Share good ideas and conversation.   Login, Join Us, or Take a Tour!
comment by veen
veen  ·  297 days ago  ·  link  ·    ·  parent  ·  post: Pubski: September 27, 2017

Finally, the beginning of the end.

Of my master's study group that we formed two years ago, we now have the first graduate among us. I went to her thesis defense this morning. Much like a PhD defense, master students also defend their thesis at my university as part of their graduation ceremony. She did great! Only a month or two and it's my turn.

Over lunch, I caught up with someone I've known since 2006. Never as a friend, always as an acquaintance - imagine a Venn diagram with 'socially inept' and 'mildly annoying' and you can put him in the middle. He went to the same high school as I, but was in a different class. He did the same bachelor's degree as I, but made different friends. And now he's attempting the same master's degree that I am finishing. He even lives in the same block of buildings. Whereas I am almost done after 5,5 years of higher education with two degrees, he was already delayed with his bachelor's and has managed to cobble together less than ten percent of all credits in more than a year. His situation is exactly what I feared when I started my master's. It's a bit like looking into mirror of what I feared when I made the jump two years ago.

On the one hand I pity him, but on the other hand, I think he should know himself better. But then again, it's not like I was super confident two years ago...nor am I confident about what to do next. A PhD position opened up which I am interested in. The professor is nice, the topic is 'public transport data science' (choice models and forecasting mostly) and they got the Amsterdam transit agencies to provide and help with data. So I can seamlessly continue to build my data science expertise in a topic that I like, while continuing to live like I do now (but with a salary instead of student loans). But maaan...four years of full-time is a long ass time. That's until 2022, which seems decades away.

Anyway, this week I finally finished my code. As in, it now includes everything that I want (including the sensitivity analysis) and when run on a 22-core remote beast it only takes a few hours to run all my scripts. What's left is a bit of refactoring and commenting. The output is 30 CSVs, each of them with 158 indicators for 1192 areas, which is what I wanted.

So finally my cool maps and graphs are actually correct! And the best thing is that everything seems to work as expected. My reasoning is sound, my data is sound and my results look sound.

Tomorrow I present these initial findings to a bunch of colleagues at my internship. Next Tuesday my thesis committee gets the same story. Hopefully after that week I can focus on writing the thesis itself and making it look cool.

Cumol  ·  296 days ago  ·  link  ·  

Can you explain your figures? I have no clue what I am watching :D

veen  ·  296 days ago  ·  link  ·  

Sure! I'll explain bottom-to-top, that makes more sense.

My thesis is about accessibility in cities, and how it is distributed over cities. Accessibility here means "how many useful places can you reach in a reasonable time". This depends on whether you travel by car, PT or bike so I distinguish between those. Anyway, the lowest image shows most of my study area. Each zone in my study area (which is one city) is a few blocks of houses big. I've coloured it depending on its accessibility score for bikes, where green is very good (relatively) and blue is very bad. As one would expect, centrally located zones are green and edge zones are blue. However, there are a bunch of patterns you can see already - like the blueish spots to the left and right of the center. Those reflect geographical barriers.

In the middle picture, I have mapped each zone four times depending on how good they score on two factors: accessibility and mobility. Accessibility is the vertical axis, mobility the horizontal axis. Up and to the right is where you want to be. Each zone in my study area has an accessibility and a mobility score, so they each get a place. Because these scores depend on the mode of transportation, I have mapped each zone four times (the four colors you see). This allows me to see what would be an 8 by 1200 table in one graph. The thick black lines represent the average car accessibility and mobility, with the dotted lines representing 75 and 50% of that average.

The top graph is a different way of making the same, inter-modal comparison of accessibility and mobility. The axes are the same, but instead of drawing points, I draw a line from public transport accessibility to car accessibility. In other words, how much better would your accessibility be if you had a car or not? As you can see, even in a city with amazing public transport (compared to the US), you would totally gain a lot by switching to a car. That graph also shows a cool asymptotic effect - when you already live in an area with good accessibility, your gains when switching modes would be more in the domain of mobility and not actually in the domain of accessibility. So you get to your destination faster, but you don't really get to good new places.

The cool thing—in my opinion—is that I have a treasure trove of information and a whole lot to say about it. :)

Cumol  ·  293 days ago  ·  link  ·  

Took me a while to understand the middle graph but maybe my eye is not used to it.

Do I understand that correctly that public transport is actually not that bad compared to a car (as the red dots cluster not so far from the car dots)?

Another question, is parking possibilities etc. taken into account when talking about car accessibility etc.? In Tel Aviv for example, the parking situation is so bad (really bad) that many people are switching to electronic bikes in the past 2-3 years.

veen  ·  293 days ago  ·  link  ·  

When I present what I have, I usually use a bunch of different slides to slowly build up to that middle graph - so it's not just you.

    Do I understand that correctly that public transport is actually not that bad compared to a car (as the red dots cluster not so far from the car dots)?

Yep! Remarkably close is my observation. My study area is highly urbanized and public transportation is generally really good here, so I would point to those as explanatory factors.

The time it would take to park your car and walk to your destination is accounted for partly - it's taken into account as a rigid amount of time on top of the actual travel time.

Cumol  ·  291 days ago  ·  link  ·  

Thanks for the info! Really cool stuff :)