2021 - Week 10

Another year, another DREU! I didn’t get quite as far as I wanted to, but considering everything that’s going on with Covid, and that I’ve spent an entire year now virtually. That’s a lot to deal with! Last year I really excelled at the online format, but this year I think I struggled. I’m an introvert and a self-starter, but I do think I need some form of in-person interaction, or at least an office to go to that has a giant whiteboard. Never underestimate the power of a change in scenery!

What’s next after DREU? Well I have two conferences to prepare for! I’m presenting my research as a poster at both the Tapia 2021 and the 2021 Grace Hopper Celebration in the next couple of weeks. I’m so excited! I’ve been brushing up my science communication skills and I’m so pleased with my poster. I am also using this project to apply to the 2022 NCWIT Collegiate Award! My pre-classifier from last year earned me an honorable mention. I wonder what this project will earn?

Once again, I’ve had a wonderful experience with DREU and I cannot recommend it enough. It’s a great program and I couldn’t have gotten as far as I’ve gotten without them.

Big thank you to DREU, the Computing Research Association, and AccessComputing!

Read More

2021 - Week 9

I wish I could say a whole bunch of work got done this week and I built a model, but unfortunately that’s not the case. I did, however, read a whole lot more and I am now working out the math of my deep learning model. I know I want to use Keras, have a convolutional neural network, and have pooling layers. However, there is a lot of prep work. All the input data needs to be the same “shape”, which means we have to force our data to be the same shape. We also have to make sure our down sampling is done in a way that works with the data. There’s a lot going on and a lot to think about, and I’m not sure I’ll be able to get a model done by next week.

Read More

2021 - Week 8

This week has been a lot of studying! I took a deep learning class last spring, the first full Covid quarter, but the sudden change in modalities and the new learning curve that came with it meant that I didn’t get as much out of it as I wanted. However, there are a lot of articles and books out there! Although I don’t think I’ll use their Python module, I’m reading the machine learning section of AstroML. It’s a great book and easy to read! Their examples and graphs are fun, and it’s making me consider using their module. Another book I’m reading is “Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow” put out by O’Reilly. I recognize a lot of what this book covers from my deep learning class, which is making the book a lot easier to read. I like Keras and I think that’s the module I’ll be using.

Read More

2021 - Week 7

During the TESS Science Conference, we found out that there’s another group of scientists working on a catalog of eclipsing binaries. We don’t know what methodology they’re using, but it’s likely they’re also doing a form of machine learning. We’ve heard rumblings that this was happening, but this week was when it finally settled in that someone else might beat me to the goal I’ve been working for for a little more than a year. To say I’m bummed is an understatement! As someone with a disability, I work a little slower than others. Will this happen a lot in my future? I don’t know, but I also don’t feel like my research has been worthless. The likelihood of both of us using the same type of machine learning model with the same parameters is slim to none, so I’m going to keep going!

Read More

2021 - Week 6

You might notice a time skip between weeks 5 and 6! Last week was quite hectic for me. This whole summer has been stressful, far more than last year, but it hit a breaking point last week and I had to take a break. Luckily I’m working with wonderful people who fully understand burnout!

This week I took it slow because I didn’t want to aggravate any lingering burnout. And by “slow” I mean… I got to attend a virtual conference! The TESS Science Conference II was held from August 2nd through 6th and it contained loads of helpful workshops and panels and discussions. I even presented a poster on my pre-classifier from last year! Conferences like these are so important for undergrads, and making it online meant that I was able to attend. If it was in person, I’m not sure I would have had the energy for it. Since Covid is still here, I think virtual will be the option for a while, but I hope conferences in the future keep online options available.

Read More

2021 - Week 5

I have a list of periods that are decently accurate! I could spend time manually fixing them, but that is a lot of time. The next step is using the periods to isolate the eclipses (both the primary and the secondary) and clip them to a window size with the maximum eclipse depth in the center. Since eclipses can vary even within the same system, taking snapshots of each eclipse gives us more data to train on. If we have a light curve with 10 eclipses in it, and we isolate all 10 eclipses, then we now have 10 eclipses instead of one light curve. This allows us to slide through a light curve comparing each eclipse in the training data with the data points in the curve! Plus it can give us more accurate results when the model determines the eclipse depth of the light curve we feed it. :)

Read More

2021 - Week 4

I often wondered why machine learning projects took so long, and I think I have the answer: the label generation. Trying to fit a gaussian curve to the data and then finding differences is tricky! I spent the whole week trying to figure it out, mostly because it took a lot of math and computation, but I think I have it down now. I’m still working on determining a good window size for the data, but that isn’t too much of a problem. Eclipsing binaries are unique and their eclipses come in many different sizes, shapes, and frequencies! There might not be a good one-size-fits-all and I’ll just have to accept that.

