Skip to content

Bayesian inference and probabilistic programming (Episode 6)

Interview with Alex Andorra

Interviewing Alex Andorra about bayesian inference, probabilistic programming, and more was a pleasure.

Alex is a data scientist and modeler at the PyMC Labs consultancy. He’s also an open-source enthusiast and core contributor to the python packages PyMC and ArviZ. Alex is also a contributor and instructor in the “Intuitive Bayes Introductory Course”. This self-paced course is designed for data scientists and developers, where you’ll learn Bayesian modeling with code, not math.
Alex also runs the amazing “Learning Bayesian Statistics” podcast!
If you love Python and Bayesian inference, catch this episode.

Stay tuned while we discuss the following points:

  • How Alex began his path on statistical modeling and how he finds this a challenging, versatile, and creative
  • How Alex and his agency do with clients using multilevel regression and post-stratification and tracking opinion through time
  • How the community around PMYC is growing and contributing to development in an open source format
  • Alex hosts a podcast that covers a wide variety of topics, including political elections, healthcare, neuroscience, and the market
  • Alex’s suggestions of what books to read and how to effectively contribute to open source projects, and stay active in the community
  • What are the benefits of joining online communities for learning
  • and more..

Listen to this episode now and share this with your friends and colleagues!

Alex Andorra
Co-Founder of PYMC Labs


[00:00:00] Paolo: Welcome to the Effective Data Scientist Podcast. The podcast is designed to help improve your skills, stay focus, manage successful projects and have fun, hard work. Be an effective data scientist. Now!

[00:00:43] Alexander: Welcome to this episode today and Paolo and myself have a great guest. Hi Alex. How are you doing?

[00:00:51] Alex Andorra: Hello! Alex and Pao lo. Yes, doing very well. A bit hot these days in Budapest, but hanging in there. And most all, thank you for having me on the podcast. Yes, I’m psyched to talk a bottle all the topics you have in mind for Tiffany.

[00:01:06] Alexander: Awesome. Very good. And maybe, for all those that don’t know you, can you introduced a little bit yourself, your background, your experience. What brought you to where you are now?

[00:01:18] Alex Andorra: Yeah, sure. Yeah, my path is actually quite seniors , so I’ll go fast and then we can complete if you want to, but actually didn’t start at all by doing what I’m doing right now. I studied management in political science in France and in Germany but was not satisfied with my first job. And quite randomly while writing a book about the two politics of the US I encountered the statistical models that nightsilver does for the US elections that was in 2016. And I was hooked from that day. And nerdy me was awoken. And so I immediately thought, that’s awesome. I need to do that in France. And so to work and build a Beijing model a statistical model. I didn’t know it was Beijing at the time for elections I dived. I went again into statistics, learned how to program. I was lucky because I just picked up Python quite randomly. And then python’s book the year after just skyrocketed. Which helped me a lot in the job market.

And yeah. Long story short today I am no longer doing political. I, although I am of course, from time to time working on electoral forecasting and also just tracking public opinion sentiments and so on, actually doing that with PCI Labs clients that we have. But now my interest is more generally in the methods themselves and how do we apply in patient methods like most efficiently and basically writing statistical models most days. And also a big part of my job is communication and teaching. So teaching workshop with fine see labs, teaching an online course with two of my friends, Tamaki and Ryan Kumar. And also the podcast Learning Patient Statistics. So yeah, like all those different activities, which are all related because they’re all talking about patient stats. But that’s basically where I am, and as you can see, like very random and sinuous. But that’s how life is sometimes and that’s cool. I picked up, I learned stuff along, around, along the way.

[00:03:40] Alexander: Yeah. I love it. It’s really weird where, all of us get this buck in terms of statistics and data science and programming. I started doing, pure mathematics, ultra analytics, all these kind of different things. And only by chance got into statistics course because another student had that was friend, said, oh, that looks interesting. That goes there. I said, oh, yeah, I joined you.

[00:04:05] Paolo: I started in doing economics and found of course quantitative methods useful, and then completely passionate about the statistics. So it’s always interesting to hear about the different path and why you are so passionate about Beijing modeling.

