WEBVTT 00:00:00.001 --> 00:00:02.620 This episode, we have an optimization twofer. 00:00:02.620 --> 00:00:07.540 We begin by looking at optimizing a subset of Python code for machine learning 00:00:07.540 --> 00:00:12.740 using the LLVM compiler with a project called PyLLVM. 00:00:12.740 --> 00:00:16.840 It takes plain Python code, compiles it to optimize machine instructions, 00:00:16.840 --> 00:00:19.920 and distributes it across a cluster to do machine learning. 00:00:19.920 --> 00:00:23.860 In the second half, we'll look at a fabulous new way to work with MongoDB 00:00:23.860 --> 00:00:25.980 for Python writing data scientists. 00:00:26.240 --> 00:00:31.340 The project is called Bison NumPy and provides direct connections 00:00:31.340 --> 00:00:33.300 between MongoDB and NumPy. 00:00:33.300 --> 00:00:37.140 It's 10 times faster than working with PyMongo directly 00:00:37.140 --> 00:00:39.320 if you plan to end up in NumPy anyway. 00:00:39.320 --> 00:00:43.660 You're about to meet the woman behind both of these projects, Anna Herlihy. 00:00:43.660 --> 00:00:49.500 This is Talk Python To Me, episode 103, recorded February 6, 2017. 00:00:52.480 --> 00:01:17.560 Welcome to Talk Python To Me, a weekly podcast on Python, 00:01:17.560 --> 00:01:20.620 the language, the libraries, the ecosystem, and the personalities. 00:01:21.260 --> 00:01:22.760 This is your host, Michael Kennedy. 00:01:22.760 --> 00:01:24.760 Follow me on Twitter where I'm @mkennedy. 00:01:24.760 --> 00:01:28.640 Keep up with the show and listen to past episodes at talkpython.fm 00:01:28.640 --> 00:01:31.180 and follow the show on Twitter via at Talk Python. 00:01:31.180 --> 00:01:36.380 This episode is brought to you by Talk Python Training and Hired. 00:01:36.380 --> 00:01:39.720 Be sure to check out what we both have to offer during our segments. 00:01:39.720 --> 00:01:40.740 It helps support the show. 00:01:40.740 --> 00:01:43.640 Anna, welcome to Talk Python. 00:01:43.640 --> 00:01:44.520 Thank you. 00:01:44.520 --> 00:01:45.780 Thank you for having me. 00:01:45.780 --> 00:01:46.940 Yeah, it's great to have you here. 00:01:46.940 --> 00:01:49.660 We got a couple of really cool things to talk about. 00:01:49.660 --> 00:01:55.460 We're going to talk about PyLLVM, which is a really cool project that you worked on. 00:01:55.460 --> 00:01:59.800 And we're also going to talk about a super high-performance sort of data science-y thing 00:01:59.800 --> 00:02:02.180 with MongoDB that you're also working on right now. 00:02:02.180 --> 00:02:04.800 So I'm looking forward to talking about both of those. 00:02:04.800 --> 00:02:06.980 But before we get into those, what's your story? 00:02:06.980 --> 00:02:08.240 How did you get into programming in Python? 00:02:08.640 --> 00:02:12.560 I had my first real experience with programming. 00:02:12.560 --> 00:02:18.380 I mean, I had done some basic when I was pretty small, but I thought it was not very fun. 00:02:18.380 --> 00:02:20.880 So I kind of put that off for another five or six years. 00:02:20.880 --> 00:02:24.820 But in university, I was kind of all over the place. 00:02:24.820 --> 00:02:27.080 I didn't know if I wanted to be a writer or an engineer. 00:02:27.440 --> 00:02:31.280 But I figured that computers are probably going to be relevant. 00:02:31.280 --> 00:02:38.460 So I took a CS class because I didn't know the difference between computer science and computer 00:02:38.460 --> 00:02:38.900 literacy. 00:02:38.900 --> 00:02:43.260 But it ended up actually working out in my favor because I really, really enjoyed it. 00:02:43.260 --> 00:02:47.800 And I ended up prioritizing my CS classes over all my other classes. 00:02:47.800 --> 00:02:50.100 And then I kind of figured, maybe I should just do this full time. 00:02:50.100 --> 00:02:51.380 That's really cool. 00:02:51.460 --> 00:02:52.680 What was that first CS class? 00:02:52.680 --> 00:02:59.900 That was Andy Van Dam's intro to computer science, which for anybody who went to Brown will recognize 00:02:59.900 --> 00:03:02.000 that it's quite a popular course. 00:03:02.000 --> 00:03:05.700 I mean, when I took it, it was maybe 150 people. 00:03:05.700 --> 00:03:09.380 But I think just last year, they had 250 or something. 00:03:09.380 --> 00:03:11.120 It's really, really blowing up. 00:03:11.120 --> 00:03:11.920 Oh, that's excellent. 00:03:11.920 --> 00:03:13.420 What language did you study? 00:03:13.420 --> 00:03:13.940 Do you remember? 00:03:13.940 --> 00:03:15.860 Yeah, it was a Java course. 00:03:15.860 --> 00:03:17.200 We did a lot of swings. 00:03:17.200 --> 00:03:18.520 Okay, excellent. 00:03:18.520 --> 00:03:19.400 Yeah. 00:03:19.500 --> 00:03:22.100 And so you decided, hey, I kind of like this programming stuff. 00:03:22.100 --> 00:03:23.700 Let's just do that as a job, right? 00:03:23.700 --> 00:03:29.320 And how did you go from learning Java and CS 101 to working in Python? 00:03:29.320 --> 00:03:34.860 So I really liked programming languages in general. 00:03:34.860 --> 00:03:38.100 So I wrote some compilers, some interpreters. 00:03:38.100 --> 00:03:44.120 My area in computer science was mostly looking at different programming languages. 00:03:44.260 --> 00:03:48.280 And I found that Python was kind of the most elegant, in my opinion. 00:03:48.280 --> 00:03:53.000 And I didn't feel bogged down by really complicated syntax. 00:03:53.000 --> 00:03:54.580 But I also felt very powerful. 00:03:54.580 --> 00:04:02.180 So I kind of felt that as opposed to being pure Python, I wanted to do more like Python and C stuff, 00:04:02.180 --> 00:04:07.540 sort of more from the very beginning to how it gets down to the computer code at the end. 00:04:07.660 --> 00:04:07.920 I see. 00:04:07.920 --> 00:04:11.740 So you really were interested in like the internals and stuff and whatnot, huh? 00:04:11.740 --> 00:04:12.080 Yes. 00:04:12.080 --> 00:04:12.440 Okay. 00:04:12.440 --> 00:04:13.520 Cool. 00:04:13.520 --> 00:04:18.740 And so the first project we're going to talk about, which I want to ask you a question quick before then, 00:04:18.740 --> 00:04:23.380 but PyLLVM, that was part of your university work, right? 00:04:23.380 --> 00:04:24.400 Yes, exactly. 00:04:24.400 --> 00:04:31.820 The project that it was based on Tupleware is a research project with some of the database professors at Brown. 00:04:32.260 --> 00:04:36.040 And my project was a senior thesis that was built sort of on top of that. 00:04:36.040 --> 00:04:36.340 Okay. 00:04:36.340 --> 00:04:36.700 Wow. 00:04:36.700 --> 00:04:37.280 Very cool. 00:04:37.280 --> 00:04:37.700 Very cool. 00:04:37.700 --> 00:04:38.400 Okay. 00:04:38.400 --> 00:04:40.460 So today you work at MongoDB, right? 00:04:40.460 --> 00:04:40.840 Yes. 00:04:40.840 --> 00:04:41.200 Yeah. 00:04:41.200 --> 00:04:41.840 What do you do there? 00:04:41.840 --> 00:04:47.580 So I actually recently switched over from spending the majority of my time on Python work. 00:04:47.580 --> 00:04:54.280 So like PyMongo, Mongo Connector, other sort of Python related projects to MongoDB Compass, 00:04:55.000 --> 00:04:58.820 which is basically a user interface for MongoDB. 00:04:58.820 --> 00:05:04.500 It's actually written in JavaScript, which was something new for me since I had done a pretty 00:05:04.500 --> 00:05:06.780 excellent job of avoiding JavaScript until now. 00:05:06.780 --> 00:05:08.440 It finally got you, huh? 00:05:08.440 --> 00:05:09.380 Yeah, I know. 00:05:09.380 --> 00:05:10.080 It was inevitable. 00:05:10.080 --> 00:05:11.580 But now I really like it. 00:05:11.580 --> 00:05:16.980 I can't say that I am converted and I still think that Python is the community and the language 00:05:16.980 --> 00:05:17.860 that I like best. 00:05:17.860 --> 00:05:19.940 I'm probably preaching to the choir here. 00:05:20.180 --> 00:05:26.420 But I now split my time between working on Compass and working on this BSON NumPy package, 00:05:26.420 --> 00:05:28.140 which I hope we'll get to talk about later. 00:05:28.140 --> 00:05:30.400 Yeah, we'll definitely talk about it because it's very cool. 00:05:30.400 --> 00:05:31.700 Okay. 00:05:31.700 --> 00:05:32.180 Excellent. 00:05:32.180 --> 00:05:35.080 So it sounds like you have a lot of cool projects going on. 00:05:35.080 --> 00:05:39.380 Let's talk about the project that you did first with this LLVM thing. 00:05:39.380 --> 00:05:45.560 So I suspect a number of people out there know what LLVM is, but the audience is diverse. 00:05:45.560 --> 00:05:46.900 There's people from all sorts of places. 00:05:46.900 --> 00:05:49.880 Let's start with just talking about what is LLVM? 00:05:49.880 --> 00:05:50.380 Cool. 00:05:50.380 --> 00:05:54.640 So despite the name, it doesn't really have too much to do with virtual machines. 00:05:54.640 --> 00:05:58.360 It's basically just a collection of compiler technologies. 00:05:58.360 --> 00:06:01.340 So the LLVM project itself is huge. 00:06:01.340 --> 00:06:07.080 But what I worked with is LLVM IR, which stands for intermediate representation. 00:06:07.080 --> 00:06:15.280 And LLVM IR is basically a way of representing code that's about halfway between a top level 00:06:15.280 --> 00:06:15.780 language. 