Saturday, April 8, 2017

Yaya's solo

It's easy to forget she's in 7th grade. As you can see from the picture, she's full adult sized now and has been for quite some time. She also plays better than all but the best High School students. This isn't just proud papa talking; she's been placed into every honors ensemble that she's eligible for.

Anyway, she played an adjudicated solo today and, frankly, it wasn't the best she's ever played it. In fact, it was a pretty average practice run. That's not to say it was bad by any means; she practices pretty hard. But, it wasn't turned up a notch like most of her actual performances are.

She obviously perceived this as well and seemed a bit despondent as we left the room. I told her that there will always be times when you don't perform as well as you know you can, but if you've really done your work, even less than your best can carry the day. The important thing is to not give up while there's still life in the game. To her credit, she definitely did not give up. She fought to the end of the piece and the last bit was probably the best part.

That last paragraph is somewhat fictionalized because I actually only got about halfway through the above statement when her accompanist ran up and told her she had gotten the highest score. Apparently, the judges are better at remembering that she's a Middle School kid than I am. By Middle School standards, she crushed it.

I thought back to my first Middle School solo competition, also as a 7th grader. The piece wasn't nearly as hard as the one Yaya played today. As beginner trombone pieces tend to be, it was a bit raucous and not only allowed, but encouraged some sloppiness. I got full marks, but I don't kid myself for a minute that I actually played it perfectly. Yaya's piece was a legit performance piece where any flaw was immediately exposed.

What impressed me most was that she spent several hours later in the day practicing. Most kids (and adults) would take a max score in a County-wide competition as permission to goof off for at least the rest of the day.

I've always wondered how things would have gone if I had chosen music instead of cycling. I'm pretty sure I know now. I had technique and I understood music, but I never had the passionate drive that Yaya shares with people who really get good. I would have been toast. What lies ahead for her is anybody's guess but only a fool would bet against her.


Friday, April 7, 2017

A tough week, but a good one

Going into last week, I noted that I was not finished on any of my major items. I'm still not, but I've certainly have made a lot of progress.

I turned in a reasonable stab at HW3 for Set Theory. It will need to be resubmitted after I get the comments back, but I wasn't embarrassed by what I turned in (though, I might be when I get the comments).

I've finished the query generator for Blocker and have a good direction for the evolutionary algorithm. I'm going to drop the Bayesian stuff for now and just use a heuristic. I give it 50/50 that I'll have something I can run by the time I leave for Boston.

At work we got through quite a bit of stuff prepping for production. Go-live is still tentatively scheduled for next Wednesday, though I'm not seeing how we get final signoff in time for that. That's kind of a bummer since it means I'll be in Boston when they turn it on. At any rate, now that we've got our Vertica instance running on a real cluster, we're getting billion-row queries back in 3-4 seconds, which is a big improvement over the old system. Load times are also faster by an order of magnitude. The old system loaded at just under 100 million rows per hour. The new one is around 1.2 billion and most of the time is extracting the data from Oracle; it will be even faster than that once the source system is writing directly to Hadoop. Given that the requirement is that we load 8 billion overnight, that increase is pretty crucial.

Thursday, April 6, 2017

Evolution

With the data layer and query generator done, it's time to get serious about the actual algorithm I'm going to employ for Blocker, my term project in Evolutionary Programming. Here are some thoughts.

The general flow is as follows: run a set of queries on the data and score the blocks. High scoring blocks get carried on without modification. Low scoring blocks are dissolved and their contents are copied back into the data base using the block assigner. The evolutionary part is that the block assigner changes with each iteration as more information is gained about the distribution of the data and the queries.

The novelty, (it's not much of one, but the term paper doesn't require any original work; just an implementation) is in the way the block assigner evolves. In a normal Evolutionary Strategy setting, the evolution comes by perturbing the data randomly and then comparing the fitness of the new generation to the old. That's too inefficient for this application because reblocking the entire database takes quite some time.

Instead, I'll have a prior distribution on blocking preferences. From the query results, a posterior will be computed and that will be used to set the distribution for preferences in the next round. For this to really work well, it has to take into account the correlations between attributes. I'm not sure if I can go beyond pairwise correlations for this effort, but even that should help quite a bit.