[00:04:24] Alex Andorra: So when I started it’s mainly because that was the most useful for electoral forecasting because that’s the perfect framework for business statistics. You don’t have a lot of data. The data you have are. Like not very reliable. They are political polls and we all know these are like the best we can have, but it’s clearly not the ideal data that we would like. And so you need to add domain knowledge and so a lot of information is not in the data. You need to add it in the model structure. And also what’s very important is estimating uncertainty. And yeah, looking at different scenarios and estimating the probabilities for all of these scenarios and trying to like quantify your uncertainty as precisely as possible. So that was perfect for basic statistics and that’s how I ended up like learning deeply about that kind of statistics. And then diving into fine see, because I was learning Python and fine see is one of the main package to do to do patient stats in Python. And yeah, that’s basically how it happens. So again, quite a lot of serendipity. But afterwards, why I found that really interesting, and I still find it extremely interesting is I would say two things mainly. One. It’s complicated . You keep learning all the time. There is never a point where you can say, tell people, oh yeah, I’m good. Like I’m a Beijing master and I don’t have to learn anything anymore. So it’s really cool because any project I start I’m like, okay, I have a vague idea of how I could do that. But I have to work on it and I have to learn, and often I have to learn new methods and new way of modeling. And the second thing I would say is it’s fun because it’s like playing with Legos but foreign nerds so you have, as if Legos were not nerdy enough, but yeah, it’s like you have all these building blocks and you play with them to get to the most customized version of the model that you want, and that of course depends always on the use case. So it’s not boring, it’s never the same thing, and you always learn something. And that’s fun because you get to learn about all these different Lego blocks and have to be creative actually.

Contrary to what a lot of people think statistical modeling at that level when it’s very customized is actually quite a creative endeavor because most of the time you don’t have the answer before you start, so you have to create it. So you have to imagine how you can like, assemble all those building blocks together. And yeah, that’s I have to say that’s very fun. That might not be the idea of fun to a lot of people, but it is one to me. So yeah, intellectually challenging. Very versatile and yeah, it’s like keeps, if you want, it keeps humans in the loop because that’s what patient statistics do mainly. It’s like it uses the horsepower of computers. But humans always have to be in the loop to have that scientific knowledge in the model and the structure. And so I find that pretty cool because he tries to use the best of what humans can do, which is like scientific reasoning and then using what computers are very good at, which is computing very fast. Yeah. So like I’d say that’s the answer.

[00:07:47] Alexander: I’ve never thought about it. That, election data and these polls that happen before elections are perfect source for it because, you get these, lots of these polls up to the date or the period of the vote, and of course they come from all kind of different areas. They have different sizes different populations in it. And of course also the time aspect plays particular role in it. I would get, you closer, you get to the specific time point. Yeah. It’s probably much more kind of predictive for the expo results.

[00:08:21] Alex Andorra: Yeah, again, I was quite lucky to be interested in these topics because it keeps being one of my nerdy sandboxes. Basically each time there is an election that I’m interested in or that some clients are interested in, I get to work on that and keep learning. And yeah, the time aspect also was something hard for me at the beginning to include the first model I ever did in 2017, was basic and didn’t have that time component. And finally in 2022 for the French election I worked again on that model for the presidential elections. And finally five years after having started. I was able to build the model that I actually wanted to build in 2017. And I was quite happy, I have to say. I was like, oh, okay. Because at the time I didn’t understand it statistically and I wasn’t able to Pro, write it programmatically. And so yeah, like that’s again, an illustration of these perpetual learning motion that you get here. Where yeah, if I five years later I was like finally in a way, in a place where I could. and all those components. So yeah, definitely the time aspect is super interesting. And if you’re lucky, also you get a geographical aspect depending on your polling data. And also if you have sensor data, then you can use multi-level regression and post stratification to devise even more your estimates which is something always super interesting. And that we’re actually working on for some clients at labs. And yeah, if you can also get that geographical aspect in the data. That’s super cool. And of course when you talk about time dependency and and geographical patterns and Sian processes often appear which is always cool because, they’re a very interesting kind of model.

[00:10:00] Alexander: Cool. That’s so cool. And it becomes pretty obvious that this is not follows, this recipe work. It’s very creative. You basically need to design your model and every new data set requires different adjustments.