00:06:15.780 --> 00:06:19.340 So for example, Python and some machine code. 00:06:19.340 --> 00:06:21.160 So something that you wouldn't really want to read. 00:06:21.160 --> 00:06:25.160 You could probably look at the assembler and know what was going on, but it's not something 00:06:25.160 --> 00:06:26.440 you would ever really want to write. 00:06:26.440 --> 00:06:30.900 But runs much, much faster than something that a human would be able to write. 00:06:30.900 --> 00:06:31.180 Sure. 00:06:31.180 --> 00:06:32.220 So does... 00:06:32.220 --> 00:06:34.020 I don't know all that much about it, actually, that myself. 00:06:34.200 --> 00:06:42.200 So do you compile the source language into this LLVM IR and then do a further compilation 00:06:42.200 --> 00:06:43.460 towards some final target? 00:06:43.460 --> 00:06:44.940 Yes, exactly. 00:06:44.940 --> 00:06:50.460 And that's actually what makes it so powerful, is that a lot of the time you have compilers 00:06:50.460 --> 00:06:51.660 that have to be very specialized. 00:06:52.060 --> 00:06:54.940 So they take a language and they take a platform. 00:06:54.940 --> 00:07:01.860 And that is what the compiler does, is it compiles Python for a particular architecture. 00:07:01.860 --> 00:07:07.940 What LLVM IR does, which is not unique to LLVM, it's true for all intermediate representations 00:07:07.940 --> 00:07:14.660 or most of them, is that it's a way of making things both language and platform agnostic. 00:07:14.660 --> 00:07:23.180 So you can take any popular language you want from Java, R, Python, and you can compile it 00:07:23.180 --> 00:07:24.580 down to LLVM IR. 00:07:24.580 --> 00:07:30.680 Then you can take that LLVM IR and you can ship it anywhere and it can be compiled down to most 00:07:30.680 --> 00:07:31.300 platforms. 00:07:31.300 --> 00:07:36.100 So it doesn't even matter what the original language was and it doesn't matter what platform 00:07:36.100 --> 00:07:37.280 you're eventually going to run it on. 00:07:37.280 --> 00:07:39.260 It makes the code really... 00:07:39.260 --> 00:07:43.260 It makes it very cross-platform and it makes it fast. 00:07:43.260 --> 00:07:44.380 Yeah, sure. 00:07:44.480 --> 00:07:49.600 So basically, if you can get something to compile down to LLVM IR, you can get it to 00:07:49.600 --> 00:07:52.280 run quickly on many, many places, right? 00:07:52.280 --> 00:07:59.680 Because there's existing infrastructure to turn LLVM IR into executable code on all sorts 00:07:59.680 --> 00:08:00.460 of platforms, right? 00:08:00.460 --> 00:08:01.040 Exactly. 00:08:01.040 --> 00:08:02.340 And it's super powerful. 00:08:02.340 --> 00:08:09.180 The optimizations that the LLVM IR compilers do make it so that it doesn't matter if you wrote 00:08:09.180 --> 00:08:11.900 your code in C or if you wrote your code in Python. 00:08:12.180 --> 00:08:15.780 I'm of the opinion that it's much easier to write your code in Python than it is to write 00:08:15.780 --> 00:08:16.200 it in C. 00:08:16.200 --> 00:08:21.540 And so it's pretty nice because you get the benefits of maybe a more difficult syntax with 00:08:21.540 --> 00:08:23.700 pretty much whatever you're comfortable in. 00:08:23.700 --> 00:08:24.880 Yeah, that's really cool. 00:08:24.880 --> 00:08:29.360 What's the weirdest thing that you can execute this stuff on? 00:08:29.360 --> 00:08:30.180 What's the weirdest platform? 00:08:31.020 --> 00:08:31.940 That's a good question. 00:08:31.940 --> 00:08:33.140 Actually, I don't know. 00:08:33.140 --> 00:08:39.300 But I think it's pretty easy to if you just look up the LLVM IR or the LLVM compiler project, 00:08:39.300 --> 00:08:44.580 I think it's a really good way to get to know compilers in general. 00:08:44.580 --> 00:08:51.200 So there are a lot of proof of concept compilers that use LLVM because the infrastructure is so 00:08:51.200 --> 00:08:51.460 good. 00:08:51.580 --> 00:08:57.780 So I'm sure somebody has written something for an extremely obscure platform just to show 00:08:57.780 --> 00:08:58.640 that they could do it. 00:08:58.640 --> 00:08:59.380 Yeah, I'm sure. 00:08:59.380 --> 00:09:05.120 Do you think that it makes a lot of sense for people to create from scratch compilers these 00:09:05.120 --> 00:09:05.360 days? 00:09:05.360 --> 00:09:08.040 Or should most people just be building on LLVM? 00:09:08.040 --> 00:09:12.620 I think it depends on what the goal of your project is. 00:09:13.060 --> 00:09:19.120 So because this was a project that was done in the world of academia, I pretty much just 00:09:19.120 --> 00:09:21.580 picked whatever seemed like would be the most fun. 00:09:21.580 --> 00:09:28.520 And I know that's not necessarily true for a lot of people who have needs and users and they 00:09:28.520 --> 00:09:29.320 have to work quickly. 00:09:29.320 --> 00:09:35.940 And so I think it just depends on if you want something to work faster or if you want to learn 00:09:35.940 --> 00:09:37.120 a lot while you're doing it. 00:09:37.120 --> 00:09:37.360 Sure. 00:09:37.360 --> 00:09:37.600 Yeah. 00:09:37.600 --> 00:09:41.860 And I guess also like you only had, it was like a senior project sort of thing, right? 00:09:41.880 --> 00:09:45.440 So you couldn't start from scratch and do a ton of work. 00:09:45.440 --> 00:09:48.140 You had a time frame and a little amount of time and energy, right? 00:09:48.140 --> 00:09:49.120 So I guess that's a- 00:09:49.120 --> 00:09:49.120 Exactly. 00:09:49.120 --> 00:09:49.860 I had to graduate. 00:09:49.860 --> 00:09:50.600 Exactly. 00:09:50.600 --> 00:09:52.540 So that's a pretty good testament to LLVM. 00:09:52.540 --> 00:09:53.340 Cool. 00:09:53.340 --> 00:09:58.920 And I know it's like used for Swift and some other things on the Apple platform and it's 00:09:58.920 --> 00:09:59.240 pretty cool. 00:09:59.240 --> 00:10:00.200 Okay. 00:10:00.200 --> 00:10:05.680 So you also talked about a project to do with machine learning called Tupleware. 00:10:05.680 --> 00:10:07.540 And this is a project at Brown, right? 00:10:07.540 --> 00:10:08.820 Yes, it is. 00:10:08.820 --> 00:10:18.920 So Tupleware's tagline is that it is a distributed analytical platform that leverages LLVM IR to 00:10:18.920 --> 00:10:23.220 be totally platform agnostic and totally language agnostic. 00:10:23.880 --> 00:10:31.840 So the way that PY LLVM fits into Tupleware, it was basically a proof of concept that you 00:10:31.840 --> 00:10:38.040 could take Python code and you could compile it down to LLVM IR and then you could ship 00:10:38.040 --> 00:10:44.360 it to your clusters and then have them be automatic and have your code be automatically run on a 00:10:44.360 --> 00:10:45.320 very large dataset. 00:10:45.540 --> 00:10:46.020 Okay, cool. 00:10:46.020 --> 00:10:52.760 So this is like a distributed machine learning type of system and your project PY LLVM basically 00:10:52.760 --> 00:10:57.420 made it possible to feed Python instead of say C++ code or something like that to it, right? 00:10:57.420 --> 00:10:57.940 Exactly. 00:10:57.940 --> 00:11:04.200 And I chose Python because I didn't really want to write an R compiler, but also because at 00:11:04.200 --> 00:11:10.900 the time I was working on this, which is about three, three and a half years ago now, the language 00:11:10.900 --> 00:11:14.900 that most scientists were using for programming was still MATLAB. 00:11:14.900 --> 00:11:21.940 I actually had a summer job where I converted MATLAB code to scientific Python code and it was 00:11:21.940 --> 00:11:26.260 becoming more and more common that people who were not programmers would be programming in Python 00:11:26.260 --> 00:11:29.820 instead of just inheriting these old MATLAB scripts that they would reuse. 00:11:29.820 --> 00:11:32.860 So that's why I've picked Python. 00:11:32.860 --> 00:11:33.980 Yeah, that's cool. 00:11:33.980 --> 00:11:35.220 I've definitely seen that as well. 00:11:35.220 --> 00:11:40.880 When I was in school, MATLAB was definitely the thing and people would have all these scripts 00:11:40.880 --> 00:11:47.360 and one of my first programming jobs was to take a bunch of MATLAB code and turn it into like 00:11:47.360 --> 00:11:54.140 graphical visualizations on Silicon Graphics supercomputers and I had to convert the MATLAB code to C++ 00:11:54.140 --> 00:11:57.900 but if it was today, I very well might have done that in Python instead, right? 00:11:57.900 --> 00:11:59.560 Yeah, that's funny. 00:11:59.560 --> 00:12:00.560 Sounds like the same job. 00:12:00.560 --> 00:12:01.600 Yeah, exactly. 00:12:01.600 --> 00:12:02.100 Exactly. 00:12:02.100 --> 00:12:03.380 Awesome. 00:12:03.380 --> 00:12:08.440 And so you said PY LLVM started from an abandoned open source project. 00:12:08.440 --> 00:12:09.780 What's the story there? 00:12:09.780 --> 00:12:15.180 Basically, once I had pinned down what exactly I wanted to do for my thesis, 00:12:15.180 --> 00:12:20.160 I went on a search spree where I tried to figure out if somebody else had done it for me, 00:12:20.160 --> 00:12:21.700 which would have been very convenient. 00:12:21.700 --> 00:12:32.900 But at the time, I couldn't find any Python 2 LLVM IR compilers that were already up and running or already published. 00:12:32.900 --> 00:12:36.680 So what I did is I found this GitHub. 00:12:36.680 --> 00:12:37.840 Actually, it wasn't even GitHub. 00:12:38.020 --> 00:12:39.500 It was a Google code, right? 