Winter Break, Finals, and SpecJBB!!

Hello everyone!

We hope everyone had a great Thanksgiving Break (mmm food…) and we hope you will have a great Winter Break (CHRISTMAS AGH) too. However, we should get some specs out of the way first.

Jolie and I are going to use this week (probably not next week because finals) to completely finish the report and make it spiffy and fantastic. All we really have to do is read over our analyses, make sure they sound educated (I really need to do that), and make sure it is production ready. It’s not too much to do, but it will keep us busy.

Also, now that we have SpecJBB, Jolie and I are going to spend winter break running the benchmark on our machines with different input sets. We hope that we will have all of the traces recorded before break ends so that next semester we’ll be ready to start using those traces along with all of our other ones and basically redo everything we did this semester with the new data set. Yay (not yay).

We were only able to get SpecJBB set up thanks to our lovely colleague, Lowell. Thanks, Lowell!

Well, as short as this post is, I’m going to have to cut it. Toodles!

Hidden Under Tables

Hey everyone,

So Jolie and I have been putting all of our data into tables for our report and it is SO INCREDIBLY TEDIOUS. Thankfully, we’ve been doing analyses along the way, so we’re not entirely behind on figuring out what we’re working with. Jolie and I split up the data so that she’s looking at across system results and I’m looking at by system results.

Remember the question we need to answer: What exactly does our data mean in the context of our problem?

Our problem is classifying power traces into their proper groups regardless of system, input size, and duration. We’re using Random Forest to classify our traces and we have different feature vectors to help Random Forest to narrow the options that it uses to create each tree.

Now, for my data, I’m pretty sure that there are some features or combinations of features that have more weight than others on each trace. It’s hard to tell when we look at all of our options (MMSD + one other feature). However, Jolie ran our traces through Random Forest again using all of the features (which we call Stat 14) and it seems that my conjecture is true. Now, when we have just MMSD + one other thing, there aren’t that many misclassifications (at least among LC groups). But when we have the entire stat 14, the misclassifications are more leveled out between RF and LC groups.

The problem there is that there are so many different combinations of three features that trying to calculate them all for the number of traces that we have is damn near impossible. Well, not impossible, but that’s a lot of extra work.

Anyway, I’ll leave the rest of this post for Jolie, since I haven’t looked at her side of the data yet.

Toodles!

 

Analyzin’, Reportin’, Halloweenin’

Hey everyone,

Jolie and I are just about to throw ourselves into the piles of data we’ve collected. Although we’re not completely finished gathering it all up, we figured it’s high time we start analyzing what we do have already. We would waste precious time gathering every single bit of data we might need (though, trust us, we will do that) when we should be figuring out exactly what it all means. We’re also going to take our findings and shove them all into a report so we can look at it later if we forget things.

Personally, I feel like this is the hardest and most rewarding part of the research process. It’s totally easy to sit back and let calculations run, but to take a step back and tell yourself, “Oh, this means that” is actually pretty difficult. It’s like doing a math problem — when you know the formula, evaluating the equation is just a matter of plugging things in. But once you get your answer you have to figure out what it means in the context of your problem.

So it is with all of the data we’ve collected about our power traces. The calculations themselves only took the effort of plugging in the right inputs. Now, we have to step back from being completely immersed in our problem to answer the broader question: What does it mean?

Unfortunately, we don’t yet have the answer. When we do, it’ll all be in our report which, I hope, we can make available to you.

On the upside of all this, Halloween is this week! Jolie is dressing up as herself (which is cool, be awesome, but DRESS UP IT’S SO FUN) and I’m going to dress up as Sherlock Holmes from BBC. I hope everyone has a fun night of trick-or-treating while we’re trudging through all of our data!

Toodles,

Rachelle & Jolie