[00:10:13] Alex Andorra: You have foundations, you have common foundations. So that’s also the cool thing is that then the foundations can be made in common. And they can be stronger because it, like you learn from different cases and then everybody enjoys having stronger foundations. But on top of that, you have To customize your house, right? It’s a bit like building houses, right? Like the way you build a foundation is I think, always the same. The principles are the same, but then you customize your house however you want and however you need it’s the idea here. Using the muscle foundation as possible and then customizing it however you want and need.

[00:10:50] Alexander: Yeah. Yeah. Now you’ve already mentioned your podcast and it’s always great to have a fellow podcast on the show. Why, what got you started to with a podcast and is there any key takeaways that you have from the podcast? Up to now?

[00:11:05] Alex Andorra: Yeah. Why did I start? Because I started that even before actually working with fine see labs like fine see Labs. We started that two years ago and the podcast study started three years ago. No. There is a one year gap between the two. And yeah, the podcast is actually at the very beginning of my learning path. And if you listen to the first episodes from three years ago, I actually say it in the very first episode that I’m just starting to learn Beijing Statistics and I’m using that podcast as both a commitment device and a way to happen while learning. And basically I, listen, I listened to a lot of podcasts and I loved to learn through podcasts. And at the time I looked around and I was like, oh, that’s so shame. There is no podcast about Beijing stats. And then, because maybe I’m crazy, I thought maybe I should start one. I don’t know. That could be interesting. And my reasoning was quite selfish. It was like I don’t really care if nobody listens to it because like I get to record the episodes and so that means I get to spend one hour each of the guests. And I was inviting people that I found were doing fascinating work. And I was like if I can get one hour one-on-one with them talking about what they are doing and how they would advise me and people to learn basin stats and get to where they want. Do it to where they are. I’m like, I’m down for that. And I don’t really care if people listen or not.

[00:12:35] Paolo: And I must say that the quality of the guest is special. I think in your podcast. I think that you had incredible people like, for example Frank Arre. I dunno. Many other outstanding people in the field. So I think that it’s really great. And, I’m addicted to your podcast, so right now I’m quite likely because Alexander started at the effective statistician podcast was my number one podcast in data sciences statistics. And number two is learning base. It’s quite incredible.

[00:13:12] Alex Andorra: For that feedback that’s that’s always awesome to hear.

[00:13:14] Paolo: Yeah. A lot of different topics. You mentioned political elections, but it’s of course’s about healthcare neuroscience, so market. So many topics and excellent guests. What do you think what are your planning in terms of podcast episodes? Would that be different topics you will explore?

[00:13:33] Alex Andorra: Yeah, as you say, I like basically the idea of the podcast is like anything Beijing can be on the podcast if it’s good basically. So that’s also the cool thing is that because Beijing statistics are a method. So they can be applied to a lot of topics. And so I can talk about any topic if it’s interesting and done in a amazing way. And since I’m a pretty curious person, then yeah, I end up talking about different, very different topics from one episode to the next. Yeah. Like the good thing is that I don’t honestly, I don’t really have a big plan to like six months where I’m like, oh, I wanna talk about that, and that. It’s often not last minute, but like in a month or two, I know about what I’m, about, what I’m gonna do in record. So this month has been more focused about. Marketing the new plan C 4.2 release, and some nerdy pleasures that I have. So mainly around physics and astrophysics these days. So yeah, like the next step episode is gonna be, With Ricardo Vierra. So it’s gonna be like diving into the New Pine Sea release. And I’m also gonna have a, an astrophysicist on the show very soon, so it’s gonna be gonna be super fun. So yeah things like that. . And of course, I’m always happy to talk about the topics. I always love, so electoral forecasting and other, let’s say more classic topics like software engineering and basic stats, stuff like that. But yeah, that’s basically, that’s what I’m doing. And for now the roadmap is quite blurry because I am actually planning right now the next batch of recordings. Actually what I’m thinking about right now, so I don’t have a clear answer for you yet, Paulo, so I’m afraid you’ll have to wait for the to update.

[00:15:13] Alexander: That’s fun. Yeah. I also get a lot of inspiration just from things that I see that I’m curious. Oh, that would be interesting. I really would love love to learn more about it and then reach out to the people. So you also started an online course, intuitive based. Tell us a little bit about it. For who is it for and what will you learn from it?

