ECS 132 Term Project

Important dates

• Progress report: Due Monday, March 8, 11:59 pm, in the form of an e-mail message to me. Must state the contribution of each group member so far.
• Submission: Scheduled final exam day (no written final), Wednesday, March 17, 11:59 pm. NO LATE SUBMISSIONS.

Downloads

In Problem B, you will be using my regtools package. There are various ways to install it on your computer, e.g. for a Unix-family system.

git clone http://github.com/matloff/regtools
mkdir ~/R
R CMD INSTALL -l ~/R regtools

and then add this line to your R startup file, ~/.Rprofile:

.libPaths("/home/yourusername/R")

Note that regtools in turn requires several other packages to be installed. If you don't have them already installed, you'll get an error message telling you which ones you need.

For Problem B, you will also need to download the Porto taxi trip data. We will use only train.csv, which you can read using read.csv().

Problem A

("Scavenger hunt.") Each person in your group must find, and write a report on, a research poster somewhere in a campus building. The poster must use p-values or equivalents such as saying something like "The confidence intervals for the difference between two means does not contain 0."

Broadening due to pandemic: Many of you are not in Davis, so you can find a research paper online. It still must have a UCD professor as one of the authors.

In a paragraph or two, state the scientific topic being researched, and discuss how using confidence intervals instead of significance tests might result in a more insightful analysis, and discuss possible dangers reliance of p-values might bring in this case.

Each section must have the name of the team member who did the analysis for this example and wrote this section of your report. Again, note that there must be a different example written by each team member. The sections for Problem A should be titled accordingly.

You may wish to cite authority here, e.g. this announcement and its links.

Be sure to include the URL for the paper you analyze.

Problem B

As noted, this problem will involve analysis of the Porto taxi trip data. Here are some general rules:

• Note that many of the parts of this problem will be rather open-ended, using terms like explore and investigate. But when possible, formal statistical inference methods must be used.
• So, there is flexibility here, but note that (a) you are restricted to using the methods of our course and (b) you are restricted to using R throughout, including for generating graphs.
• For graphics, use either (i) base-R graphics, (ii) the lattice package or (iii) the ggplot2 package. Our book contains a tutorial on the latter, but there are tons of examples of all three on the Web. There are no specific graphical tasks delineated below, but you'll definitely need good pictures to write a good, professional quality report.
• The tapply() function is one of the most powerful in R (also, the related split()). You may find it useful.
• Do not use another package or statistical method without getting my approval.
• The UCI data repository is quite famous in the machine learning world. Thus, its datasets tend to have many analyses on the Web. You can probably find some for this Porto taxi data, and there are many, many analyses of the various NYC taxi datasets on the Web. You are very welcome to browse through these, but if you use anything that you find there, you must cite it in your report.
• It is imperative that you start early. There are many things to be done that you may not realize at the outset. You'll need to figure out how the software works, which can sometimes be challenging in the case of graphics. You'll encounter various idiosyncrasies in the data, and will need to figure out how to handle them. You'll find yourself wondering, "Well, in actuality, what does that statistical method do?" This is a rather large dataset; some computations may have long run times. Etc. We are happy to help you through problems you encounter, but clearly you need to allocate time to deal with these issues.

Here are your tasks:

• Let T denote trip duration. Explore finding a model for f_T from one of our density families.
• Let B denote the proportion of time a driver is busy, i.e. actually driving rather than waiting for the next fare. Explore finding a model for f_B from one of our density families.
• Investigate whether the type of call used to summon the taxi makes much difference in mean trip time.
• Develop models for predicting trip time from other variables, both the explicit ones and also trip distance. (This task has many subparts, and will probably take up most of your time.)

Details of the prediction task:

• First try a linear model.
  • Note that "linear" means "linear in the beta parameters." You can and should also try polynomial models, using qePolyLin().
  • Find an approximate 95% confidence interval for E(Y | X = u) (X, u vectors), your choice of u, for your straight line model.

• Then try two machine learning models, chosen from qeKNN(), qeRF(), qeLASSO(), qeGBoost() and qeNeural(). In each case, you may wish to try some nondefault values for the hyperparameters. (Some of the models have many hyperparameters, some of them very esoteric; you need not learn them all.)

  I suggest you use the rough draft of my forthcoming book to learn what these methods do. Or, there are many good, usually brief, tutorials on the Web. I will give brief overviews of ML methods in class.

• Compare the predictive power of the methods you used above (several linear models plus two ML).

Important General Rules

PLEASE FOLLOW THESE RULES 100%!

• Don't skimp on time devoted to the writing. The phrasing matters! This is true both in terms of having good, professional writing quality, but also in terms of proper statistical interpretation, which will be vital in your report.
• You must use LaTeX and R throughout.
• Submit your report, including all files, i.e. .tex, .pdf, R code (see below), any image files, etc., to my handin site on CSIF (NOT the TAs' sites), username matloff, directory 132project.
• Your LaTeX file must be named ProjectReport.tex, and the corresponding PDF ProjectReport.pdf. The grading script will look for these; don't disappoint the grading script!
• The name of your submitted file must be of the form email1.email2....tar , where each emaili is the UCD e-mail address of group member i, e.g. bclinton.gwbush.bobama.dtrump.tar. Note the periods separating fields. Don't get the address wrong! Otherwise the grading script may not give someone credit.
• Your .tar file must contain only regular files, NO SUBDIRECTORIES!!!! And .tar does NOT mean .tar.gz or .tar.bz2 (or for that matter .rar, which one student used once). The grading script will execute

  tar xf youraddresses.tar  # unpack
  ls ProjectReport.tex ProjectReport.pdf  # check for proper files
  xpdf ProjectReport.pdf  # view PDF
  

  Note that the script will NOT do cd.

• Place all your code in a file TermProject.R, as well as in an Appendix to your report (LaTeX \appendix \section{}). I may execute your code, so make sure it is runnable.
• Absolutely NO late reports will be accepted. As you near the deadline, keep submitting what you have (each one will overwrite the last), so that at least you will get a lot of credit even if you don't finish.
• Include a section listing each team member's contribution to the project -- who did what. If a member did not participate, do not include him/her in this section, nor in the .tar file name. Don't forget this section!
• DOUBLE-CHECK THAT YOU ARE MEETING ALL SPECS! Whoever does the actual turning in of your submission, impress upon him/her that this is a huge responsiblity that can affect everyone's grade.

Grading

General commnets:

• Groups that put in a reasonable amount of time -- and thought! -- almost always receive an A or A+ grade on the project. Groups that do not complete the project usually get a D grade. PLEASE START EARLY!
• As explained in class, groups that do good work on the project receive an extra bonus in their course grades, beyond what your quiz and homework grades are. The boost is always at least one notch (e.g. B to B+) and typically two notches (e.g. B to A-). E.g. a student could have strictly B work in the homework and quizzes and yet still get an A- in the course.
• A+ grades are very possible, and can have a significant impact on your course grade, letters of recommendation, knighthoods, marriage prospects, coronations, etc.

Criteria:

• Technical content of the work (correctness, thoroughness etc.).
• Adherence to instructions.
• Professional quality of the work: Clear, engaging writing, using correct grammar; it need not (should not) be pretentious, but avoid being too colloquial ("the mean was kinda low"). Presentation need not be fancy, but graphs and tables should be used where helpful.