00:12:39.500 --> 00:12:46.060 Yeah, it was a Google code project, which had been documented almost zero. 00:12:46.060 --> 00:12:52.200 There were some slides in Japanese that I found, which I tried to feed through Google Translate, which did not work. 00:12:52.800 --> 00:12:55.180 So it was basically just some code. 00:12:55.180 --> 00:13:04.560 And it ended up being almost exactly what I needed because it was a pretty simple compiler outline without any of it really implemented. 00:13:04.560 --> 00:13:06.100 But the structural stuff was there. 00:13:06.280 --> 00:13:11.660 So a lot of the scoping and a lot of the variable tracking had been started but not finished. 00:13:11.880 --> 00:13:21.440 So it was kind of the perfect project to pick up because a lot of the design decisions that would have resulted in a lot of cost versus value arguments. 00:13:21.440 --> 00:13:28.740 So, you know, we could implement it this way, but it would take a lot of time and we might not actually get that much out of it was sort of already decided for me. 00:13:28.860 --> 00:13:38.620 So I had a pretty clear path in terms of getting the subset of Python it supported to be something that I thought a user could actually make use of. 00:13:38.620 --> 00:13:42.680 Yeah, that seems really great that, you know, that's out there and you can pick it up. 00:13:42.680 --> 00:13:49.760 And it's funny it was on Google Code because not only was the project kind of abandoned, like Google Code itself is kind of in archive mode, right? 00:13:49.760 --> 00:13:51.740 I guess it's a real testament to GitHub. 00:13:51.740 --> 00:13:55.080 I think it was in archive mode when I found it even. 00:13:55.080 --> 00:13:57.580 It's definitely in archive mode now. 00:13:57.580 --> 00:13:58.700 It's several layers. 00:13:58.700 --> 00:14:07.660 Yeah, but that's helpful that it's like, well, we already kind of put the thinking into place, but hey, I need to – they just didn't take it to completion, right? 00:14:07.660 --> 00:14:13.140 And so, like, okay, these are the final to-dos on this project or whatever until I can actually use it on Tupleware. 00:14:13.140 --> 00:14:15.760 Yeah, I mean, when I found it, it wouldn't even compile. 00:14:15.760 --> 00:14:28.440 So I have really a lot of respect for the person that wrote it because it really had everything – it was clear that this person didn't actually run it, but they had thought about it a lot. 00:14:28.440 --> 00:14:34.000 And the skeleton was there, but it hadn't actually gone through the process of becoming a tool that could be run. 00:14:34.000 --> 00:14:36.240 And that, to me, is very incredible. 00:14:36.400 --> 00:14:39.960 I mean, I run my code every 10 seconds just to make sure that it's working. 00:14:39.960 --> 00:14:40.560 Exactly. 00:14:40.560 --> 00:14:43.020 Yeah, I'm with you on that. 00:14:43.020 --> 00:14:43.860 Like, run it often. 00:14:43.860 --> 00:14:56.420 So, yeah, I guess some people – you know, probably more in the early days when compiling took longer and stuff, but some people really sit there and work through it, and then they run it. 00:14:56.420 --> 00:14:59.360 You know, hours after hours of work, that's just how I work. 00:14:59.360 --> 00:15:00.260 I run it often. 00:15:00.260 --> 00:15:02.400 Yeah, it's very impressive. 00:15:02.400 --> 00:15:03.460 Yes, indeed. 00:15:03.460 --> 00:15:05.320 It takes a subset of Python. 00:15:05.320 --> 00:15:08.900 I can't feed it like Django and have it spit that out, right? 00:15:08.900 --> 00:15:16.540 Because what happens is it takes the Python language and it turns it to LLVMIR, which then finally compiles down to machine instructions. 00:15:16.540 --> 00:15:19.940 But there's no interpreter. 00:15:19.940 --> 00:15:24.680 There's no standard library underneath there for it to run, right? 00:15:24.720 --> 00:15:26.660 So, you've got to be pretty focused on what you give it. 00:15:26.660 --> 00:15:27.020 Yes. 00:15:27.020 --> 00:15:31.240 What it does have is sort of the C standard library. 00:15:31.240 --> 00:15:43.820 So, for example, if you want to call, like, Printline, you can do some somewhat complicated gymnastics in order to get LLVM to make that external call. 00:15:44.240 --> 00:15:49.740 So, that's really nice because getting Printline to work when you're writing a compiler is so important. 00:15:49.740 --> 00:15:53.120 And when it does work, it's, like, really one of the best feelings ever. 00:15:53.120 --> 00:16:05.560 But generally, no, if you wouldn't be able to import a Python package, for example, since the goal of the project was really just to provide people an alternative to using MATLAB. 00:16:05.680 --> 00:16:10.040 So, instead of writing your machine learning algorithms in MATLAB, you can write them in Python. 00:16:10.040 --> 00:16:14.660 Or, you know, in this case specifically, it'd be instead of writing in C++, you're writing it in Python. 00:16:14.660 --> 00:16:18.620 But the nature of those algorithms is that they're quite simple. 00:16:18.620 --> 00:16:27.500 So, we didn't – I didn't spend a lot of time trying to get objects or more complicated data structures working because I didn't anticipate it would be that common. 00:16:27.500 --> 00:16:28.060 Right. 00:16:28.200 --> 00:16:33.680 So, you're working with, like, loops and join and map and stuff like that. 00:16:33.680 --> 00:16:36.980 And then you feed it basic algorithms, huh? 00:16:36.980 --> 00:16:40.440 So, like, Bayesian stuff or linear regression. 00:16:40.440 --> 00:16:42.300 Yeah, exactly. 00:16:42.300 --> 00:16:43.160 You feed those off, huh? 00:16:43.160 --> 00:16:43.420 Okay. 00:16:43.420 --> 00:16:49.240 And I'm not a machine learning – I had not really done much machine learning at that time. 00:16:49.240 --> 00:16:59.320 So, I basically just went to my ML professor and was like, give me, like, your top five most common machine learning algorithms that you would expect to want to run on large data sets. 00:16:59.320 --> 00:17:02.680 I'll make these run in Python through this thing, right? 00:17:02.680 --> 00:17:03.780 Nice. 00:17:03.780 --> 00:17:05.080 And so, yeah. 00:17:05.080 --> 00:17:06.660 So, you basically – you write the algorithms. 00:17:06.660 --> 00:17:12.700 They compile down and then finally get transformed to run basically on the C, C++ runtime, like you said. 00:17:12.700 --> 00:17:15.740 And that goes through something called Boost Python, right? 00:17:15.740 --> 00:17:16.540 What is Boost Python? 00:17:16.840 --> 00:17:25.640 So, Boost Python – Boost is a bunch of packages for C++ that provide a lot of really, really powerful capabilities. 00:17:25.640 --> 00:17:31.620 And so, Boost Python is actually just an easier way of calling the Python C API. 00:17:31.620 --> 00:17:38.740 If I had done this now, I probably just would have skipped using Boost entirely because I'm pretty comfortable with the Python C API. 00:17:38.740 --> 00:17:45.380 But at the time, it was sort of the first Python plus another language interfacing I had done before. 00:17:45.760 --> 00:17:51.760 And this was pretty simple because it was already being used by Tupperware itself. 00:17:51.760 --> 00:17:55.840 So, I didn't have to do a lot of – I didn't actually have to incorporate it into the ecosystem. 00:17:55.840 --> 00:17:56.800 It was already there. 00:17:56.800 --> 00:17:57.120 Yeah. 00:17:57.120 --> 00:17:57.780 That's cool. 00:17:57.780 --> 00:17:58.080 Yeah. 00:17:58.080 --> 00:18:02.060 And by now, you've done so much with PyMongo and all the stuff in MongoDB. 00:18:02.060 --> 00:18:02.480 Yeah. 00:18:02.480 --> 00:18:02.900 Yeah. 00:18:02.900 --> 00:18:03.260 Cool. 00:18:03.260 --> 00:18:03.700 Okay. 00:18:03.700 --> 00:18:08.500 So, give us, like, a short example of what this algorithm might look like. 00:18:08.560 --> 00:18:11.040 I mean, code over audio is really hard. 00:18:11.040 --> 00:18:14.300 But just, like, what kind of stuff would you feed off to this system? 00:18:14.300 --> 00:18:19.300 So, I would expect there to be – so, you've got to assign to your variables. 00:18:19.300 --> 00:18:25.880 You most likely have an array that you are assigning values to. 00:18:26.420 --> 00:18:31.240 You need to iterate through the array or you need to iterate through the data that you've been given. 00:18:31.240 --> 00:18:35.280 And you need to do a lot of – or at least a decent amount of error checking. 00:18:35.280 --> 00:18:42.520 So, what that means in terms of syntax is you would want to have reassignable variables. 00:18:42.520 --> 00:18:47.460 You would need to have conditionals, loops, and arithmetic. 00:18:47.980 --> 00:18:49.780 Those are basically the most important things. 00:18:49.780 --> 00:18:53.640 This portion of Talk Python To Me is brought to you by us. 00:18:53.640 --> 00:18:59.540 As many of you know, I have a growing set of courses to help you go from Python beginner to novice to Python expert. 00:18:59.540 --> 00:19:01.580 And there are many more courses in the works. 00:19:01.580 --> 00:19:05.820 So, please consider Talk Python training for you and your team's training needs. 00:19:05.820 --> 00:19:12.260 If you're just getting started, I've built a course to teach you Python the way professional developers learn, by building applications. 