[00:15:36] Alex Andorra: Yeah. Good question. So yeah that’s very recent. We just released that one month ago with my friends Rav Kumar and Thomas Vicky. And so we had the first cohort already. The second cohort is gonna open very soon, actually at the end of July. And yeah, there are like an Idea was that the three of us, like all the whole pine sea developers team we had to learn patient statistics basically by ourselves. And also we had to go on the internet and look for every possible resources and then filter that out and ask people, do we think about that and do back and forth and trial and error. And so that means in the end you get to that point where you know what’s useful and not. But it takes a lot of time and trial, and error and back and forth to understand what’s useful to learn now, what’s not what you can put aside when you are a very beginner and so on. And so the idea was like let’s try to develop a basically one stop shop where people are like, okay. I heard about that basin stats stuff. What is that about? And how can I apply that in my work? Like very fast and without too much hassle, like reading through extremely complicated statistical mathematical books which are awesome, but when you’re beginners it can like, just make you, discuss it . So yeah, that was the idea. Like how do we pack as much information about beijing statistics. If you’re a complete beginner about patient statistics, you already are familiar with some python or the core is in Python and Pine Field and Pine C of course, but or in Python or not that far. So it’s if people are it’s not hard to pick a python.

So a bit of Python, you probably know about data science or your software engineer and you are curious about that, like you. Like shifting into more patient data science in your work. That’s the course we wanted to create. What do I do? And that’s the one stop course. You just need that and then you’re good. You can start applying that in your work. That was the idea. And the second idea was, we wanna do that with a code first and example, first approach. Instead of mathematics first approach. Which is also the way Thomas Raven and I learned because none of us have a mathematical degree. We of course studied mathematics, but it’s, it was not like our major or something like that. And the way we learned based statistics and me, the way I discovered it with the power was, simulating stuff on my computer and understanding about probability distributions and random draws that way.

So that’s what we wanted to do with the course. To be like this concept is that basically on your computer you can recreate it, you can see what that means? . And then, yeah. We’ll tell you a bit more about the math and what that means, because sometimes you will need it but probably not if you’re a beginner, like it’ll be later down the road, so you don’t need all that math at the beginning. And again, not that math is not cool. It’s one of the coolest things in your universe, but you get my point.

[00:18:33] Alexander: Yeah. I kinda can see some mathematicians screaming. Although you are serious first, I can completely see myself enjoying such of approach where you kinda first get some kinda quick success with your work. And you can see oh, that’s interesting that it works. I can tweak a little bit here, tweak a little bit there, and then see what happens. And then understand the foundations behind it.

[00:18:59] Alex Andorra: Exactly. Yeah. And actually I had that idea of I was reading some Getta Christi novel and the power of Aha Christi like you read the novel at the end, you understand who the murderer is, and then if you reread the novel, you actually notice that you could have kissed on page three. Everything was there. It’s like incredible. I don’t know how she did that. It’s like just the masterpiece, like most of the time you reread it and then you see everything, like all the pieces of the puzzle are here. But you just didn’t see them at the beginning. And that’s what we wanted to do with the course. Like you go through the course and like histograms or you do that model, etc. You code the pine sea model and then at the very end we’ll tell actually what you did is like linear aggression or, and that’s the form, mathematical form of an linear aggression. But that’s what you just did. It’s not, that’s scary. It’s just that’s exactly what you just did. So like these kind of christi approach if you want.

[00:19:55] Paolo: At the end, the mathematics is the model.

[00:19:58] Alex Andorra: Oops. Yeah. I didn’t think that’s perfect. Yeah. I mean it’s the murderer of bad thinking, so of non methodical thinking. So yeah, in a way, yeah. Then that’s a good murderer.

[00:20:10] Paolo: I think that the applications make mathematics and physics. It’s really fun to learn and apply. And you mentioned lab, which is a company . And could you please tell us a bit more about 30 company? What kind of clients do you have industries can benefit from your work, for example?

[00:20:36] Alex Andorra: Yeah basically Labs is asian consultancy. It’s a collection of pine sea car developers who work there mostly full-time .And basically, clients come to us because they are using pine C models in production, or they are well, or they want to get more into beijing modeling. And they have an idea of what they wanna do, but they don’t know how to do it because they don’t have the skills in house. And so yeah, they come to us because they have some difficulty. And then we help them, like we build custom solutions and models for them that they can then use and deploy in production, in their systems. And usually make money out of it because that’s what they want. Or save money or, yeah, basically that’s the idea.

