Transcrição completa de PsychoPy with Jonathan Peirce

but but they were wondering whether or not they would ever support Macs or or or see support for their Macs. 23 00:02:18.070 --> 00:02:30.585 And I started hearing about Python as this other alternative that was free and and, you know, the cool kids were were having a play. But these were really early days for Python. Right? These were there was no matplotlib. 24 00:02:31.045 --> 00:02:32.105 There was no pandas. 25 00:02:32.805 --> 00:02:33.305 NumPy 26 00:02:33.685 --> 00:02:37.140 didn't exist. So there was a a debate between 27 00:02:37.440 --> 00:02:39.060 whether we should use num array 28 00:02:39.440 --> 00:02:40.420 or numeric. 29 00:02:41.040 --> 00:02:44.415 And and and later on, you know, the NumPy 30 00:02:44.795 --> 00:02:54.700 sort of came out of the the the burning embers of those 2. So it was really early days for Python. But what it had that that was really useful to me was it had OpenGL bindings. 31 00:02:55.480 --> 00:02:59.019 And so I could write my OpenGL code 32 00:02:59.535 --> 00:03:04.995 and get hardware accelerated graphics right there in a Python script. Whereas in MATLAB, I had to 33 00:03:05.535 --> 00:03:08.330 write MEX files, which is like a compiled 34 00:03:08.710 --> 00:03:11.050 bridge from from MATLAB to c. 35 00:03:11.830 --> 00:03:14.165 A bit like, I I guess the, 36 00:03:14.805 --> 00:03:16.344 I can't remember the name. The, 37 00:03:18.245 --> 00:03:21.704 a bit like, no. It wasn't c types. Anyway, compiled files 38 00:03:22.030 --> 00:03:27.090 that that you had to write for MATLAB, and I I didn't want to write those. MATLAB actually improved, 39 00:03:27.870 --> 00:03:30.290 in in many ways, and it did support Apple's, 40 00:03:31.465 --> 00:03:32.345 Apple Macs, and, 41 00:03:33.145 --> 00:03:40.125 they did get wrappers for various different things. But by the time MATLAB was getting better, I was already in love with the Python syntax, 42 00:03:40.570 --> 00:03:45.210 and I moved on. So so, yeah, it was early days, basically, and and, 43 00:03:46.090 --> 00:03:50.265 and I just kind of fell in love with the syntax of Python and decided this was a way to go. 44 00:03:51.705 --> 00:03:56.605 And can you start off by telling us what PsychoPy is and how the project got started? 45 00:03:58.400 --> 00:03:59.140 So in 46 00:04:00.000 --> 00:04:00.500 2003, 47 00:04:00.959 --> 00:04:10.375 I then moved to the UK and and got my own faculty position, and I was starting up my own lab. And PsychoPy then began so it started off as a proof of concept 48 00:04:10.675 --> 00:04:15.175 where I was trying to show other people, look. This is what I could do in trying to get other people to write it, to be honest. 49 00:04:15.480 --> 00:04:15.980 But 50 00:04:16.440 --> 00:04:21.180 when I moved to the UK, I started my own lab, and and this became the library 51 00:04:21.800 --> 00:04:26.795 that that my lab used for for Python script. So we would write code in other editors, 52 00:04:27.335 --> 00:04:29.275 and

then we would use the the PsychoPy 53 00:04:29.655 --> 00:04:30.155 library 54 00:04:30.695 --> 00:04:34.235 in order to, you know, provide some convenience functions for 55 00:04:34.600 --> 00:04:36.460 saving data, checking keyboards, 56 00:04:36.840 --> 00:04:38.540 presenting stimuli in a window. 57 00:04:39.080 --> 00:04:39.580 And 58 00:04:40.040 --> 00:04:46.045 gradually, that changed. So I put it online. It was on SourceForge as of 2, 002. But 59 00:04:46.825 --> 00:04:49.885 gradually, people were sort of sending me messages saying, 60 00:04:50.665 --> 00:04:53.965 hi. I saw your thing. I downloaded it. It was cool, but 61 00:04:54.320 --> 00:05:08.725 I don't understand. Where is it? I installed it, and it's not in my start menu. Right? Because it was a library, and people didn't know what that meant exactly. And and so people were struggling also to get all the dependencies from Python and that sort