00:19:12.260 --> 00:19:17.320 Check out my Python Jumpstart by building 10 apps at talkpython.fm/course. 00:19:17.320 --> 00:19:20.460 Are you looking to start adding services to your app? 00:19:20.460 --> 00:19:23.620 Try my brand new Consuming HTTP Services in Python. 00:19:23.620 --> 00:19:29.060 You'll learn to work with RESTful HTTP services as well as SOAP, JSON, and XML data formats. 00:19:29.060 --> 00:19:30.920 Do you want to launch an online business? 00:19:30.920 --> 00:19:35.120 Well, Matt Makai and I built an Entrepreneur's Playbook with Python for Entrepreneurs. 00:19:35.120 --> 00:19:40.360 This 16-hour course will teach you everything you need to launch your web-based business with Python. 00:19:40.360 --> 00:19:43.900 And finally, there's a couple of new course announcements coming really soon. 00:19:43.900 --> 00:19:49.360 So, if you don't already have an account, be sure to create one at training.talkpython.fm to get notified. 00:19:49.360 --> 00:19:53.140 And for all of you who have bought my courses, thank you so much. 00:19:53.140 --> 00:19:55.080 It really, really helps support the show. 00:19:55.080 --> 00:20:00.360 One of the things that seems a little challenging to me is like you're building something in Python 00:20:00.360 --> 00:20:09.580 and it's being more or less just compiled into the results of C++, which is a typed system, right? 00:20:09.580 --> 00:20:14.120 So, it expects, you know, here's a four-byte integer. 00:20:14.120 --> 00:20:16.000 Here's a Boolean. 00:20:16.000 --> 00:20:17.560 Here's a string pointer, right? 00:20:17.560 --> 00:20:20.840 Whereas in Python, you don't really have that, right? 00:20:20.940 --> 00:20:24.800 So, what do you have to do to make the type system sort of fit together there? 00:20:24.800 --> 00:20:32.800 That is a problem that comes up sort of time and time again because no matter how many layers of abstraction you're talking about, 00:20:32.800 --> 00:20:36.680 ultimately, machine code is pretty strictly typed. 00:20:36.680 --> 00:20:40.840 And we don't really like working in strictly typed languages. 00:20:40.840 --> 00:20:53.740 So, there's a lot of cost-benefit analysis going on there where you could say I demand that I can reassign my variable from an array to an integer 00:20:53.740 --> 00:20:56.920 or between different types of numbers, for example. 00:20:56.920 --> 00:21:04.820 Or you could say that's going to save me a ton of work if I just tell my user who I anticipate is probably used to MATLAB. 00:21:04.820 --> 00:21:06.360 Don't do that. 00:21:06.360 --> 00:21:08.900 This is Python, but it's not that Pythonic. 00:21:08.900 --> 00:21:13.000 Like, you know, maybe down the line when this is no longer a proof of concept. 00:21:13.000 --> 00:21:22.840 Ultimately, what we ended up doing was just sticking to pretty static types because whenever you write a compiler, 00:21:22.840 --> 00:21:32.960 it's kind of a debate between how much do I want to reinvent the wheel and how much are my users going to be willing to sort of have a more limited experience 00:21:32.960 --> 00:21:37.520 for the sake of, like, my sanity or how much time I'm willing to put in on this project. 00:21:37.520 --> 00:21:39.140 Sure, that makes a lot of sense. 00:21:39.140 --> 00:21:45.800 I feel like because it's such a limited set of types, really, that you can write your algorithm in, 00:21:45.800 --> 00:21:51.640 it's not so bad to sort of restrict it and talk about the types, right? 00:21:51.640 --> 00:21:57.880 You've got, like, the fundamental numerical types, strings, lists, and a few other things that they can really work with, right? 00:21:58.140 --> 00:22:02.980 Yeah, and I definitely, that's what we're trying to sell here. 00:22:03.060 --> 00:22:11.680 But I remember having this argument where I came up with what I thought was, like, a super clever solution for how we're going to do dynamic types. 00:22:11.680 --> 00:22:16.420 And it was, like, twice as much code as the entire compiler was before that point. 00:22:16.420 --> 00:22:17.560 And I brought it to my advisor. 00:22:17.560 --> 00:22:19.300 I was like, this is such a good idea. 00:22:19.440 --> 00:22:21.400 Like, I can't wait to implement this. 00:22:21.400 --> 00:22:24.840 And he kind of looked at me and he was like, this is not what we need right now. 00:22:24.840 --> 00:22:33.080 What we need is something that is working, which is, I guess, was my first experience with, you know, writing code that actually needs to do something. 00:22:33.080 --> 00:22:34.920 And it needs to do something as soon as possible. 00:22:35.320 --> 00:22:38.240 As opposed to code that's, like, beautiful and elegant. 00:22:38.240 --> 00:22:40.580 And I've, like, prototyped it three times. 00:22:40.580 --> 00:22:41.960 And I have all the time in the world. 00:22:41.960 --> 00:22:42.380 So. 00:22:42.380 --> 00:22:42.860 Yeah. 00:22:42.860 --> 00:22:44.720 You know, shipping is a feature, right? 00:22:44.720 --> 00:22:45.720 Yeah. 00:22:45.720 --> 00:22:49.860 No one's going to use your code if you don't actually get it out there and get it working, I guess. 00:22:49.860 --> 00:22:50.320 Yeah. 00:22:50.320 --> 00:22:52.020 I mean, I still struggle with that now. 00:22:52.020 --> 00:23:00.120 I definitely like to write code that is fully baked as opposed to just, like, getting stuff out of the door. 00:23:00.120 --> 00:23:03.460 But I'm definitely getting used to that now that I'm working in JavaScript more. 00:23:04.600 --> 00:23:05.540 Yeah, it's one benefit. 00:23:05.540 --> 00:23:06.200 That's awesome. 00:23:06.200 --> 00:23:07.220 Yeah. 00:23:07.220 --> 00:23:12.320 Yeah, I guess I really appreciate getting something out there so that people can use it and give me feedback. 00:23:12.320 --> 00:23:13.220 Like, this is working. 00:23:13.220 --> 00:23:14.040 This is not working. 00:23:14.040 --> 00:23:16.280 But, yeah, you just have to have some flexibility. 00:23:16.280 --> 00:23:19.940 You can't get locked into, like, some early prototype API or something, right? 00:23:19.940 --> 00:23:20.760 Yeah, exactly. 00:23:20.760 --> 00:23:21.260 Yeah. 00:23:21.260 --> 00:23:25.600 So you said that some of the LLVM IR features made your life easier and some made it harder. 00:23:25.600 --> 00:23:26.820 How'd that work out? 00:23:26.820 --> 00:23:41.320 So probably the most involved issue that comes up between converting from Python to a intermediate representation is like what you mentioned with types and reassigning. 00:23:41.940 --> 00:23:46.400 LLVM IR is written in SSA, which means static single assignment. 00:23:46.400 --> 00:23:56.300 And what that basically means is that you have your registers, which are your smallest unit of storage, and you can only assign to them once. 00:23:56.980 --> 00:24:07.120 So if you assign a number to your register, even if you want to assign something of the same type, it is frozen for that function call. 00:24:07.700 --> 00:24:13.600 So what I needed to do to get around that was basically move everything onto the stack. 00:24:13.600 --> 00:24:23.120 And once everything is on the stack, then you need to keep track of stuff using what people call a symbol table, which is basically just a dictionary where you can look up. 00:24:23.120 --> 00:24:31.880 I have a variable named X, and it lives at this memory address, and it's been around for this long and all sorts of other metadata like that. 00:24:31.880 --> 00:24:32.160 Okay. 00:24:32.160 --> 00:24:33.820 So, yeah. 00:24:33.820 --> 00:24:34.660 Does that make sense? 00:24:34.660 --> 00:24:35.140 Yeah, yeah. 00:24:35.300 --> 00:24:36.220 That's pretty interesting. 00:24:36.220 --> 00:24:38.060 You can only assign to them once. 00:24:38.060 --> 00:24:38.380 Okay. 00:24:38.380 --> 00:24:39.760 Pretty interesting. 00:24:39.760 --> 00:24:42.520 And then memory management, how did that work, actually? 00:24:42.520 --> 00:24:48.120 Because everything was on the stack, for the most part, things basically just took care of themselves. 00:24:48.120 --> 00:24:51.320 I didn't have to write a garbage collector, thankfully. 00:24:51.320 --> 00:24:57.120 But there was one particular instance that was really awkward. 00:24:57.840 --> 00:25:08.820 So if you are making a function call and you want to return something that can't fit into a register. 00:25:08.820 --> 00:25:17.360 So say, for example, you have an array, or you have a function call that populates an array, and then you want it to return the array to the original caller. 00:25:17.820 --> 00:25:22.280 The problem there is that you can no longer save that on the stack. 00:25:22.280 --> 00:25:26.920 Because as anyone who's ever programmed in C or C++ knows, it'll go out of scope. 00:25:26.920 --> 00:25:32.340 So the solution there is either, oh, do I have to keep track of scopes now? 00:25:32.340 --> 00:25:34.300 Do I actually have to write a garbage collector? 00:25:34.460 --> 00:25:40.660 Do I have to reinvent memory management in Python myself in two months? 00:25:40.660 --> 00:25:45.240 And the answer to that was, okay, no, that doesn't make sense either. 00:25:45.240 --> 00:25:56.760 So what ended up happening is I would just move the data that you put in your array onto the heap temporarily and then pass back a pointer to it. 00:25:56.920 --> 00:26:10.860 And then either copy it into the stack again and free it, or just chalk that up to a memory leak and let the stuff that you return from a function just call it a memory leak. 00:26:10.860 --> 00:26:13.940 So that's sort of a lose-lose situation. 