And most of lot of clients are in marketing for instance. So a lot of variations around media mix models where the idea is we’ve got all those bunch of ad channels and we pour money into them. But we wanna know rightfully if it’s efficient, like how efficient are our channels and in converting the persons who see the ad into new customers basically. And okay, that’s the big part of our job is helping clients do that. So if you’re interested in that, we have a whole Globe Post series on the pine sea Labs website about what we did in that regard for HelloFresh which is really cool model in the end with hierarchical gas processes. Like way to take into account the fact that there is a saturation in the ad market and that some channels have delay effects or stuff like that. Like for instance, an advertisement is more efficient, usually on Facebook than on TV because it takes less time for people to see it on Facebook, for instance, stuff like that.

So the model takes all of that into account. So there is a whole blog post series on the Pine Labs website by Vincent, one of the great guys working at labs. And we’re gonna make a video version of these of these posts for the Pine Lab YouTube channel. So if you prefer that medium just hang around a bit and then it’ll be released. So that’s a big part of the client’s marketing basically, and understanding the efficiency of the marketing. And then we have a lot of biostatistics. So we’ve worked a lot and we still work a lot for with clients who are developing medicine, developing vaccines testing new drugs, et cetera, where understanding the efficiency of the treatment is extremely important. And also here, causal inference is often in the balance. And also, the cost of being wrong is extremely high. Because you can kill people, so you don’t want that. And so you have you can also take that into account like the, with the cost functions in the optimization then afterwards of the beijing of the Beijing models. Basically this is another topic like Basian decision making optimization, which is taking the full poster distribution of the model. Then taking all of those scenarios into account for optimizing the decisions you make afterwards, whether for marketing or for drug development, things like that. And Then we have other, these are like the two blockbusters, let’s say marketing and biostatistics. And then we have other applications which are a bit more original, let’s say for agriculture, for instance. Which is ironic when you know that RA Fisher developed most of frequentist statistics control data so you know, the revenge of the patients.

But yeah actually we are doing. Actually, I think it’s happening right now. A live webinar with one of our clients, indigo. And there will be, so Billos Thomas, Vicky and a person at Indigo during the webinar. You can watch that afterwards. It’ll be recorded where they will dive into the model that we did for them. And basically it was, how do you understand the efficiency of new crop? Like how do you discriminate between different crops and they are like testing new crops all the time and they wanna understand the efficiency of those crops, which is extremely complicated because you have a lot of confounding factors like the environment where the plants are. Where the crops are which plots they are in, et cetera. And bill and a lot of other people at land have worked on that model, which ended up being again, like involving a lot of gosh processes, but here, not for time correlation, but for geographic variation. So awesome application again, of course and processes is really fascinating.

I just love that model each time I go in there and look at it. So yeah, if you have a chance to go and watch that webinar to achieve, you will get a better idea of that. And then final final point to that long answer. As I mentioned already, we also do some. Let’s say electoral political science for clients. Multilevel regression with post stratification mainly. So ba working on polling data and sensor data and try to make sense of all of these through Beijing model to get uncertainty estimation, to get better estimates than the biased polling results that you have. So to track opinion also through time with the time dependency we already talked about. And then after that, you can also do based on decision making optimization, where you try to understand which coalitions, for instance, in a parliament would be possible and which would be the scenarios with which probability and things like that. So it’s also something we work on from time to time, and of course I always love working on that as you can please.

[00:26:12] Alexander: Cool. That is so cool. As I’m, working for over 20 years now in the healthcare space I’m really interested in the applications there. So is that basically pretty much cross the value chain from Preclinical development up to kinda things in marketing and things like this? Or is it really, do you really focus on the both stuff there?

[00:26:34] Alex Andorra: So you mean for the bio statistics client?

[00:26:37] Alexander: Yeah.

