Homework 6

ISyE 6416: Computational Statistics
Homework 7 (Last Homework)

(100 points total. )
• This homework is due on April 19.

• Please write your team member’s name is you collaborate.
1. Bootstrapping. (25 points)

Efron (1982) analyzes data on law school admission, with the object being to examine the
correlation between LSAT score and the first year GPA. For each of 15 law schools, we
have the following pair of data points:
(576, 3.93) (635, 3.30) (558, 2.81) (578, 3.03) (666, 3.44)
(580, 3.07) (555, 3.00) (661, 3.43) (651, 3.36) (605, 3.13)
(653, 3.12) (575, 2.74) (545, 2.76) (572, 2.88) (594, 2.96)
(a) calculate the correlation coefficient between LSAT and GPA.

(b) use the nonparametric bootstrapping to estimate the standard deviation and confi-
dence interval of the correlation coefficient. Use B = 1000 batches, and each batch
consists of N = 15 re-samples. For confidence interval, use α = 0.05.
(c) use the parametric bootstrapping to estimate the standard deviation of the correla-
tion coefficient. Assume that (LSAT, GP A) has bivariate normal distribution and
estimate the five parameters. Then generate 1000 batches of 15 samples from this
bivariate normal distribution.
2. Random forest for email spam classifier (25 points)

Your task for this question is to build a spam classifier using the UCR email spma dataset
https://archive.ics.uci.edu/ml/datasets/Spambase came from the postmaster and
individuals who had filed spam. Please download the data from that website. The collec-
tion of non-spam e-mails came from filed work and personal e-mails, and hence the word
’george’ and the area code ’650’ are indicators of non-spam. These are useful when con-
structing a personalized spam filter. You are free to choose any package and any language
to choose for this homework.
One would either have to blind such non-spam indicators or get a very wide collection of
non-spam to generate a general purpose spam filter. Load the data.
(a) (5 points) How many instances of spam versus regular emails are there in the data?
How many data points there are? How many features there are?
Note: there may be some missing values, you can just fill in zero.
1
(b) (10 points) Build a classification tree model (also known as the CART model). In
R, this can be done using library(rpart). In our answer, you should report the tree
models fitted similar to what is shown in the “Random forest” lecture, Page 16, the
tree plot. In R, getting this plot can be done using prp function in library(rpart).
(c) (15 points) Also build a random forrest model. In R, this can be done using li-
brary(randomForest).
Now partition the data to use the first 80% for training and the remaining 20% for
testing. Your task is to compare and report the test error for your classification tree
and random forest models on testing data, respectively. To report your results, please
try different tree sizes. Plot the curve of test error versus the number of trees used
in random forest, similar to our lecture.
3. Importance sampling. (25 points)

Similar to the example we had in lecture, using importance sampling to evaluate tail
probability of Gaussian random variable. Assume X is N (µ0 , 1). We want the right tail
probability α = P{X ≥ z}. For z µ0 , α is very small, and estimating this small
probability accurately is not easy. Now to improve accuracy, we samples from another
Gaussian random variable mean µ1 = z and variance equal to 1.
(a) Derive the important ratio.

(b) Write down the importance sampling algorithm
(c) Now assume µ0 = 1, z = 10. Using N = 100 Monte Carlo trials. Evaluate the tail
probability using direct Monte Carlo Ib1 , and using importance sampling using Ib2 .
Now repeat this 500 times, to compare the variance of Ib1 and Ib2 .
4. Bayesian inference using Metropolis-Hastings algorithm. (25 points)

Implement the Metropolis algorithm. Parameter for binomial distribution is probability
of success θ ∈ [0, 1], n = 20. Assume the observed data vector gives Sn = 5.
(a) Assume the prior distribution as in our lecture, π(θ) = 2 cos2 (4πθ). Generate samples
from the posterior distribution π(θ|Y ). Discretize θ to be a uniform grid of points [0,
1/10, . . . , 1]. Run the chain for n = 100, 500, 1000, and 5000 time steps, respectively.
For each time step, compare the empirical distributions with the desired posterior
distribution π(θ|Y ). (Hint: you may use ergodicity: hence the distribution of states
can be estimated from one sample path when the number of time steps is large (e.g.
500).)
(b) Following from the previous question, evaluate the mean of the posterior distribution
(this gives an estimator for the parameter value), and Eπ(θ|Y ) {[θ − 1/2]2 } = (θ −
R
1/2)2 π(θ|Y )dθ.

(c) Now assume the the prior distribution is given by π(θ) is a uniform distribution over
[0, 1]. Repeat the above questions, (a) and (b).

Homework 6

Uploaded by

Copyright:

Available Formats

Homework 6

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Homework 6

Uploaded by

Copyright:

Available Formats

ISyE 6416: Computational Statistics

Homework 7 (Last Homework)

• This homework is due on April 19.

1. Bootstrapping. (25 points)

(a) calculate the correlation coefficient between LSAT and GPA.

2. Random forest for email spam classifier (25 points)

3. Importance sampling. (25 points)

(a) Derive the important ratio.

4. Bayesian inference using Metropolis-Hastings algorithm. (25 points)

1/2)2 π(θ|Y )dθ.

You might also like