00:26:13.940 --> 00:26:18.120 It was definitely one of those things that if I had had more time, I would have liked to dig into that. 00:26:18.120 --> 00:26:24.440 And dynamically reassigning variables to two parts of the compiler, I feel like, are unfinished. 00:26:24.440 --> 00:26:25.280 Sure. 00:26:25.280 --> 00:26:25.560 Okay. 00:26:26.260 --> 00:26:29.920 Yeah, that's a big challenge in adding your own garbage collector. 00:26:29.920 --> 00:26:31.240 That sounds like a lot of work. 00:26:31.240 --> 00:26:34.640 Sounds like it could be really fun, but it sounds like a different thesis. 00:26:34.640 --> 00:26:35.800 Exactly. 00:26:35.800 --> 00:26:37.520 It's not my problem right now. 00:26:37.520 --> 00:26:37.760 Okay. 00:26:37.760 --> 00:26:38.420 Interesting. 00:26:38.420 --> 00:26:40.540 So how was the performance? 00:26:40.540 --> 00:26:48.060 Say, like, one of the options was I could write my code in C++ and I could give it to Tupleware, or I could write it in Python and give it to Tupleware. 00:26:48.060 --> 00:26:50.320 What was the trade-off there in terms of performance? 00:26:50.320 --> 00:26:52.400 Was it huge or pretty close? 00:26:52.400 --> 00:26:58.220 So in terms of performance, I actually do have the numbers that I ran that I can dig up. 00:26:58.220 --> 00:27:07.840 But the bottom line was that the LLVM compiler itself was much faster. 00:27:08.000 --> 00:27:29.900 If you take C++ and you compile it down to LLVM IR using a compiler that I think is written in LLVM IR itself, or at least in C or C++, compared to a compiler that is written in pure Python that takes in Python code, does all the, builds the syntax tree, does the parsing, all the semantic error analysis, that kind of thing. 00:27:29.900 --> 00:27:35.720 It takes longer, but it doesn't take an order of magnitude longer. 00:27:35.720 --> 00:28:01.380 So if you're running ML algorithms on huge data sets, having the compilation of your algorithm, which only happens once, if it takes one second or if it takes five seconds, it doesn't really matter because we're talking about hours and hours of work. 00:28:01.380 --> 00:28:08.140 So it's definitely slower, but for what the project needed, it wasn't inhibitively slower. 00:28:08.140 --> 00:28:08.860 Yeah, yeah, sure. 00:28:08.860 --> 00:28:13.560 And for the execution speed, like would the algorithms run about the same or was it really different? 00:28:13.960 --> 00:28:20.480 So the algorithms itself, once they got down to LLVM IR, pretty much ran the same. 00:28:20.480 --> 00:28:22.620 There were a couple interesting cases there. 00:28:22.900 --> 00:28:40.640 So what actually happens when the code gets put into LLVM IR and is then shipped to the other half of the distributed system where it gets actually compiled, analyzed, and run, is that is all handled by LLVM and C++. 00:28:40.640 --> 00:28:42.820 There's no Python involved in that. 00:28:43.280 --> 00:28:49.420 And so the optimization passes that the LLVM IR compilers actually do are incredibly powerful. 00:28:50.240 --> 00:29:04.020 So the reason that there's not a lot of optimization happening in the compiler itself, I mean the compiler from Python to LLVM IR, is that it's pretty much going to get squashed no matter what with the LLVM passes themselves. 00:29:04.020 --> 00:29:11.920 So that was a huge benefit because I didn't really have to sweat optimizations, which is another huge part of compiler writing. 00:29:12.140 --> 00:29:12.720 Yeah, that's awesome. 00:29:12.720 --> 00:29:17.420 Just let LLVM do its analysis on the intermediate representation, yeah? 00:29:17.420 --> 00:29:20.060 Yeah, I mean it's a huge selling point of LLVM IR. 00:29:20.060 --> 00:29:25.800 But there is some interesting stuff about optimizing function calls. 00:29:25.800 --> 00:29:31.760 And basically if you have a recursive call, it becomes a lot more difficult. 00:29:31.760 --> 00:29:45.360 I basically discovered that the LLVM IR compiler is not as good at unrolling these recursive calls as it is when you give it Python code because there's sort of, if you're doing... 00:29:45.360 --> 00:29:47.920 It just basically doesn't like the recursion so much, huh? 00:29:47.920 --> 00:30:00.440 Yeah, there's some things that you have to do in Python that you don't have to do in C++ because it's closer to the end result that ends up tripping up the optimizer. 00:30:00.440 --> 00:30:06.760 And so for an algorithm that involves recursion, it will actually perform slower. 00:30:06.760 --> 00:30:07.720 Okay, interesting. 00:30:08.500 --> 00:30:21.180 Yeah, so this sounds like a really cool project if you have something super focused like Tupleware where you can take a really small subset of Python and execute it against that system, right? 00:30:21.180 --> 00:30:25.000 There's a bunch of different implementations or runtimes out there. 00:30:25.000 --> 00:30:29.980 So we've got things like IronPython and Jython that try to take a different take on Python. 00:30:29.980 --> 00:30:34.280 There's PyPy, there's Pigeon, there's Cython, and Numba. 00:30:35.180 --> 00:30:42.060 It sounds like you're much closer to something like Numba with this project than you would be, say, with Cython. 00:30:42.060 --> 00:30:52.820 Yeah, I think that a lot of the sort of the line in the sand that gets drawn between projects is everybody is trying to get Python code to run really, really fast. 00:30:52.820 --> 00:30:59.260 But the way that LLVM IR actually factors into it can vary a lot between projects. 00:30:59.260 --> 00:31:04.720 So the goal of Numba is to run your Python code super fast. 00:31:04.720 --> 00:31:10.400 LLVM IR is just one step in what is, I think, a six-step process. 00:31:10.400 --> 00:31:19.960 And there's not actually, there wasn't actually a way to extract the LLVM IR directly from Numba, which has changed now. 00:31:19.960 --> 00:31:23.160 But when I initially wrote it, they didn't have that ability. 00:31:23.660 --> 00:31:35.860 So I would have had to basically go in and pick and choose bits from their code and then move it into a separate project because there was no elegant way to really pull it out. 00:31:36.420 --> 00:31:48.140 Another huge difference is that a lot of these compilers are JITs, which work great for what they're trying to do and generally work faster. 00:31:48.140 --> 00:31:50.800 But the thing about JITs is that they're lazy. 00:31:51.580 --> 00:31:57.580 And if you have a lazy compiler, it won't actually compile anything unless it's run. 00:31:57.580 --> 00:32:02.200 But for Tupleware, we're completely unconcerned with running the code. 00:32:02.200 --> 00:32:07.480 We just want to compile it and then we want to take that compiled code and pass it off to somebody else. 00:32:07.480 --> 00:32:07.860 I see. 00:32:07.860 --> 00:32:14.340 So that's basically part of the mechanism for deploying to this distributed cluster, right? 00:32:14.340 --> 00:32:16.220 Because you've got to give it the executable code. 00:32:16.220 --> 00:32:16.660 I see. 00:32:16.660 --> 00:32:22.080 So sort of ahead of the time JIT compilation would be as close as you could get or something. 00:32:22.080 --> 00:32:26.920 I did some performance comparisons between Numba and PyLVM. 00:32:27.080 --> 00:32:33.360 But the problem there is that there's no way to actually run code using my compiler. 00:32:33.360 --> 00:32:37.520 But you have to run the code in order to compile stuff with Numba. 00:32:37.520 --> 00:32:48.760 So if you're doing benchmarking, the actual cost of the algorithm itself, it doesn't negate the data, but it makes it a pretty big asterisk. 00:32:48.760 --> 00:32:51.980 Like, by the way, we also had to run the algorithm. 00:32:51.980 --> 00:32:53.040 Yeah, sure. 00:32:53.040 --> 00:32:55.400 So it's hard to compare apples to apples. 00:32:55.400 --> 00:32:55.880 Yeah. 00:32:55.880 --> 00:32:56.520 Okay. 00:32:56.880 --> 00:33:02.540 So I guess maybe we should kind of wrap it up on PyLVM and talk about your MongoDB stuff. 00:33:02.540 --> 00:33:05.200 But what's the two quick questions? 00:33:05.200 --> 00:33:09.340 Is this a Python 2 or Python 3 project or both? 00:33:09.340 --> 00:33:13.460 This is Python 2.7 is what I wrote it in. 00:33:13.460 --> 00:33:13.980 Yeah, sure. 00:33:13.980 --> 00:33:15.200 Okay, cool. 00:33:15.200 --> 00:33:18.380 And what's the future for this project? 00:33:18.380 --> 00:33:23.560 Do you know if anyone's picking it up or, you know, people out there listening, if it sounds interesting, you can pick it up. 00:33:23.560 --> 00:33:24.300 It's on GitHub, right? 00:33:24.300 --> 00:33:25.540 Yes, it is on GitHub. 00:33:25.540 --> 00:33:30.640 I would really recommend people to look at the project and to contribute. 00:33:30.640 --> 00:33:35.300 But that is also from like a curiosity slash selfish interest. 00:33:35.300 --> 00:33:44.920 I think that if you're going to do it, there are a lot of design decisions that were made in the interest of this specific project. 00:33:45.560 --> 00:33:47.940 And so if it matches your use case, that's excellent. 