[00:26:37] Alex Andorra: Yeah, it’s mainly, not marketing is really drug development.. And understanding the efficacy, the efficiency of a new treatment. And that, that can be along like a variety of different treatments. But many, it’s, yeah. In the very, in the research and development process. And there will actually be a research paper core authored by Luciano Pass, the one of guys working at labs with us. A really brilliant guy. I had him on the podcast very recently. I think the episode 63. And yeah he’s coauthoring a paper with one of our clients, so Roche to nine, which dive a bit more into the methods. They used, I, I think the paper of course is not out yet but will be at some point.

[00:27:17] Alexander: Will put references to all these kind of things into our show notes. So that you can easily find that.

[00:27:25] Alex Andorra: Sure.

[00:27:25] Paolo: Okay. And I have one question about the pros and cons of using pine and C in beijing application and what are the advantages of using pine and C instead of other software like stamp for example, or just it’s only a matter of using Python or there are other advantages?

[00:27:50] Alex Andorra: Yeah. So I think understand by saying that’s a good problem to have. Like a few years ago you didn’t have enough beijing package to work on beijing on your beijing models, and now you have a whole array of package of high quality packages that you can choose from, and that’s just amazing. And that’s, I just love that first and second yeah like it’s indeed important to think about what would be the best package for you in a way. But I think honestly there is no really wrong choice. Like stunt is amazing and we actually often update also the algorithms in Pine C when the stunt team does so it’s like basically, yeah, stunt is never a wrong choice. And I would say Pine C is never a wrong choice either. So it’s mainly based, I think, on your personal preferences. Yeah, clearly the fact that Pine C is Native Python helps you a lot because if Python already is just Installing a new package and then learning the syntax and then you’re good.

That’s a huge advantage. And also there is the, this whole ecosystem around prime C now. So pine C is the flagship, but you also have Bambi, which is the b RMS in of Python if you want. So a wrapper around P M C that helps you work on generalized linear models with the how do you call the Patsy formula syntax. It’s actually a course, an online course I’m working on with Tommy Capto one of the core developers of Bambi. Yeah, that’s one that’s of course we’re working on for the intuitive base series actually, so you have that and you also have. Then the backends, which are SR App and apple, which are extremely powerful. And I think that’s one of the really great advantages of pine C here because then you can use symbolic computation and symbolic transformation of your models and I’m gonna go very fast here, but if you want more details about that, actually the next episode of Learn Based Stats so 65 is gonna be with Ricardo Vierra, and we’re gonna dive into all of that and what this enables you to do.

But basically Yeah, you are gonna be able to do symbol computations. And so first the shape issues that you had in pine C3 are basically gone from pine C 4.0 because now you have that Symbolic computational graph behind the scenes. And also more and more is apple is developed. So APPLE is developed by also Ricardo and other people Brenton Willard in particular who had this whole vision about symbolic computation and who was in, in learning based and statistics. So you can refer to that piece of those. So if you wanna, the big, like very, let’s say, mathematical vision around that. And so that will allow you to do automatic parametrization of models, for instance. So if you’re using a model with the centered parameterization you could also ask P M C to under the who’d try to use the self-centered prioritization and then just choose which one works best. And the user would not have to change anything in his model code, which would be absolutely amazing.

So that’s that’s almost ready. We just need someone to work on the whole request to push it towards the finish line. So if someone wanna work on that please get in touch because that’s gonna be very high. But basically the infrastructure is here. We just need to add the feature. But that’s a very powerful thing. And last thing I’ll mention is you talked about Jax The good thing is that pine C 4.0 was built to work with Sarah which is, let’s say, agnostic on that end, and that way you can use different samplers. You can use the NumPy sampler you can use some Jacks features. When you’re sampling an in your models and you can also use I think that it’s a relo working on developing the nuts algorithm completely symbolically in ESARA. So you can also use that simpler. It’s not yet in the package, but I think it’s going well. It’s gonna come at some point.

So yeah, you have basically, those building blocks because it’s abstracted away. And so that way you can plug some of other samplers and use some jacks features, which I think allows you to have kind of the best of both worlds here because you can keep your prime syntax and models, but under hood use something that could be more efficient in a special use case.

That’s a very ideal answer again. But yeah, if you want more details about that NBS 65, Ricardo is gonna talk about that much better than me in much more details.

[00:32:32] Paolo: And, is the community growing around P M C? Do you have more contributors or it’s a small group. The way now, all the work.

