitting a model using predictors (X) to predict a categorical response variable(Y)
what is a tuning parameter & suggest a common approach to optimize it
a set values to help tune your methods
train test RF
Random forests use the nonparametric bootstrap and build trees for each bootstrapped sample. As such,each tree will be fit on approximately 2/3 of the data. So, conveniently, misclassification rates (classification)or MSE (regression) can be calculated using out-of-bag observations.
PCA vs NMF
Principal component analysis (PCA) is an orthogonal rotation of the original data. There is no dimen-sionality reduction or loss of information (unless we remove components) or assumptions placed on thedistribution of the data (aside from it being numeric). In practice, we often need to scale the data.Non-negative matrix factorization (NMF) assumes that the observed dataXis non-negative, and can bedecomposed into two non-negative matricesWandH. The dimensionality ofWandHhas to be pre-specified and so there will be a loss of information. In contrast to PCA: solutions are non-unique, there isno inherent ordering to the basesH(which are the NMF equivalents to components from PCA), and thenon-negativity assumption results in reconstructions ofXthat canonlybe additive.
P value definition
fill
4 steps of K means
1.
2.
3.
4.
Whats a hidden layer in a Neural Networks
artificial neutron inbetween input n output, takes in a weighted output and provides and output from an activated function
Describe support, confidence, and lift from association rule mining.
fill
Describe support, confidence, and lift from association rule mining.
Support : the proportion of time that the item is in the dataset
Confidence: probability that if LHS then RHS
Lift:
Describe what multidimensional scaling achieves. What is its relationship to principal component analysis?
PCA: clusters groups based on correlation
MDS: clusters groups based on distance
they have the same analysis, so you could reverse