Read More

2021 - Week 3

This week was spent building up the dataset that I want to train on. Based on the Stella paper, I will be picking out eclipses from the light curves and determining a “best window range” that will give the model at least one full eclipse to look at. To do this, I have to find a couple of things first: 1) what is the period and 2) what is the eccentricity. I also have to do something called phase folding, which is folding the light curve over onto itself after every period. If it works out correctly, the eclipses will line up with each “fold” and we will get two eclipses, the primary and the secondary. Pretty cool, huh?

So I calculate the period using the same statistical classifier I created last year, which generated two different periods (one from the Autocorrelation Function and another from the Box Least Squares) and then I graph the phase folding to make sure the period is correct. But which period is correct? Well, I’m still working on it, but the goal is to apply a gaussian curve to the phase fold based on the periods, and pick the one that fits the data best. That will take a bit of work, though. Wish me luck!

Read More

2021 - Week 2

Not a lot of programming got done this week, but a lot of literature review is still productive! For those looking back on this blog, this week was not a good one. A massive heatwave swept much of the US, especially the pacific northwest. For four days, I could not use my laptop during the day without overheating (and there are only so many chai lattes I can buy at coffee shops before I spend all my DREU stipend!). Luckily my mentor gave me several papers to keep me occupied last week, which made the heatwave somewhat bearable. Unfortunately, as soon as the heatwave lifted, I faced another catastrophe: my internet broke. Thankfully, I still had more reading to do, and I fixed my internet just in time for the fourth of July.

Read More

2021 - Week 1

Hello again DREU, I missed you! I had a fantastic summer last year! I did loads of research that made me even more excited about graduate school. I also presented that research at several conferences! The biggest win for my DREU project came from NCWIT, though… I won an honorable mention for the NCWIT Collegiate Award! They sent me a fancy trophy and everything!

Read More

Week 10

This is my last official week of the DREU program but I’m still going strong and plan to continue it through the school year, and into next summer. This program has made me so much more confident in my ability to research and I’m so thankful for the opportunity I was given. I can’t even begin to say how much I love what I’m doing, and the fact that I’m being paid to do what I love by a wonderful program makes it even better.

My classifier is almost ready for a final run through, but I’m a perfectionist who will still be toying with it for the next few weeks before classes start up again. However, we’ve built up a satisfactory training sample and we found a new direction to present this research. Instead of focusing on the ML model, we’re focusing on the creation of the sample training set. Only 2% of the light curves in TESS are the type we want to see. Can you imagine going through hundreds of thousands of plots to look for that 2%? That’s what undergraduates and new graduate students do all the time in astronomy and it’s time we make it better.

So, this research has a new spin and a poster title: Leveraging Statistical Analysis to Develop Labels for Astronomical Time Series Data.

Pretty snazzy, eh? A bit too wordy for my taste, but it was hard to convey what I wanted to! I’ll be presenting my poster at the Grace Hopper Celebration in October and I’ll be submitting it for approval to the American Astronomical Society for their annual winter meeting in January. Wish me luck! (By the way, DREU funded me for Tapia AND Grace Hopper!! I wish I could have done a poster at Tapia but at least I’ll get to show folks what DREU helped me accomplish at GHC!)

Finally, I also plan on using this research when applying for the NCWIT Collegiate Award. Working on this project and doing research for it made me realize that there’s not a lot of work being done in the development of time series training data. We need to ease the burden placed on new academics who are classifying these things by hand day in, day out. My research not only trained my eyes, but it helped me learn actual statistics and metrics of my data. That’s pretty stellar (pun intended) if you ask me!

So thanks again, DREU and Dr. Davenport, for the wonderful summer and I hope y’all will stick with me as I keep this research going.

Read More

Week 9

This week I’ve added 3 more iterations, not a lot more compared to last week but these iterations are deeper. I noticed that clipping wasn’t clipping the right amount of data. I’m still working on it, and this is the trickiest line of code I’ve ever had to work with. There’s no reason that I can see that the data wouldn’t be clipping like it should. It’s frustrating! But that’s how it goes.

Right now the classifier is identifying light curves on several thousand randomized TESS light curves, so I’ve moved beyond the test data and am testing the real thing. So far it’s flagging 1% of the light curves as eclipsing binaries, and that’s about half of what we’d expect to see. Still, when only 2% of your data set is the “true” class, finding half of them is a BIG deal! And it’ll make a future machine learning model work nicely.

