Hello, and thank you for taking the time to look at this code. There is a lot of commentary throughout this repository and I hope you find it helpful in understanding my thought process. Before we jump into it, I'd like to thank Kay Ehni for her incredibly succinct Ruby example (though I think you got Q5 wrong ... sorry!) and Rob Simmons for posting your results - this was exceptionally helpful validation! I'd also like to thank Ad Hoc for designing and publishing this excellent challenge.

Hi! I'm Peter Edstrom. I build teams to solve complex software problems. I see myself as a straightforward and methodical team player. I have about 20 years of software experience ranging from UX to data science to machine learning. As an engineering leader, I work to actively bring about positive change in technical organizations.

I have a LinkedIn Profile and you can reach me by email: [email protected].

With the preliminaries out of the way, let's start unpacking this challenge. I've organized the code as follows:

• byte-reader/ - These are the original challenge files, as presented to me.
• 02-proto.Rmd - The R Markdown Notebook that contains my thought process while coding the solution, as well as the final code.
• 02-proto.pdf - This includes the code, as well as the execution results from RStudio. There is also an HTML version, but if you are not viewing this locally 02-proto.nb.html may show you the html source instead of a rendered page. I'd recommend downloading the file and open the local copy in your browser of choice.
• Commits - Not only do you get all of my commentary in the aforementioned notebook, but you get all of my commit messages as well!

There was no requirement to choose a specific language. Python, Ruby, JavaScript, and Java were called out as possibilities, but I chose to address this challenge using R Project.

There are a few reasons for this choice:

• Data science is an active interest of many people right now, and the scientists seem to favor either R or Python languages.
• I was not able to find a pre-made solution in R through googling.
• I didn't have any idea if R had binary operators that could handle this challenge and I was curious to find out.
• Notebooks - a technology available in R and Python (plus a few others) - is an excellent way to document a thought process.
• And I like R. It's been around for a long time ( 24 years! ) and I feel it is an under-appreciated tool given the power it brings to processing data.

Before you ever write a single line of code, it's best to think through the problem ahead of time and conduct some research. I'm not talking about the architecture of the solution, but the strategy in which the architecture will be decided. These can be found in questions like: How am I allowed to solve this problem? If I find another solution I can copy & paste, is that considered cheating? What resources are at my disposal? Can I pair-program on the solution? What constraints exist?

Let's talk about the constraints and then do some research on previous solutions.

• I was given 6 calendar days to complete the challenge. I could request additional time if needed.
• My primary computer for the last 15 months has been an iPad Pro. While the iPad has been sufficient for more than 90% of my computing tasks, coding is one area that it falls short in. I needed to make a shift to address the coding challenge.
• My secondary computer is an aging 2011 iMac. This is a shared computer used heavily by all of the other members of my family.

You might ask why I even considered searching Google for previous solutions when it may be construed as plagiarism. Software development at it's best is a collaborative effort and I was curious about other approaches. However, the solution I have put forward is nothing like any of the published solutions.

I was also aiming to answer a simple question: If the answers are all that matter, are those answers available in a trustworthy form without writing any code? I found:

• a potential Elixir solution.
• a potential C solution which also have final results published.
• a potential JavaScript solution.
• a potential Ruby solution.

Notice I'm saying "potential". None of these are guaranteed to work, nor are they guaranteed to have correct answers.

Of the 4 solutions, only the C solution had published the results of their code. I wanted the results of the Ruby solution as well so I downloaded the code 01-sample-proto.rb and ran it locally (Ruby version 2.3.3p222). The corroboration of the two results against each other follows next.

C solution published results:

Reading version 1
Expecting 72 records

All transactions of user 2456938384156277127:
Debit  248.584932 for 2456938384156277127 Sun Jul 13 16:17:53 2014
Credit 248.584932 for 2456938384156277127 Sun Jul 13 16:17:53 2014
User's balance is: 0                   

Summary
Total debits:     18203.699533
Total credits:    10073.359933
Autopays started: 10
Autopays ended:   8

Ruby solution local-execution results:

{:question_1=>18203.69953340208, :question_2=>10073.359933036814, :question_3=>10, :question_4=>8, :question_5=>248.58493153382494}

The Ruby answers match the C answers with a few exceptions:

• The ruby code appears to have more precision on question 1 and 2.
• They differ on the answer to question 5 (0 vs 248.58493153382494)

Given these two samples, I'm inclined to believe that we have found answers to the initial 4 questions. Using only the digits that are in agreement:

• What is the total amount in dollars of debits? Answer: 18203.699533
• What is the total amount in dollars of credits? Answer: 10073.359933
• How many autopays were started? Answer: 10
• How many autopays were ended? Answer: 8

I’d like to refine the first two answers slightly. The challenge README.md does specify that the amounts are in dollars. So it would be appropriate to display the results in dollars. I favor the traditional format used in the United States which includes the unit “$”, a comma separators for thousands, and round to the nearest penny. So:

• Debits? $18,203.70
• Credits? $10,073.36

But then again, both coding examples may be wrong. I find it suspect that any amount recorded in dollars would have precision past the hundredths place.

There are instances where fractional pennies are important (for example, the cost of a gallon of gas) but it is equally likely that the fractional pennies are simply an artifact of translating a base-10 number into a base-2 representation. It might be more appropriate to round each amount as it is extracted from the data file - prior to any calculations - instead of after calculations are completed.

To sum up where we are right now in the thought process:

• Rounding questions aside, I have pretty strong confidence that we have correct answers for the first 4 questions.
• There are two contending answers for the 5th question.
• I have not yet written a single line of code.

During my research, I found that the original authors of the code challenge are from a company called Ad Hoc. They provide a quick tip worth including: “We want to see how you approach the problem, understand what's being asked, and solve it. We tend to favor solutions that are simple, elegant, and efficient. Submit a solution as if you were giving it to a colleague for code review.”

This seems like a good approach and is one I'll be thinking about as I create my solution in R.

I had a few tasks to complete before writing any code:

• Update my local installation of GitHub Desktop.
• Set up a GitHub repository for the project.
• Install RStudio, which gave me a run-time error Unable to locate R binary by scanning standard locations. I was able to resolve by installing R 3.4.3. Whoops!

From here on out, please follow my thought process in 02-proto.Rmd. The pdf and html files will also include the code-execution results.

Thank you for taking the time to review my approach to this challenge!

Peter Edstrom [email protected] January 2018