A couple other quick thoughts on how I can get generations through faster:

  • Time box the operation. That is, keep generating and running queries as until a fixed time has passed rather than a fixed number of queries. This allows early generations, which tend to see big gains, get done quickly. Later generations, where the improvements are much smaller, will get a larger query base because the query times will drop as a result of the optimization so far.
  • Change the sample threshold over time. Start with sparse sampling, maybe only actually reading 10% of the selected blocks. Increase the percentage of blocks actually read as the posterior distribution gets better and can benefit from more data.

Wednesday, April 5, 2017

Sourdough

I spent 12 hours at work today, so writing about math is not really at the top of my agenda tonight. However, I promised my friend Bill that I'd give him instructions for the sourdough starter I shared with him last weekend so, why not, I'll post it here.

This starter was created in 1874 by Mrs. Sheldon Goodwin of El Dorado, Arkansas. It was then shared with Gladys Edwards who subsequently shared it with Nathaniel Pyron. Nathaniel shared it with me in 2016. The original recipe creates a fairly sweet version of sourdough. I’ve modified it to bring it more in line with a traditional sourdough, though the yeast does not produce the really heavy San Francisco sour. I always use weight when working with flour since volume can change significantly due to settling. However, 5oz – 1 cup is a reasonable approximation if you prefer to measure by volume.

As with all yeast, rising times are very dependent on temperature. At 70F, expect the rising process to take 36 hours. It can be accelerated by warming the dough, but the longer rise time will yield a more robust flavor. My personal preference is to go the other way and let the yeast rise in the refrigerator, at least most of the way, to bring out the most flavor.

Preparing the starter

The starter is 1 pound, divided 50-50 between water and flour. Give the starter at least two days in the refrigerator after the last division. The starter will keep indefinitely at refrigerator temperatures provided it is fed a tablespoon of sugar every couple weeks.

Put the starter in a medium mixing bowl (opinions vary on whether to include any surface water; I always include it to keep the ratio right). Add 8 ounces water, 8 ounces bread flour, and 1 ounce sugar and stir until combined. Cover the bowl and return to the refrigerator for 24 hours. The starter should double in size during this period.

Divide the yeast, placing one pound back in the refrigerator for future generations (of dough, not people). The other pound is ready to be used.

Preparing the dough

You are looking for a 5:3 ratio of flour to liquid. However, keep in mind that all liquids are not created equal. I like using buttermilk, which has a large quantity of fat. Therefore, I have to increase the amount to get the ratio right. The following table should help:

floursaltoilwater*buttermilk*makes
7oz.3oz1.5oz1oz1.5ozsmall loaf or 14" pizza
12oz.4oz2oz4oz5ozsmall loaf and some rolls
17oz.5oz2.5oz7oz8.5ozsmall loaf and 14" pizza
22oz.6oz3oz10oz11oztwo loaves

*In case it's not obvious, these are extreme points; only add one or the other. For liquids in between (milk, broth, beer, etc.), start with the water ratio and add as needed while kneading. The amount of oil depends on what I'm making. For pizza crust, I add slightly more. For bread, a bit less and, if I'm using something like beef broth as the liquid, I'll leave it out altogether.

Mix with wet hand or spoon until the dough comes together. Let rest for a few minutes. Knead with mixer or by hand until dough is not sticky (using a Kitchenaid with a dough hook, I find that 1 minute on low, 2 on medium, and 3 on high works well). Remove dough from bowl and fold several times on floured countertop. Cover dough with plastic wrap and return to refrigerator for 12-24 hours.

Baking

Dough should have roughly doubled in size. Fold dough a few more times to further develop the gluten and release most of the gas. Where you go from here depends on what you are making.

For bread, form into final shape and place in pan. Allow to rise (doubling in size again). This may take a while if you were doing the first rise in the refrigerator and the dough is still cold. I'll often do the forming right before bed and bake first thing in the morning. Bake at 375 for 10 minutes then reduce to 350 for another 30-35 minutes until internal temperature is at least 180F.

For pizza dough, form into dough ball and let rest for at least 30 minutes (up to 2 hours if it's coming out of the fridge) Flatten with palm of hand. Let rest for another 30 minutes. Roll or throw dough to desired thickness. Top and bake. For thin crust, 500-550F for 6-7 minutes. Thicker crust and heavy toppings require more time at lower temp.