00:33:47.940 --> 00:33:57.820 And so if it matches your use case, because there are so many variables, I recommend just writing one because it's probably one of my favorite projects I've ever worked on. 00:33:58.060 --> 00:34:04.820 Just because it's such a neat, like a very well defined and very satisfying problem to solve. 00:34:04.820 --> 00:34:05.080 Nice. 00:34:05.080 --> 00:34:10.640 Do you feel like you understand how a lot of these compilers and execution fits together better now? 00:34:10.640 --> 00:34:18.200 Yeah, I think it's probably the best learning project anyone could have is to actually understand what goes on under the hood of the language they use. 00:34:18.880 --> 00:34:26.040 It also made me much, much better at writing optimizations work. 00:34:26.040 --> 00:34:26.360 Okay. 00:34:26.360 --> 00:34:27.240 Yeah, very cool. 00:34:27.240 --> 00:34:28.400 All right. 00:34:28.400 --> 00:34:32.280 So let's talk about this, what you're up to these days at MongoDB. 00:34:32.280 --> 00:34:34.040 And you said you'd worked on PyMongo. 00:34:34.040 --> 00:34:37.360 And just for everyone listening, like that's the primary driver. 00:34:37.360 --> 00:34:43.440 The primary way to speak to MongoDB is to, you know, pip install PyMongo and import it. 00:34:43.440 --> 00:34:44.920 And then you just start talking. 00:34:44.920 --> 00:34:48.440 And basically the data exchange is dictionaries, right? 00:34:48.520 --> 00:34:55.380 You write a prototypical query sort of thing as a dictionary and you get back rows, which are documents in the forms of dictionaries, right? 00:34:55.380 --> 00:34:56.380 Yes, that's correct. 00:34:56.380 --> 00:34:59.520 So you don't actually have to get dictionaries anymore. 00:34:59.520 --> 00:35:04.320 For a long time, PyMongo would just automatically read your data into dictionaries. 00:35:04.320 --> 00:35:07.860 But now you can actually get raw BSON out of the driver. 00:35:07.860 --> 00:35:10.620 And that opens up a lot of doors for what you can do with it. 00:35:10.620 --> 00:35:10.820 Right. 00:35:10.820 --> 00:35:18.380 So BSON is binary JSON, which is the actual in-memory on-the-wire representation that you get talking to MongoDB. 00:35:19.140 --> 00:35:27.420 You had a cool talk about something called Monary, which is kind of getting superseded by the project that you're working on now. 00:35:27.420 --> 00:35:35.780 But you had some interesting performance numbers about getting dictionaries back or just in terms of analysis in general. 00:35:35.780 --> 00:35:37.180 Forget the database for a minute. 00:35:37.180 --> 00:35:42.320 Like working with dictionaries versus working with lists versus something like NumPy, right? 00:35:42.320 --> 00:35:42.740 Yes. 00:35:43.280 --> 00:35:59.700 So that was pretty enlightening for me as a relatively new Python programmer to realize that Python dictionaries, which I considered was kind of the most basic way of storing data in Python, was actually pretty slow. 00:36:00.040 --> 00:36:10.800 And compared to like ND arrays, which are C-style arrays that come with the NumPy package versus just something like a list, they are significantly slower. 00:36:10.800 --> 00:36:11.160 Yeah. 00:36:11.160 --> 00:36:12.020 You had some cool numbers. 00:36:12.020 --> 00:36:18.440 You said something like for a certain algorithm, working with a bunch of Python dictionaries, you could do like 12 million a second. 00:36:18.440 --> 00:36:18.880 Yes. 00:36:18.960 --> 00:36:22.980 And with lists, it was close to 10 times as much, 110 million a second. 00:36:22.980 --> 00:36:25.340 And with NumPy, you could do 500 million a second. 00:36:25.340 --> 00:36:25.840 Yes. 00:36:25.840 --> 00:36:35.600 And that's just a really simple like take a bunch of data that's in this form and just add them all together or like make pairs or something like that. 00:36:35.600 --> 00:36:36.020 Right, right. 00:36:36.020 --> 00:36:38.060 Like basically read through it or something to that effect. 00:36:38.060 --> 00:36:38.980 Exactly. 00:36:39.400 --> 00:36:39.640 Okay. 00:36:39.640 --> 00:36:49.900 And so you said, look, the way with the BSON capability notwithstanding, like basically the way that PyMongo works is you do a query, you get a bunch of dictionaries back in Python. 00:36:49.900 --> 00:36:57.000 But if you're doing data science or something computational, you probably want to work in, say, NumPy, right? 00:36:57.000 --> 00:37:00.600 But the workflow would be I make a query in my Python layer. 00:37:00.600 --> 00:37:01.400 It goes to Mongo. 00:37:01.400 --> 00:37:03.160 That comes back across the wire. 00:37:03.160 --> 00:37:08.800 It gets turned into Python dictionaries and then serialized back in down in the C layer into NumPy. 00:37:08.900 --> 00:37:10.020 And that was a problem, right? 00:37:10.020 --> 00:37:20.840 Yes, that's a huge problem because you're taking something that is pretty fast, namely MongoDB, and you have something on the other end of your line, which is also really fast, NumPy. 00:37:20.840 --> 00:37:24.360 And then you have this bottleneck, which is Python dictionaries. 00:37:24.360 --> 00:37:36.020 And it's kind of a shame that there hasn't been more stuff related to this until recently because MongoDB is an incredibly powerful database and it's very easy to use. 00:37:36.400 --> 00:37:40.980 And NumPy also has a selling point that it's extremely powerful and pretty easy to use. 00:37:40.980 --> 00:37:48.980 So you'd think that for a lot of data scientists or for people who don't love coding in MATLAB, they would want to put these two tools together. 00:37:48.980 --> 00:38:01.500 But MongoDB hasn't been used in this context super often because of the limitation that in order to get the data out, you have to put it through this kind of clunky data structure before getting it back into your super fast arrays. 00:38:01.500 --> 00:38:01.940 Right. 00:38:01.940 --> 00:38:12.220 So you said basically with some tests you did, like going through PyMongo, you could read about from MongoDB through Python into NumPy about 150,000 documents a second. 00:38:12.780 --> 00:38:24.840 And there's this other project called Monary that you were talking about when I saw this presentation where it basically says, let's stay down in the C layer the entire time. 00:38:24.840 --> 00:38:30.780 And even though we're calling it from a Python app, it sort of connects it directly to NumPy, right? 00:38:31.120 --> 00:38:39.620 Exactly. So now we are taking this raw BSON format and basically moving it directly into C style arrays. 00:38:39.620 --> 00:38:47.160 So you have two things that are in sort of natural machine order and you no longer have to take it out of that. 00:38:47.160 --> 00:38:50.540 Nice. And that's something like 10 times faster, right? At least it was. 00:38:51.000 --> 00:38:56.900 Yeah, it was 1.7 million reads per second compared to 150,000. 00:38:56.900 --> 00:38:58.380 That's a big difference. You might. 00:38:58.380 --> 00:39:07.460 That's like a difference between going and getting a cup of coffee while your algorithm runs on your data versus going to lunch or something. 00:39:07.460 --> 00:39:14.880 Yes, exactly. And so it's really exciting to me that we can leverage these two really awesome technologies. 00:39:14.880 --> 00:39:19.100 And Monary itself is actually not under active development anymore. 00:39:19.540 --> 00:39:22.360 Monary is a project that actually was community based. 00:39:22.360 --> 00:39:33.160 It was started by somebody who did not work at MongoDB, David Beach, who basically wrote it because he was sick of having to lose so much time to Python dictionaries. 00:39:33.160 --> 00:39:48.600 And so now we are writing a codec that is 100% in C that takes this raw BSON, which before, I think maybe a year or two ago, you couldn't actually get raw BSON from PyMongo, but now you can. 00:39:49.120 --> 00:39:58.960 So now you can take this raw BSON and you can have this super lightweight package that just converts it directly into NumPy and you are good to go. 00:39:58.960 --> 00:40:00.800 That's really cool. So what's this thing called now? 00:40:00.800 --> 00:40:05.220 So this is called BSON NumPy, which is a deeply uncreative name. 00:40:05.220 --> 00:40:09.020 And I am definitely open to alternatives, but it is descriptive. 00:40:10.080 --> 00:40:11.920 Basically, it is in beta. 00:40:11.920 --> 00:40:20.960 So maybe closer to alpha because we stopped working on it about two weeks ago and we released, we actually released about two weeks ago. 00:40:20.960 --> 00:40:29.360 So it's the project of both me and a Jesse Giryu Davis, who is my coworker who you spoke to twice, I think. 00:40:29.360 --> 00:40:30.920 Yeah, yeah. Jesse's a friend of the show. 00:40:30.920 --> 00:40:31.860 So hello, Jesse. 00:40:31.860 --> 00:40:33.680 And nice work on this project as well. 00:40:33.800 --> 00:40:36.520 We are both really eager to hear people's feedback. 00:40:36.520 --> 00:40:43.140 We really want MongoDB to become more useful for data scientists. 00:40:43.140 --> 00:40:46.680 And we think that this is really the tool that is going to make it happen. 00:40:46.920 --> 00:40:48.640 Yeah, it seems like a really great idea. 00:40:48.640 --> 00:40:54.080 Just we'll skip the serialization where it's slow and just flow the data straight into NumPy. 