ISYE 6501 MIDTERM 1 EXAMPREP TESTED
QUESTIONS AND DETAILED SOLUTIONS
◉ Soft classifier.
Answer: (SVM) Observations separated as well as possible by the
classifier line
◉ When the classifier line is vertical....
Answer: (SVM) only x-axis matters in determining class
◉ When the classifier line is horizontal....
Answer: (SVM) only y-axis matters in determining class
◉ What is the goal of maximizing the space between support
vectors?.
Answer: minimize classification error (more generalized model)
◉ In terms of the equation, how do we maximize space between
support vectors?.
Answer: minimize Σ(aj)^2
◉ In the. equation for a classifier line, if aj is close to 0....
,Answer: the attribute is not relevant
◉ As λ increases....
Answer: (SVM) more emphasis placed on maximizing the margin
◉ As λ decreases....
Answer: (SVM) more emphasis placed on minimizing classification
error
◉ How do the relationships of λ differ when using C in
computation?.
Answer: (SVM) C has inverse relationships to λ. As C increases,
minimization of error is prioritized. As C decreases, maximization of
margin is prioritized
◉ What is a weakness of randomly selecting data for
test/train/validation sets?.
Answer: possible to have clumps of data
◉ What is a weakness of the rotation method of selecting data for
test/train/validation sets?.
Answer: can introduce bias if there are repeating patterns in the
data (i.e., days of the week)
, ◉ If a model requires a bounded range, you need to....
Answer: (scaling) normalize
◉ What types of models prefer standardized data?.
Answer: PCA, clustering
◉ What is a possible effect of using unscaled data in a classifier
model?.
Answer: introduce bias into the distance metric. Changes in features
with larger ranges will have a disproportionate impact
◉ When should you use KNN over SVM?.
Answer: >2 classes
◉ What is the KNN model sensitive to?.
Answer: noisy predictors and near-0 variance features
◉ KNN: A low k-value may _______.
Answer: overfit
◉ KNN: A high k-value may _______.
Answer: underfit
QUESTIONS AND DETAILED SOLUTIONS
◉ Soft classifier.
Answer: (SVM) Observations separated as well as possible by the
classifier line
◉ When the classifier line is vertical....
Answer: (SVM) only x-axis matters in determining class
◉ When the classifier line is horizontal....
Answer: (SVM) only y-axis matters in determining class
◉ What is the goal of maximizing the space between support
vectors?.
Answer: minimize classification error (more generalized model)
◉ In terms of the equation, how do we maximize space between
support vectors?.
Answer: minimize Σ(aj)^2
◉ In the. equation for a classifier line, if aj is close to 0....
,Answer: the attribute is not relevant
◉ As λ increases....
Answer: (SVM) more emphasis placed on maximizing the margin
◉ As λ decreases....
Answer: (SVM) more emphasis placed on minimizing classification
error
◉ How do the relationships of λ differ when using C in
computation?.
Answer: (SVM) C has inverse relationships to λ. As C increases,
minimization of error is prioritized. As C decreases, maximization of
margin is prioritized
◉ What is a weakness of randomly selecting data for
test/train/validation sets?.
Answer: possible to have clumps of data
◉ What is a weakness of the rotation method of selecting data for
test/train/validation sets?.
Answer: can introduce bias if there are repeating patterns in the
data (i.e., days of the week)
, ◉ If a model requires a bounded range, you need to....
Answer: (scaling) normalize
◉ What types of models prefer standardized data?.
Answer: PCA, clustering
◉ What is a possible effect of using unscaled data in a classifier
model?.
Answer: introduce bias into the distance metric. Changes in features
with larger ranges will have a disproportionate impact
◉ When should you use KNN over SVM?.
Answer: >2 classes
◉ What is the KNN model sensitive to?.
Answer: noisy predictors and near-0 variance features
◉ KNN: A low k-value may _______.
Answer: overfit
◉ KNN: A high k-value may _______.
Answer: underfit
