Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Machine Learning and Ketosis

github.com

Post reply on HN
Print view View original

71–80 of 309 posts

Re: Machine Learning and Ketosis

#71

Post by Waterluvian »

hoodwink wrote: post #67
hrq wrote: post #48

Earlier quoted context omitted.

I've been trying to be low carb for the last couple years and I think it's overall gone well, but I really struggle with what to eat still. I hear your points about protein, cheese, and nuts, but then what do you eat? Can you list out a few complete days of your meals?

Breakfast: eggs + bacon or sausage or leftover steak Lunch: salad and leftover protein (with oil, vinegar, and mustard dressing) Dinner: meat, fish, or poultry (simply prepared) with roasted vegetables and salad Snacks: coffee with butter and coconut oil, raw veggies, leftover meat

I want this so badly but I feel I'm set up to fail due to how much I hate vegetables. I just can't enjoy salads.

I wonder if the "this tastes awful" response will change as I move away from carbs?

Re: Machine Learning and Ketosis

#72

Post by samuelbrin »

I might have missed it, but doesn't look like ketone levels was one of the factors being measured. You can actually buy ketone pee strips at any drugstore for cheap. Would have been an interesting thing to track, since conventional wisdom says that ketosis is binary, you're in it or your not, and the actual ketone level doesn't affect the weight loss. I'm sure this has been put to the test in some experiments already, but maybe applying learning could teach us more and validate/invalidate this hypothesis

Re: Machine Learning and Ketosis

#73

Post by spearo77 »

PerfectElement wrote: post #47
tunnuz wrote: post #3

This is all good if your goal is weight loss. However weight loss doesn't necessarily means higher fitness. Glycogen is fundamental if you do sports, and exercise is a major ingredient in getting fit. If I read correctly, exercise was not a big part of your experiment, how would you suggest modifying the experiment to accommodate one's exercising needs? On a side note, I reached a similar conclusion on the role of "c…

Another anecdote. I tried keto on and off for 2 years while trying to achieve my former triathlete performance. I felt good at the beginning; lost some weight and felt more energy, but it quickly declined and I was never able to perform well. Last year I was introduced to books like Finding Ultra, Whole and Proteinaholic, and went on a high carb diet. Today my diet is based on beans, lentils, potatoes and fruits and…

I was in Ketosis for 12 months, and the last 6 months was training for my first Ironman (age 37).

I would say most of the benefit from Ketosis while training is the recovery. I had plenty of energy for my long rides, but the lower inflammation was making recovery quicker.

My race was cancelled (Tahoe) and I went to Spain two weeks later and did Barcelona instead.. 10h and 3 minutes. Pretty happy with sub-5h for the bike leg!

Re: Machine Learning and Ketosis

#74

Post by CuriouslyC »

hoodwink wrote: post #39
tunnuz wrote: post #3

This is all good if your goal is weight loss. However weight loss doesn't necessarily means higher fitness. Glycogen is fundamental if you do sports, and exercise is a major ingredient in getting fit. If I read correctly, exercise was not a big part of your experiment, how would you suggest modifying the experiment to accommodate one's exercising needs? On a side note, I reached a similar conclusion on the role of "c…

Glycogen is necessary for high intensity exercise, but you are mistaken that carbs are necessary for glycogen. When you are keto-adapted and eat low carb, two things happen that most people overlook. First, you get enough "trace carbs" in your diet that you have glycogen in your muscles for exercise. Second, your body can synthesize glycogen! I've eaten LCHF for 10 years and regularly exercise with CrossFit (arguably…

The rate of glycogen synthesis is directly mediated by glucose availability, and the glycogen synthase enzyme is also activated by an insulin induced signalling cascade. Glycogen replenishment still occurs on a low carb diet, it just takes longer.

If you only do crossfit every 2-3 days, and rotate the exercises you do somewhat (as is typical at most boxes) you might not notice a big difference between a low carb and higher carb diet. If you train daily and you're doing similar movements every day, you will definitely notice a difference. Since that is pretty much the modus operandi of serious athletes, that is where the whole "athletes don't do well on keto" idea comes from; it definitely doesn't apply to your garden variety recreational athlete though.

Re: Machine Learning and Ketosis

#75

Post by spearo77 »

philip1209 wrote: post #2

I've been doing ketosis for about 6 months, after having done twice in the past. From my heaviest to today over the last 2 years, I'm down 50 pounds. The data in this post was cool. I found Dr. Peter Addia's analysis to be one of my favorite resources. He's a medical doctor who was an overweight endurance athlete, then began doing ketosis. I appreciated an honest scientific analysis of the benefits and drawbacks of k…

s/Addia/Attia/

Re: Machine Learning and Ketosis

#76

Post by ariel-faigon »

aab0 wrote: post #32
benkuhn wrote: post #31

Earlier quoted context omitted.

