2019 Resolutions Revisited

There is only a little more than a month before the end of 2019 and it seems like a good time to look back and see how much progress I’ve made on my 2019 resolutions. I have achieved some of my goals which I am quite happy about and there have been some re-routes and detours along the way as well.

1 Things I wanted to learn and/or get better at

• Causal Inference (PO and Graph Frameworks and the difference)

This endeavour has been a success so far. I read Pearl, Rubin, Imbens, Holland, Hernan, Shalizi, Peters. I formed a good understanding of the potential outcomes framework as well as the DAG framework. Of all the approaches and amongst a plethora of jargon (ATEs, LATE, CATE, IVs, potential outcome, DAGs, mediators, collidars etc), what helped me most was to see things as falling under 3 main branches as Shalizi does in his book.

a. Identification: This is best achived by Directed Acyclic Graphs. Given how your variables causally affect one another, which causal effects can we estimate using data and which causal effects we simply cannot. Without drawing a DAG, I personally find it very difficult to get a good sense of things when they are embedded in independence assumptions. b. Estimation: This is pure statistics, associational methods that statisticians have developed over centuries. Having said that, because the Rubin framework mashes identification and estimation together, it makes the practitionars life a lot easier. One can always pick a paper written in PO framework and immediately apply them whereas those using DAGs have to put the identification and estimation parts together a bit more manually. c. Causal Discovery: This is basically going in the other direction of identification. This time given the data, we try to uncover the causal mechanism that might have generated the data. Using the dependencies observed in the data we can restrict the graphs that could have led to the data we ended up observing. Fascinating stuff intellectually but this was very difficult to apply in real data at my job. One fun takeaway for me was it was maybe the first time normality assumption made things more difficult not easier. In the case of two variables only, if the cause and noise are independent and noise is non-Gaussian, then the causal effect is actually identifiable whereas without the non-Gaussian assumption we can never know which of the two variables is the cause and which is the effect. More on this here.

What I could apply in my job was mediation analysis and instrumental variables. I found very interesting use cases that come up in our data at booking.com and it was so much fun applying what I learned in my day job. I have a blog post on our company website on the work I did on mediation analysis.

I also had the opportunity to attend European Conference on Causal Inference 2019 and met some fantastic people mostly working in epidemeology. One thing in the opening talk of Miguel Hernan did not sound too convincing to me and that was “no causal effects without an experiment”. Hernan wants to bring the focus from causality to causal effects. So even when the data comes from an observational study, he insists the methods should help create ‘target trials’ that allows the estimation of causal effects.

This approach mitigates the ‘problems’ of methaphysics if one wants to keep the focus on causality insread of causal effects. So instead of changing the focus to make things more practical, I ended up taking a detour and reading on the metaphysics of causation. In another post, I should to put my ideas and notes together on what I learned from reading Al-Gazali, Avirreos, the early modern empricists and Plato, Aristotle and Neo-Platonists.

• Population Genetics and how the same theory applies to Cultural Evolution

I bought books to learn more about the subject but causality took most of my free time. So this will have to wait for 2020.

• Information Theory

This will be my main area of focus in 2020. I intend to start from the basics, learn about the connection of information theory to hypothesis testing and on a slightly unrealted note, read more on the information theoretical approach to aging.

• Bayesian Model Fitting and Checking

I never got the chance to go beyond McElreath Statistical Rethinking. I did fit a Stan model to booking data to test for heterogeneous treatment effects (HTEs) though. I should make a blog post under data science on how one can utilzie hierarchical models for HTEs.

2 Books to Read

Progress is marked by check marks on the original 2019 resolutions page.

3 Diet

Stick to a Paleo diet as much as possible. This has been another of the successes of this year.

I pretty much eliminated sugar from my life. First, I kept one piece of fruit a day like a banana or an apple but later in the year I replaced that with mixed nuts, then walnuts only and these days i do not snack at the office anymore. The only fruit I eat is berries after dinner.

I eat an animal protein heavy diet. One of the difference between the paleo and keto folks is the type of the cut and how the animal is fed. Paleo diet prefers lean cuts and grass fed animals (game animals especially) whereas in the Keto diet more fatty cuts are preferred. I don’t necessarily look for lean cuts but try to make sure my meat is coming from grass fed animal raised in local farms, though not always successfully.

I do not drink at home anymore. And even when I am out socializing, I do not binge drink as much as I used to. I still drink a beer or two or a glass of red wine occasionally which I may try to reduce more in 2020.

Another thing i comppletely eliminated from my diet is vegetable oils. I stick to coconut oil, olive oil, butter, and ghee.

In addition to what I eat, I also changed when I eat as well. I have ben doing intermittent fasting with 16 hours fasting and an 8 hour eating window a day. I started with fasting a couple of days a week and now these days I pretty much fast everyday unless there is something extra ordinary that stops me from fasting. On Saturdays I even do one meal a day fasting.

Al these changes in my diet combined had a positive impact on my health. Even though it was not the actual goal, I lost about 10-12 kilos and went down to 70 kilos. Most of that was fat as I managed to hold on to my muscle mass for the most part via calisthenics training though some of it was inevitably still muscle mass.

In 2020, I plan to add cold showers to my routine and focus more on my sleep which has been quite erratic unfortunately.

4 Exercise

High intensity workout 1–2 times a week I managed to stick to this goal as well. I bought gymnastics rings and a pull up bar to my home and cancelled my gym membership. Bringing my workouts to home and keeping them to a maximum of 20-30 minutes really removed all the barriers that prevented me from going to the gym.

I got hooked on calisthenics. I managed to bring my pullups from 5-6 rep range to 16 reps, started doing ring dips (10-12 reps in a set) and managed to learn how to do a one leg tucked front lever.

All my progress came to a halt when I pushed my limits a bit too much with weighted calisthenics and injured my wrist in October. I stopped all exercises save some leg and abs workouts at home and giving my wrist a rest until January next year. If I can recover fully, I intend to continue with calisthenics in the coming year with the goal of holding a front lever for 5-10 seconds.