For rolls, divide into 2-3oz rolls and roll into balls using palms of hands. Place either in muffin tins or together in a pan. Allow to rise at room temperature for another 2-3 hours. Bake at 375 for 20-25 minutes.

Epilogue

You'll be hard pressed to find a baker who hasn't killed their yeast at some point. The defense against this is to share it with your friends so you can get it back. If you want some, let me know. I'll happily split a batch for you.

Tuesday, April 4, 2017

Set Theory HW3

Rather pressed on this one; I'm sure I'll resubmit it. First effort is here.

Sunday, April 2, 2017

HW2 corrections

Set Theory HW2 had enough small errors (and a couple big ones) that I re-submitted it. The updated proofs are here.

Saturday, April 1, 2017

Reims to Nimes

If you came here hoping for a race writeup of the Firehouse 5K, let me not waste any time in disappointing you: 19:53, 6th overall, 1st over 50. Predicts a 3:10 at Boston which absolutely won't happen because I have no intention of running Boston full-on. Enough about that.

Instead, I'm going to write about how I celebrated passing the Q. I have a few friends who can comprehend both what the Q represents and also appreciate a good celebratory meal, but none more so than Bill & Laura Langton. Therefore, we had them over for dinner tonight. Kate and Laura are somewhat indifferent to French wine but Bill & I are quite passionate about it. The theme of the night was Riems to Nimes. In my mind, that corridor represents the absolute pinnacle of culinary excellence.

I tried to come up with a menu that followed geography north to south or vice versa, but found that a bit unwieldy. So, I settled for just keeping the wines from somewhere between Champagne and Provence and then paired the courses as best I could. Here's the final menu:

Appetizers: 2014 Domain Servin Chablis Les Pargues with Crab dip and chips. While paring seafood with a chardonnay is obvious enough, I probably wouldn't have thought of this simply because of the standard maxim: what grows together goes together. As Chablis is a land-locked appellation, crab wasn't even on my radar. However, the more I read about it, the more I kept seeing that Chablis and crab are made for each other. It's true, the match was perfect.

Salad: 2015 Miraval Cotes de Provence blush with Spring Greens, Goat Cheese, Cranberries, Pecans, and Olive Oil. Another revelation. This course was added when I found out (at the last minute) that Yaya's evening plans were cancelled so we were going to have both her and her friend at dinner. I originally had the salad as part of the next course (baking it directly on the flatbread). I asked about how one could best bring a Provence Blush into a salad course and Kevin from the Wine and Cheese Place suggested not using dressing at all, but rather feature goat cheese with just a bit of olive oil. Total home run. This pairing was sublime.

Intermediate: 2009 Moulin a Vent Jules Desjourneys with pesto, bacon, and onion flatbread. As much as Bill's CdP was fab, this was the wine of the night for me. Maybe it's just because I've never had a Beaujolais of this quality. I expected it to be good because it came with a good recommendation from a reliable source (a master sommelier who knows my preferences) but, damn, it was really, really, really good. And, the flatbread I made to go with it was one of my better efforts as well. My usual sourdough crust seems to like more savory sauces like pesto (for a normal red-sauce pizza, I use standard active dry yeast). The bacon did a great job of bringing out the deeper notes of the wine and the onions, well, what's not to like about caramelized onions?

Main: 1998 Bosquet des Papes Châteauneuf du Pape with slow cooked sirloin and peppers over cabbage. Yeah, this was pretty much the bomb. Bill supplied the wine. The stew was pretty straightforward. The one twist that worked great was that, rather than throwing the bell peppers in the croc pot with the meat, I roasted them separately and then added them as a garnish. I think they retained a lot more flavor that way. The other thing was searing the meat in the fat left over from the bacon.

Dessert: Non Vintage Bollinger Champagne with bread pudding. Ok, bread pudding isn't even French, much less from the Rhone corridor, but Kate likes it and Bill makes it really well. So, since he insisted on contributing a course, that's what we went with. And it was super good. I generally buy grower Champagnes as they tend to be better deals. Kate gave me this bottle for Christmas and, while I'm sure I could have found a comparable grower wine for less, this was certainly ample evidence that the big houses do know what they are doing. Meanwhile, Bill's bread pudding is quite good and his sauces (yes, plural, he brought three) are crazy good.

Yeah, it was a good evening for sure.