I mean, the uncertainties in the middle are less interesting than the uncertainties at the extremes. I'm sure you can't say anything useful about melon, but it would be interesting to know if (e.g.) "sleep" was consistently that big. He did regularize somewhat (`--l2 1.85201e-08`) but unclear whether that's enough regularization. Basically I'd love to see some actual diagnostics I guess :) PS I believe the "relevance…

Well, he also disables the holdout set (`--holdout_off ariel.train`), so between that and not reporting the uncertainties and his absolutely tiny dataset with high measurement error and the low prior probability that any of these effects could be that big... Nah. It's just overfitting noise. I tried to run the makefile, but apparently the version of Vowpal Rabbit that ships on my Ubuntu (7.3) is so outdated that it d…

Thanks so much for all the excellent comments. There was definitely an over-fit with 4-passes.

No more. I've updated the Makefile to run only one pass, changed the options so it runs with older-version vw, Fixed misspellings of 'gioza', removed 'mayo' which found itself on the wrong side because it appeared only twice and always alongside the bun and regenerated the chart.

All the main conclusions remain intact.

In the end, I urge everyone to use their own data, that was the main purpose of sharing this code. My data-set is small, awfully noisy and insufficient. There are no p-values and no rigorous statistics, so please don't read too much into the minute details. It is the discovery journey into the top factors that is the important part, in my view. The ML was just one aid in this discovery process. The proof for me was my actual, and sustainable, weight loss that came after (very slowly) realizing the top factors that eventually worked for me. Thanks again.

Re: Machine Learning and Ketosis

#77

Post by will_brown »

mrfusion wrote: post #33
philip1209 wrote: post #2

I've been doing ketosis for about 6 months, after having done twice in the past. From my heaviest to today over the last 2 years, I'm down 50 pounds. The data in this post was cool. I found Dr. Peter Addia's analysis to be one of my favorite resources. He's a medical doctor who was an overweight endurance athlete, then began doing ketosis. I appreciated an honest scientific analysis of the benefits and drawbacks of k…

What do you mean about insulin?

I am very interested as well and hope there is a response. Since he was working with an MD, my guess is as follows: insulin is the devil.

Whereas if you look at the old FDA food pyramid and the new food plate, carbs/sugars/fruits take up a significant percentage of the recommended daily diet. Take fruit for example, prevailing norm would be fruit is healthy (especially in its natural state where it is accompanied by fiber), fruit aside most people would believe carbs (especially in whole grain form) are perfectly fine in moderation. My big guess is that an M.D. recognizes any amount of carb (fruit or whole grain being no different that high fructose corn syrup) triggers the bodies production of insulin. More and more, I think M.D.'s will have negative perspective on any/all insulin production if it can be avoided. I have more thoughts on the impacts of insulin on the body, but I hope OP will chime in with a response to see if my instincts are on point.

Re: Machine Learning and Ketosis

#78

Post by CuriouslyC »

toomuchtodo wrote: post #7
tunnuz wrote: post #3

This is all good if your goal is weight loss. However weight loss doesn't necessarily means higher fitness. Glycogen is fundamental if you do sports, and exercise is a major ingredient in getting fit. If I read correctly, exercise was not a big part of your experiment, how would you suggest modifying the experiment to accommodate one's exercising needs? On a side note, I reached a similar conclusion on the role of "c…

http://www.ncbi.nlm.nih.gov/pubmed/24979615 TL;DR A ketogenic diet is net-positive for aerobic exercise activities.

You will see greater endurance adaptations in response to exercise in the short term on a ketogenic diet, though this effect decreases the longer you maintain the diet. The downside to a ketogenic diet is your total training volume for work above the lactate threshold is more limited.

As a result, high level athletes tend to spend the majority of their time consuming a carb-rich diet, then switch to a ketogenic diet for 2-3 weeks prior to competition. During this low carb period, they just focus on a very high volume of low/moderate intensity work. Finally, 2-3 days before competition, they will carb-load to glycogen super-compensate, getting the best of both worlds.

Re: Machine Learning and Ketosis

#79

Post by spearo77 »

raffy wrote: post #24

I guess if we're sharing fitness plots, here's 1300 measurements and 7 DXA scans: http://i.imgur.com/J4Ls2bQ.png

Woah!

I thought I might be over-doing it with 2 DXA scans :)

I should dig mine up, I got down to 10% body fat a few weeks before completing my first Ironman. I'm normally around 17-19% from memory.

Re: Machine Learning and Ketosis

#80

Post by aantix »

samuelbrin wrote: post #72

I might have missed it, but doesn't look like ketone levels was one of the factors being measured. You can actually buy ketone pee strips at any drugstore for cheap. Would have been an interesting thing to track, since conventional wisdom says that ketosis is binary, you're in it or your not, and the actual ketone level doesn't affect the weight loss. I'm sure this has been put to the test in some experiments already…

On a related note; NuSI (Gary Taubes's Nutrition Science Initiative for better nutritional testing) has not published their first results for the carbohydrate/insulin link to obesity, but Dr. Kevin Hall discusses the upcoming results in this video and some people may be surprised (including Gary).

_No Metabolic Advantage for Ketosis Found_

https://www.youtube.com/watch?v=MiUyjMjuLl0

Post reply on HN