Introduction to Statistical Learning
Save
What are parameters? (Bonus points for another name for parameters.) - (correct Answer) - Parameters
are another name for coefficients. They correspond to the "weights" of each input variable. So if you
have y-hat = e + b0x, b1x, b2x, etc., the parameters are the values of beta for each of the x's in that
equation.
What is the difference between a parametric and a nonparametric method? - (correct Answer) -
Parametric methods consist of two parts: 1. Assume the data conforms to a functional model 2. Use
training data to estimate the coefficients of each input variable in the model. A nonparametric method is
more "flexible" because it does not try to learn a function, it just learns whatever shape it thinks is
correct
What are i, j, p, and n when used as subscripts? - (correct Answer) - i: An individual observation. j: An
input variable. p: The total number of input variables in the dataset. n: The total number of observations
(points) in the dataset.
How many parameters are there in f-hat(x)? - (correct Answer) - As many as you have input variables,
plus one for the error value e. (/epsilon)
What are the pros and cons of parametric methods? - (correct Answer) - P: Interpretable, because you
can just look at what weights were assigned to which input variables (oh, it looks like zip code accounts
for a high percentage of a home value estimate); Don't need a ton of data; Easier to visualize. C:
Depending on the problem, sometimes not as accurate as nonparametric methods.
What are the pros and cons of nonparametric methods? - (correct Answer) - P: Very flexible (low bias);
Often very accurate. C: Needs a ton of data; Not interpretable; Risk of overfitting (high variance, later
causing high test error or low generalizability)
What is meant by the "flexibility" of a model? - (correct Answer) - Flexibility means that a model is free
to learn predictive attributes of the data beyond what can be captured within the constraints of a
function. This results in higher accuracy, but also potentially overfitting.
Explain prediction vs. inference problems and give a business example and a science example for each. -
(correct Answer) - Prediction: The model or method seeks to successfully predict accurate values or
classifications on real (non-training) data. Inference: We seek to construct a model that informs us about
the data-creation process and/or about which input variables have the highest influence on the
response.
Save
What are parameters? (Bonus points for another name for parameters.) - (correct Answer) - Parameters
are another name for coefficients. They correspond to the "weights" of each input variable. So if you
have y-hat = e + b0x, b1x, b2x, etc., the parameters are the values of beta for each of the x's in that
equation.
What is the difference between a parametric and a nonparametric method? - (correct Answer) -
Parametric methods consist of two parts: 1. Assume the data conforms to a functional model 2. Use
training data to estimate the coefficients of each input variable in the model. A nonparametric method is
more "flexible" because it does not try to learn a function, it just learns whatever shape it thinks is
correct
What are i, j, p, and n when used as subscripts? - (correct Answer) - i: An individual observation. j: An
input variable. p: The total number of input variables in the dataset. n: The total number of observations
(points) in the dataset.
How many parameters are there in f-hat(x)? - (correct Answer) - As many as you have input variables,
plus one for the error value e. (/epsilon)
What are the pros and cons of parametric methods? - (correct Answer) - P: Interpretable, because you
can just look at what weights were assigned to which input variables (oh, it looks like zip code accounts
for a high percentage of a home value estimate); Don't need a ton of data; Easier to visualize. C:
Depending on the problem, sometimes not as accurate as nonparametric methods.
What are the pros and cons of nonparametric methods? - (correct Answer) - P: Very flexible (low bias);
Often very accurate. C: Needs a ton of data; Not interpretable; Risk of overfitting (high variance, later
causing high test error or low generalizability)
What is meant by the "flexibility" of a model? - (correct Answer) - Flexibility means that a model is free
to learn predictive attributes of the data beyond what can be captured within the constraints of a
function. This results in higher accuracy, but also potentially overfitting.
Explain prediction vs. inference problems and give a business example and a science example for each. -
(correct Answer) - Prediction: The model or method seeks to successfully predict accurate values or
classifications on real (non-training) data. Inference: We seek to construct a model that informs us about
the data-creation process and/or about which input variables have the highest influence on the
response.