00:40:54.080 --> 00:41:08.700 This portion of Talk Python is brought to you by Hired. 00:41:08.700 --> 00:41:11.720 Hired is the platform for top Python developer jobs. 00:41:11.720 --> 00:41:16.520 Create your profile and instantly get access to 3,500 companies who will work to compete with you. 00:41:16.860 --> 00:41:19.380 Take it from one of Hired's users who recently got a job and said, 00:41:19.380 --> 00:41:24.680 I had my first offer on Thursday after going live on Monday and I ended up getting eight offers in total. 00:41:24.680 --> 00:41:28.140 I've worked with recruiters in the past, but they've always been pretty hit and miss. 00:41:28.140 --> 00:41:30.980 I tried LinkedIn, but I found Hired to be the best. 00:41:30.980 --> 00:41:33.080 I really like knowing the salary up front. 00:41:33.080 --> 00:41:35.440 Privacy was also a huge seller for me. 00:41:35.440 --> 00:41:37.120 Sounds awesome, doesn't it? 00:41:37.120 --> 00:41:39.140 Well, wait until you hear about the sign-in bonus. 00:41:39.140 --> 00:41:42.560 Everyone who accepts a job from Hired gets $1,000 sign-in bonus. 00:41:42.560 --> 00:41:45.220 And as Talk Python listeners, it gets way sweeter. 00:41:45.820 --> 00:41:50.440 Use the link Hired.com slash Talk Python To Me and Hired will double the signing bonus to $2,000. 00:41:50.440 --> 00:41:52.240 Opportunities knocking. 00:41:52.240 --> 00:41:56.000 Visit Hired.com slash Talk Python To Me and answer the door. 00:42:03.000 --> 00:42:05.640 I guess people can just go to MongoDB and find it. 00:42:05.640 --> 00:42:07.180 How do they learn more about this project? 00:42:07.180 --> 00:42:11.180 Because it's in its very, very first iteration. 00:42:11.180 --> 00:42:12.780 It's not... 00:42:12.780 --> 00:42:17.080 We released it on bsunnumpy.readthedocs.io. 00:42:17.080 --> 00:42:21.040 And you can find it on Read the Docs. 00:42:21.680 --> 00:42:22.860 Basically, we... 00:42:22.860 --> 00:42:29.700 Neither myself nor Jesse are data scientists, but we're both super familiar with Python and 00:42:29.700 --> 00:42:31.200 we're super familiar with MongoDB. 00:42:31.200 --> 00:42:37.060 But that does leave a lot of open questions that I wouldn't necessarily know how to answer. 00:42:37.060 --> 00:42:43.120 So I'm not sure what the most common data type for NumPy would be. 00:42:43.240 --> 00:42:48.660 I mean, there are all these really cool, complex data types, but for all I know, nobody ever 00:42:48.660 --> 00:42:49.200 uses them. 00:42:49.200 --> 00:42:53.620 So what I really want is to be able to reach out to the community and have people tell me 00:42:53.620 --> 00:42:54.260 what they need. 00:42:54.260 --> 00:42:59.100 Because a lot of the great features at MongoDB pretty much came out because somebody had a 00:42:59.100 --> 00:43:00.120 need and they asked for it. 00:43:00.120 --> 00:43:02.280 So we wrote them and then it became a huge selling point. 00:43:02.800 --> 00:43:07.800 So what I want is just to hear from what people who would use these technologies really want. 00:43:07.800 --> 00:43:08.400 Yeah, absolutely. 00:43:08.400 --> 00:43:09.920 So people can check it out. 00:43:09.920 --> 00:43:12.780 It's bsun-numpy on PyPI. 00:43:12.780 --> 00:43:14.940 And also I'll put the link in the show notes. 00:43:14.940 --> 00:43:20.600 So check it out and give both of you guys feedback on like, hey, this looks really cool, but it 00:43:20.600 --> 00:43:22.180 doesn't do whatever, right? 00:43:22.180 --> 00:43:23.060 Exactly. 00:43:23.060 --> 00:43:23.640 Nice. 00:43:23.640 --> 00:43:29.160 You also had in your presentation about Monary, it's still kind of the same type of question 00:43:29.160 --> 00:43:32.640 and answer type thing that you get the same kind of analysis you can do. 00:43:32.640 --> 00:43:39.360 with Bison-numpy, you had an interesting analysis you did of taxi cabs and Times Square, right? 00:43:39.360 --> 00:43:41.320 Do you want to tell people that story? 00:43:41.320 --> 00:43:41.880 Sure. 00:43:41.880 --> 00:43:49.140 So when I first joined MongoDB, I was working out of the headquarters, which is in Times Square 00:43:49.140 --> 00:43:50.020 in New York City. 00:43:50.020 --> 00:43:55.480 As a side note, right now I work out of Stockholm, Sweden, which I prefer quite a bit more to 00:43:55.480 --> 00:43:56.940 fighting my way through the crowds. 00:43:56.940 --> 00:44:02.480 But I was struck one morning when I was trying to think of sort of a nice data set to 00:44:02.480 --> 00:44:07.900 sort of show examples of how much faster Monary would be than Pymongo. 00:44:07.900 --> 00:44:14.800 I just was so sick of fighting through the crowds and fighting to get to the train station. 00:44:14.800 --> 00:44:16.420 And I kind of figured like, wow, what is... 00:44:16.420 --> 00:44:17.780 Why is there a Batman in my way? 00:44:17.780 --> 00:44:20.020 What's going on in this place? 00:44:20.020 --> 00:44:20.680 Exactly. 00:44:20.680 --> 00:44:23.900 It's like, how can I get out of here as fast as possible? 00:44:23.900 --> 00:44:25.220 Like, where can I go? 00:44:25.220 --> 00:44:26.140 How can I do it? 00:44:26.480 --> 00:44:33.660 So I also had access to the taxi data for all of New York City, which is freely available 00:44:33.660 --> 00:44:34.080 online. 00:44:34.080 --> 00:44:36.180 And it's a really interesting data set. 00:44:36.180 --> 00:44:43.180 And so I basically just took that data set and I looked at all the rides that both started 00:44:43.180 --> 00:44:44.280 and ended in Times Square. 00:44:44.280 --> 00:44:47.560 Because I kind of just wanted to know, like, where are people going? 00:44:47.560 --> 00:44:48.860 Like, what's the rush? 00:44:48.860 --> 00:44:50.200 Where are you coming from? 00:44:50.200 --> 00:44:51.440 Why are you coming here? 00:44:52.160 --> 00:44:54.620 Of all the places in New York City to be, why Times Square? 00:44:54.620 --> 00:44:55.060 Okay. 00:44:55.060 --> 00:44:55.320 Yeah. 00:44:55.320 --> 00:44:56.240 That's really cool. 00:44:56.240 --> 00:45:01.840 And you found, you had a really great bunch of visualizations that came out of MATLAB. 00:45:01.840 --> 00:45:06.160 And I'll put the link to that video up there because you have, you know, some really great 00:45:06.160 --> 00:45:10.240 maps and like bars living on flat maps and all kinds of stuff. 00:45:10.240 --> 00:45:10.960 I thought that was great. 00:45:10.960 --> 00:45:11.200 Yeah. 00:45:11.200 --> 00:45:12.520 It's like, it's MATPLOTLIB. 00:45:12.520 --> 00:45:13.040 Oh, yeah, yeah. 00:45:13.040 --> 00:45:13.300 Sorry. 00:45:13.300 --> 00:45:14.640 I keep, I don't know why I said MATLAB. 00:45:14.640 --> 00:45:16.920 MATLAB because we were talking about, yeah, yeah. 00:45:16.920 --> 00:45:18.320 MATPLOTLIB is what I had in mind. 00:45:18.320 --> 00:45:18.580 Yeah. 00:45:18.840 --> 00:45:23.800 But yeah, no, I mean, the point of those diagrams were just basically to show off how 00:45:23.800 --> 00:45:31.380 cool NumPy can be and how powerful MATPLOTLIB can be for creating really beautiful visualizations. 00:45:31.380 --> 00:45:35.400 And I know that there are a lot of different visualization tools out there. 00:45:35.400 --> 00:45:41.680 But in terms of selling Python and providing a way for data scientists to use both MongoDB 00:45:41.680 --> 00:45:44.940 and NumPy, I felt like it was a pretty good sell. 00:45:44.940 --> 00:45:45.620 Yeah, that's cool. 00:45:45.740 --> 00:45:50.060 And you also were able to leverage, I think the interesting linkage there is you were able 00:45:50.060 --> 00:45:57.760 to put a huge amount of data in Mongo and apply a geospatial index to it and then use that 00:45:57.760 --> 00:46:01.600 in your queries, but then also do the analysis with NumPy, right? 00:46:01.600 --> 00:46:02.240 Mm-hmm. 00:46:02.240 --> 00:46:02.880 Exactly. 00:46:03.060 --> 00:46:08.900 So you are taking advantage of these geo queries, which are really simple, really easy to use, 00:46:08.900 --> 00:46:10.860 but also very fast. 00:46:10.860 --> 00:46:17.240 And you are taking advantage of Python, NumPy specifically, where you have really, really 00:46:17.240 --> 00:46:20.560 fast analysis, but a lot of these algorithms are actually written for you. 00:46:20.560 --> 00:46:26.120 So you can use a lot of these scientific Python packages that have done what I consider the hard 00:46:26.120 --> 00:46:27.660 work for you completely. 00:46:27.660 --> 00:46:28.180 Yeah. 00:46:28.180 --> 00:46:28.660 Yeah. 00:46:28.660 --> 00:46:32.540 It seems like just take the tools, click them together, and you get some really great 00:46:32.540 --> 00:46:33.140 analysis. 00:46:33.140 --> 00:46:33.740 That's awesome. 00:46:33.740 --> 00:46:34.620 Yeah. 00:46:34.620 --> 00:46:40.180 So I'm looking forward to your presentation where you talk about your Bison NumPy version. 00:46:40.180 --> 00:46:41.480 Yeah. 00:46:41.480 --> 00:46:47.260 Well, I'll have to find something better, some better visualizations and some, you know, I 00:46:47.260 --> 00:46:49.720 don't have as many things to complain about now that I live in Stockholm. 00:46:49.720 --> 00:46:51.540 So I have to... 00:46:51.540 --> 00:46:51.940 Yeah. 00:46:51.940 --> 00:46:54.240 You can do something with snow or winter or something. 00:46:54.240 --> 00:46:54.660 Who knows? 