- Review the relevant analysis solution example. (See bottom of this document.)
- Compare to your own solution.
- The example solution is not the solution. It is a solution.
- Write an analysis reflection.
- Use the provided “template” below.
- Stick to markdown when writing your document. (Consider organizing your thoughts under headers.)
- You may use
R
code as seen in the template. (Remove the specific template R
code.)
- Render to PDF.
- RStudio Cloud is automatically setup to be able to render to PDF.
- To knit to PDF locally, you will need to install LaTeX.
- Submit to Compass.
Grading
Grading will be based almost entirely on effort. (Point loss could occur for missing the deadline or submitting in a format other than PDF.) A member of the course staff will quickly review the document to determine if some thought was put into the writing.
- Minimum page length: 0.5 pages.
- Maximum page length: 2 pages.
- Supported formats: PDF
Notes
- These directions will change over time as we better understand how to prompt a good discussion of your work.
- Currently an attempt is being made to write this document to apply to all analyses, but in the future we may consider some specific questions.
- The end of the document notes specifics for each analysis.
- It’s possible that some thoughts below might not apply to every analysis.
- Fall 2019: Feel free to write about things not listed below. This is being written for the first time this semester, so we might use your responses to give us some ideas for prompts.
- Try not to spend more than an hour on a reflection!
- STAT 432 probably already has too much work…
Purpose
There are several reasons we’re performing these reflections:
- From the instructor’s perspective, it forces you to read a solution!
- It will hopefully reinforce the need to view these analyses in context.
- It will hopefully allow you to think of the analysis as more than just an assignment for a course.
- Hopefully it allows the analyses to be an activity where you are encouraged to take risks, without it being a detriment to your grade.
- Do an analysis, review your work, write a reflection. There is no need to get the analysis “correct.” It’s more important that you try something and learn from it.
- It will provide a system for students to give feedback about the analyses.
What follows are some loosely organized question that could spark some reflection. These are not questions that need to be answered. You do not need to write about each topic or question. (But you should on some level think about all of them when performing analyses and reviewing your work.)
Template
Consider using the following as the start of an R Markdown document when writing your reflection.
---
title: "Analysis Reflection"
author: "Your Name Here"
output: pdf_document
---
***
# Some Header
This is some text.
## A Subheader
- This
- Is
- A
- List
***
# Another Header
```r
# a code chunk
x = 42
```
***
Reflection Topics
General
- Did you find this dataset interesting?
- Do you feel that you had enough documentation to understand this dataset?
- Did the assignment seem useful? That is, did you get something out of performing this analysis? (Other than points.)
- Does this dataset seems useful? That is, did this dataset allow you to create a useful model?
Abstract
- Do you feel that a reader has a reasonable idea of what they are about to read?
Introduction
- Do you feel that a reader would have a reasonable idea what the data is, and why the analysis is being performed?
Methods
- Do you feel that a reader would have a reasonable idea how you have modeled the data?
- Do you feel that a reader would have a reasonable idea how you are validating your decisions?
Results
- Does your results section contain useful information? That is, you’ve reported your results, but only those that are most relevant to your decision and discussion.
Discussion
- Do you feel that you properly contextualized your results?
Effort
- Do you feel that you spent enough time on this assignment?
- Do you feel like your time was well spend completing this assignment?
- Are some parts of the analysis easier than others? Where did you spend your time when you did this analysis?
Modeling
- Did you make the same modeling decision as the example solution?
- If not, what did you do differently? Can you justify this difference?
- Do your numeric results “differ” from the example solution? If so, can you find a reason why?
- A difference here doesn’t imply something is wrong.
- How are you evaluating your modeling choices? Did you make any choices that seem unjustified?
Graphics
- Do your graphics appear to be publication quality? (Well labeled, easy to read, generally well formatted.)
- Quickly made EDA plots in an appendix somewhat excluded, but should still look reasonable.
Code
- How does your code compare to the example solution?
- Do you find the code in the example solution easy to read? Is yours easier to read?
- Do you see changes you could make to your coding habits and style?
- Does your code follow the
tidyverse
style guide? (With the known exceptions for STAT 432.)
Questions?
- Do you have any questions about an assignment, course materials, or course administration?
- If so, consider asking a few! Do so in the Piazza thread for the specific analysis in question. (See below for link.)
- We will attempt to answer as many of these as possible!
- Please try to not submit duplicate questions! (You should read questions from other students!)
Analysis 02 (Heart Disease)
- Did you view the original source data? Did you use additional source data? Extra variables? Extra observations?
- If you used additional data, did you encounter missing data? If so, how did you deal with it?
- Did you do any additional EDA? (Beyond the very basic analysis in the quiz.)
- If so, did you notice anything interesting?
- Were you careful to consider the population that this data was sampled from when recommending this model for use? (It is sampled from four very specific locations, and is seemingly sampled from patients that were already seeking medical attention, in particular from cardiology.)
- It would be a terrible idea to use this model on randomly chosen individuals from the a big population, for example, the United States.
- How did you evaluate your models? What metrics did you use? How did you define and weight false positives and false negatives in your decision making?
Questions Thread: [ @193
] Example Solution: [ .html
] [ .Rmd
]
Analysis 03 (Credit Card Fraud)
- Did you use the full data or the provided subset? Did increasing the sample size make the performance better?
- How did you quantify the performance of your model?
- Did you utilize the loss values given in the quiz? Did you consider the amount of the transaction in some other way?
- Did you consider the time it takes to make predictions?
- Note that in the “solution” there are some meta “instructor notes” at the bottom.
Questions Thread: [ @248
] Example Solution: [ .html
] [ .Rmd
]