Still, there are some false positives popping up, but not as many as before. I think once I sort out clipping, I’ll be able to cut this in half. At the very least, there are only a few dozen being picked up for a few thousand light curves. That’s not bad! (And ML models need to train on false examples that look similar to true examples anyway! Win win!)

Read More

Week 8

Woohoo! I have things to report this week! Granted, I probably spent…way more than 40 hours this week, but hey–astronomy and computer science are special interests! (And as an Autistic adult, diving headfirst into special interests for hours on end is my bread and butter.)

This week was all about making minor changes trying to perfect performance, both in computation time and accuracy. I brought back smoothing (and it works now) and added a ‘clipping’ to the data. Since most of the data errors are at the very beginning or at the very end of data collection AND look like eclipses, clipping the edges might help cut down on the false positives. I tried about five different iterations this week and they keep getting better! However, the clipping is causing some serious performance issues and I need to look into that next week.

Read More

Week 7

Unfortunately this was another slow week for me. Visiting my family for the first time in months, when I normally visit them weekly during the summer, was hard and left me feeling vulnerable. I appreciate the efforts our Governor is doing to ensure public safety, and I wish more Governors treated Covid as seriously as he is, but I can’t wait for all of this to be over so I could go back to my semi-regular visits. Stay healthy and mask up, everyone!

Read More

Week 6

I’ll admit–I didn’t get much work done this week. This week was all about comet hunting! Luckily my mentor (and virtually all astronomy buffs) were out and about looking for Comet Neowise, which was in the neighborhood this week. I got some excellent pictures on my Twitter: https://twitter.com/astronomyftw/status/1283310881095520256 https://twitter.com/astronomyftw/status/1285102900536946688 https://twitter.com/astronomyftw/status/1285470844433477633

It was gorgeous and so thrilling! It reminded me of the time when Comet Hale-Bopp visited us in the mid-90s. That little two-tailed comet was one of the reasons that I started looking to the stars for inspiration. As I grew older, that love never faded, and once I found out how important my first love (computing) is to astronomy? Well, it felt like destiny.

Read More

Week 5

Well this was a tough week for sure! I spent most of my time this week working on the cleanliness of my code and debugging a few NaN issues–nothing too terrible! I made a mistake that a lot of folks make and that I was implementing several changes at once without testing them individually to make sure they worked. This is fine, but it makes debugging a lot longer than it should be! Narrowing down the problem areas in my code was hard, but I was able to figure everything out with the help of my mentor cheering me on.

I also made the decision to remove smoothing on the light curves, but I’m thinking I will bring it back soon. I’m still not quite sure if I got the smoothing right! I guess I should explain what smoothing is: it is the process of smoothing out the sinusoidal patterns in the data. If the wavy curves are flatter, the eclipse dips stick out more. At least that’s the theory! I’ll work on it and see what I can do next week.

Read More

Week 4

I want to start this week’s blog post with a very important message: I am Autistic. My brain is wired a certain way and that comes with a certain set of problems, mainly with confidence and the need for structure. With Covid-19 wreaking havoc on routine, I was in desperate need of structure in my life and Dr. Davenport has been excellent about building that support for me. In education research, his method is called scaffolding and it’s hugely important for a lot of learners, especially learners like me. He gives clear, concise instructions that are neither too big nor too small that get me to where we both want to go. He doesn’t give me a large project and tell me to wing it, he gives me the necessary steps to get there and that’s the #1 reason why I feel like I’m rocking this DREU project.

Read More

Week 3

Wowee, this week was one of the hardest for me! While I have a pretty hefty statistics background, I’ve never used the Lomb-Scargle or Autocorrelation Functions before. I’d only heard them mentioned in passing and had no idea what they were actually doing. I’m glad I had this experience through DREU, I can’t imagine learning this in grad school when I’m juggling so many other responsibilities! But first, I should probably talk about why I’m doing this.

Read More

Week 2

This week focused a lot on reading, but it was extremely helpful. I learned that eclipses aren’t the only thing that cause a star’s brightness to decrease–starspots do, too!

Read More

Week 1

Whew, what a week! And where to start? This weekend was mostly dedicated to the onboarding process. I was able to get access to University of Washington IT services (and a UW email!) that will eventually give me access to the powerful network of computers that will absolutely trounce my workstation. Speaking of my workstation, it was finished this week–just in time for my DREU! My spouse built this water-cooled system that can handle a lot of data crunching and computing quickly and efficiently, although it can’t possibly compare to the computing power that the DIRAC institute has. I’m not there yet, so my workstation is more than enough, but I can’t wait to get to running analytics on a couple hundred thousand light curves.

Read More