00:46:54.660 --> 00:46:55.220 Yeah. 00:46:55.220 --> 00:46:55.860 The darkness. 00:46:55.860 --> 00:46:56.840 Yes, exactly. 00:46:56.840 --> 00:46:58.120 Here's the light analysis. 00:46:58.120 --> 00:46:59.300 All right. 00:46:59.300 --> 00:46:59.600 Cool. 00:46:59.600 --> 00:47:01.260 So I guess let's go ahead and leave it there. 00:47:01.260 --> 00:47:02.600 That sounds like a great project. 00:47:02.600 --> 00:47:07.420 And so if you're out there working with MongoDB and ultimately your data ends up in NumPy, 00:47:07.420 --> 00:47:10.260 you know, maybe skip the dictionaries is the message, right? 00:47:10.260 --> 00:47:10.880 If you can. 00:47:10.880 --> 00:47:11.800 Yeah, exactly. 00:47:11.800 --> 00:47:17.420 And especially for people who use scientific Python and have never considered MongoDB a viable 00:47:17.420 --> 00:47:19.760 option, I think that is going to change. 00:47:19.860 --> 00:47:25.280 And I think that we are going to be able to write programs that take advantage of all this 00:47:25.280 --> 00:47:27.500 cool work that has already been done. 00:47:27.500 --> 00:47:27.820 Yeah. 00:47:27.820 --> 00:47:33.480 And I think having it officially part of the product, the library, gives it a little more, 00:47:33.480 --> 00:47:35.860 probably gives people more confidence to build upon it. 00:47:35.860 --> 00:47:36.520 Yeah, exactly. 00:47:36.520 --> 00:47:41.600 And I am personally just so excited because this is the first project that I've really had 00:47:41.600 --> 00:47:42.680 ownership over. 00:47:42.680 --> 00:47:45.880 I contributed a lot to PyMongo, but it wasn't my project. 00:47:45.880 --> 00:47:50.440 Well, I feel very proud of what I've done for this particular project. 00:47:50.440 --> 00:47:51.680 It sounds really, really cool. 00:47:51.680 --> 00:47:52.780 So, so nice work. 00:47:52.780 --> 00:47:57.020 Let me ask you a couple of questions before we get out of here, as I always do. 00:47:57.020 --> 00:48:01.820 First one, if you're going to write some Python code, maybe not C code, if you write some Python 00:48:01.820 --> 00:48:03.380 code, what editor do you open up? 00:48:03.520 --> 00:48:04.720 So I use PyCharm. 00:48:04.720 --> 00:48:13.160 And it's funny that you mentioned C code because I got a perpetual license for JetBrains. 00:48:13.160 --> 00:48:20.740 And I love that I can write, I can write Node, I can write Ruby, I can write Python, I can write C. 00:48:20.740 --> 00:48:22.660 And all my shortcuts are the same. 00:48:22.660 --> 00:48:27.800 I was a Vim user before Jesse convinced me to switch over to PyCharm. 00:48:27.800 --> 00:48:30.040 And now I'm like completely converted. 00:48:30.040 --> 00:48:34.200 I can have all my Vim shortcuts and I can also have all the power of a really great IDE. 00:48:34.200 --> 00:48:39.920 So I really recommend the JetBrains libraries because it's really nice to not have to switch 00:48:39.920 --> 00:48:45.960 between IDEs, between like SeaLion, PyCharm, Ruby9, WebStorm. 00:48:45.960 --> 00:48:47.720 You know, it's the same environment. 00:48:47.720 --> 00:48:50.580 So I feel like I don't have to lose any time to relearning stuff. 00:48:50.720 --> 00:48:51.840 Yeah, that's really cool. 00:48:51.840 --> 00:48:54.500 Jesse's a big fan of PyCharm and so am I. 00:48:54.500 --> 00:48:59.940 Just so people who are not familiar, basically there's this IntelliJ platform, which is kind 00:48:59.940 --> 00:49:02.420 of the IDE and then they plug in the language specific stuff. 00:49:02.420 --> 00:49:05.080 So like when you say it's all the same, it's like all kind of the same base. 00:49:05.080 --> 00:49:07.200 And like you said, like it behaves the same. 00:49:07.200 --> 00:49:08.160 It's pretty sweet. 00:49:08.160 --> 00:49:08.580 Yeah. 00:49:08.580 --> 00:49:09.220 Okay, cool. 00:49:09.220 --> 00:49:14.520 And we just this week passed 100,000 packages on PyPI. 00:49:14.520 --> 00:49:16.620 So that's a big milestone. 00:49:16.620 --> 00:49:17.180 Hooray. 00:49:17.180 --> 00:49:18.360 Wow. 00:49:18.360 --> 00:49:19.600 Yeah, no kidding. 00:49:19.600 --> 00:49:21.700 And so there's a ton of them out there. 00:49:21.700 --> 00:49:24.840 Do you have one that maybe like you think people don't necessarily know about? 00:49:24.840 --> 00:49:26.100 You're like, hey, you should check this out. 00:49:26.100 --> 00:49:28.260 BSON NumPy maybe? 00:49:28.260 --> 00:49:29.180 Yes. 00:49:29.180 --> 00:49:34.900 Is it too shamelessly self-promoting to say that you should download BSON NumPy and that's 00:49:34.900 --> 00:49:36.340 the best package on PyPI? 00:49:36.340 --> 00:49:38.280 It's one of the newer ones. 00:49:38.280 --> 00:49:39.120 That's a really good question. 00:49:39.120 --> 00:49:43.940 I guess I'll just go with that because I'm very utilitarian with my packages. 00:49:44.480 --> 00:49:47.300 I pretty much just keep them to a minimum. 00:49:47.300 --> 00:49:47.920 Okay. 00:49:47.920 --> 00:49:48.420 Yeah. 00:49:48.420 --> 00:49:48.880 Very cool. 00:49:48.880 --> 00:49:49.960 All right. 00:49:49.960 --> 00:49:50.120 Yeah. 00:49:50.120 --> 00:49:50.840 It's a great package. 00:49:50.840 --> 00:49:54.200 And you know, you guys need some people to try it out and actually do real data science 00:49:54.200 --> 00:49:54.500 with it. 00:49:54.500 --> 00:49:55.620 So that's great. 00:49:55.620 --> 00:49:56.500 Exactly. 00:49:56.500 --> 00:49:57.260 All right. 00:49:57.300 --> 00:49:58.300 So final call to action. 00:49:58.300 --> 00:50:00.860 What do you want people to check out or do? 00:50:00.860 --> 00:50:05.440 I would like people to check out BSON NumPy because I think it's a really cool project. 00:50:05.440 --> 00:50:12.160 I think that if anyone is interested in a really neat piece of code, they should look through 00:50:12.160 --> 00:50:17.680 the LLVM Py work, especially if you're either new to compilers, maybe new to Python, new to 00:50:17.680 --> 00:50:18.360 LLVM IR. 00:50:18.360 --> 00:50:24.500 It's a very nice sort of pet project and it's very self-contained and it is, in my opinion, 00:50:24.500 --> 00:50:25.480 pretty well documented. 00:50:25.480 --> 00:50:28.520 There's an entire 20-page thesis I wrote about it. 00:50:28.520 --> 00:50:30.820 So if you have questions, they're probably answered. 00:50:30.820 --> 00:50:34.760 I'm also always happy to answer by like email, Twitter, anything. 00:50:34.760 --> 00:50:38.080 So those are my two things I would love to see people do. 00:50:38.080 --> 00:50:38.600 All right. 00:50:38.600 --> 00:50:39.640 That sounds great. 00:50:39.640 --> 00:50:41.420 Ana, thank you so much for being on the show. 00:50:41.420 --> 00:50:42.840 It's been really interesting to talk to you. 00:50:42.840 --> 00:50:43.380 Yeah, of course. 00:50:43.380 --> 00:50:44.900 Thank you so much again for having me. 00:50:44.900 --> 00:50:45.140 Yeah. 00:50:45.140 --> 00:50:45.640 Thanks so much. 00:50:45.640 --> 00:50:45.960 Bye. 00:50:45.960 --> 00:50:46.280 Bye. 00:50:46.280 --> 00:50:50.760 This has been another episode of Talk Python To Me. 00:50:50.760 --> 00:50:56.880 Today's guest has been Ana Herlihy and this episode has been sponsored by Hired and Talk 00:50:56.880 --> 00:50:57.760 Python Training. 00:50:57.760 --> 00:51:00.640 Are you or a colleague trying to learn Python? 00:51:00.640 --> 00:51:05.320 Have you tried books and videos that just left you bored by covering topics point by point? 00:51:05.320 --> 00:51:10.640 Well, check out my online course, Python Jumpstart by Building 10 Apps at talkpython.fm 00:51:10.680 --> 00:51:13.940 slash course to experience a more engaging way to learn Python. 00:51:13.940 --> 00:51:18.720 And if you're looking for something a little more advanced, try my Write Pythonic code course 00:51:18.720 --> 00:51:21.280 at talkpython.fm/pythonic. 00:51:21.280 --> 00:51:24.640 Hired wants to help you find your next big thing. 00:51:24.640 --> 00:51:29.380 Visit Hired.com slash Talk Python To Me to get five or more offers with salary and equity 00:51:29.380 --> 00:51:33.140 presented right up front and a special listener signing bonus of $2,000. 00:51:35.020 --> 00:51:36.720 Be sure to subscribe to the show. 00:51:36.720 --> 00:51:38.920 Open your favorite podcatcher and search for Python. 00:51:38.920 --> 00:51:40.160 We should be right at the top. 00:51:40.160 --> 00:51:46.260 You can also find the iTunes feed at /itunes, Google Play feed at /play and direct 00:51:46.260 --> 00:51:49.440 RSS feed at /rss on talkpython.fm. 00:51:49.860 --> 00:51:54.560 Our theme music is Developers, Developers, Developers by Corey Smith, who goes by Smix. 00:51:54.560 --> 00:51:59.240 Corey just recently started selling his tracks on iTunes, so I recommend you check it out at 00:51:59.240 --> 00:52:01.240 talkpython.fm/music. 00:52:01.240 --> 00:52:06.360 You can browse his tracks he has for sale on iTunes and listen to the full length version of the theme 00:52:06.360 --> 00:52:06.600 song. 00:52:06.600 --> 00:52:08.680 This is your host, Michael Kennedy. 00:52:08.680 --> 00:52:09.960 Thanks so much for listening. 00:52:09.960 --> 00:52:11.160 I really appreciate it. 00:52:11.160 --> 00:52:13.300 Smix, let's get out of here. 00:52:13.300 --> 00:52:34.740 I'll see you next time. 00:52:34.740 --> 00:52:35.580 I'll put it in. 00:52:35.580 --> 00:53:05.560 Thank you.