[00:32:41] Alex Andorra: Honestly, for that I don’t really have data. That’s something I should ask Ricardo. But for sure the help on Pine C 4.0 was great. From the community. Of course the the biggest pool requests come from the car developers and actually it’s better that way because we know the quality of the code is gonna be good and those big pool requests need to be very high quality anyways, because then it’s good that goes into production in a lot of companies and things like that. So it needs to be to be very thorough. But we got a lot of help Yeah. Outside help from people starting and having their first contributions for the new version. And like audio lab and other people in the core team have starting these PINC Sprint open to anyone on Meetup.

And I was actually in one of them. The other weekend where it’s a bunch of people from all over the world getting into the discord from I think it’s PTA umbrella or something like that. The name of the organization, I’m so bad with names, so I’m so sorry about that. But yeah, we all go into the discord and basically volunteers come here. And they are paired together to work on the issue of their choice. And sometimes the developers are there in another room so that when people have troubles or issues, we can help them out. And that appeared with the development of 4.0 and that was, that’s really cool because that tree helps people enter the community and stay there because they are well welcomed and they see that it’s like the barrier to entry is lower. So that definitely.

[00:34:16] Paolo: Oh, that’s amazing. That’s amazing.

[00:34:18] Alexander: Cool. Thanks so much. Awesome discussion. I think we covered a lot of ground starting from your podcast that has no, as you said, quite a lot of episodes for people to get into it over your course Intuitive space, the company pine sea and finishing up with the pro and cons of both of different programming approaches.

[00:34:42] Alex Andorra: And yeah, well done guys. I’m impressed by your hosting skills. Like I need to, you need to give me some of your tips. Like we covered a lot of topics.

[00:34:50] Alexander: After one or 50 episodes with the effective, I’m have a little bit of training in that. Yeah. Thanks so much. . Is there one last thing that you would like the listener to take away with if he wants to learn about Asian statistics?

[00:35:06] Alex Andorra: So you mean for people who wanna start learning about Beijing statistics?

[00:35:10] Alexander: Yeah.

[00:35:11] Alex Andorra: Yeah. As I always say just do it.

[00:35:14] Alexander: Love it. Yeah. Yeah.

[00:35:16] Alex Andorra: At one point. Yeah sure. Reading books is important, but at one point you also need to just go there and go ahead. And for me that was, I think one of the main best decisions I made to learn beijing status when I was reading Statistical Rethinking by Richard Mcow. Oh, I contributed to port the code from R to pine C of the chapters. And that helped me tremendously because at the beginning it was extremely bad because I was learning pine C and also I didn’t know how to do open source contributions. At the time I was very lucky because Rav Kumar and Osvaldo Martin like who were already more seasoned, pine C developers helped me and welcomed me into the community. Jun Peng-Lao really helped me and Yeah, like that. I made a lot of mistakes on those notebooks and they helped me understand how to fix them, how to take up good practices for cutting together and developing in a team. And yeah, that just jump started my learning by 10 x I think is just because I was doing it and not only reading the book, but also contributing.

And the other thing I did was just also show up in the community. At some point, like my morning routine was going on the P discourse and reading the answers to the questions that I found interesting and often June Peng Lao was answering incredible stuff. Most of the time I didn’t understand. Sometimes I understood and with time I was understanding more and more of the answers. And also I was giving some of the answers with time. So that’s also something that helped me tremendously because it was very active. I had to understand what was written and also I had to understand what I was writing. So super important.

[00:37:00] Alexander: Love it. Love it. I love these kind of tips. Start doing it instead of just reading about it and get into the community. Awesome tips. Yeah. Thanks so much.

[00:37:10] Alex Andorra: The community is really awesome. So like for me, it help. And also that’s why we have this community in the intuitive patient, core space. We were wanted to recreate that thing that microcosm with Thomas and Ravin because it’s actually something that. Really help you instead of learning on your own, learning with other people who are at the same place or a bit more advanced than you. Will help you tremendously. Yeah.

[00:37:33] Alexander: Yeah. And it helps you to keep moving, have fun at the same time. And that’s makes so much a difference.

[00:37:41] Alex Andorra: Yeah.

[00:37:41] Alexander: Thanks so much.

[00:37:43] Alex Andorra: Yeah, thank you guys. That was really awesome. Really enjoyed the conversations, so thanks so for inviting me.