of thing. So I turned it into 62 00:05:09.345 --> 00:05:12.485 an app, which was basically just a code editor 63 00:05:13.010 --> 00:05:14.230 with all of its 64 00:05:14.930 --> 00:05:21.345 sort of dependencies built in, including its own copy of Python. So people install an app, they could write their PsychoPy 65 00:05:21.805 --> 00:05:35.650 scripts, which use the library, and get all of the dependencies built in. And then it was in their start menu, and they got it. Right? So so gradually, people started using that in the sort of 2, 005, 2, 006. We had a couple hundred users maybe, Python enthusiasts. 66 00:05:36.205 --> 00:05:42.145 And then the thing that started happening was that I was getting frustrated because I was teaching 67 00:05:42.765 --> 00:05:47.070 to undergraduates who didn't know how to program and wouldn't want to learn Python. 68 00:05:47.690 --> 00:05:52.750 I was teaching them how to use other software packages that use, like, a graphical interface, 69 00:05:53.405 --> 00:06:05.889 And and I started getting frustrated by that, and the students were frustrated by that too. So around 2008, 2009, I decided I was gonna write a graphical interface on top of this library that would 70 00:06:06.270 --> 00:06:08.930 write your script for you. You would build your experiment 71 00:06:09.470 --> 00:06:09.970 graphically, 72 00:06:10.425 --> 00:06:17.005 or you would write the script yourself if you wanted to. But but you could you could construct it graphically in the builder, and it would compile 73 00:06:17.465 --> 00:06:18.205 the Python 74 00:06:18.585 --> 00:06:20.285 syntax script for you 75 00:06:20.780 --> 00:06:28.800 from what your from your visual representation of your experiment. And and that was the thing that really sort of changed the numbers of PsychoPy, 76 00:06:29.645 --> 00:06:30.465 quite drastically. 77 00:06:30.765 --> 00:06:35.025 That enabled it to be used by an awful lot of people who who didn't want to learn a new language. 78 00:06:36.045 --> 00:06:38.465 And can the output of that builder, 79 00:06:39.099 --> 00:06:41.599 interface then be modified further? 80 00:06:42.539 --> 00:06:51.985 So you can compile it to a you can either just press run, or you can compile it to a script and hack the script, But, also, you can insert into it custom code 81 00:06:52.685 --> 00:07:00.490 of your own that can be executed at various different either on every screen refresh or just at the beginning of experiment, or wherever you like. And 82 00:07:01.110 --> 00:07:06.090 so you can customize it as well, which once you've compiled the script and hacked the script, 83 00:07:06.414 --> 00:07:10.435 you can't go back to the graphical interface from that. Right? So so, 84 00:07:11.134 --> 00:07:11.794 the way 85 00:07:12.335 --> 00:07:14.595 of providing something that you can 86 00:07:14.920 --> 00:07:17.180 edit code within the graphical interface 87 00:07:17.560 --> 00:07:22.940 was was that you got these inline code components as well. And all of this was possible because 88 00:07:23.825 --> 00:07:28.565 Python does great text handling. Right? So we could build a WX 89 00:07:29.425 --> 00:07:31.925 widgets application that would run on any 90 00:07:41.639 --> 00:07:54.789 and and it could just create the appropriate text because of because of the great text handling of Python. So that part was well, it took a lot of thinking, I guess. But but now when I look back at it, it all kind of just hangs together. It makes sense. 91 00:07:55.569 --> 00:07:59.590 So how does PsychoPy compare feature wise against some of the proprietary alternatives? 92 00:08:01.014 --> 00:08:04.955 So I think the key there is that Python and interpreted languages, 93 00:08:05.895 --> 00:08:09.115 and the open source world have meant that nowadays, 94 00:08:10.190 --> 00:08:12.530 software can be written by the people who use the software. 95 00:08:12.910 --> 00:08:13.810 Right? So 96 00:08:14.190 --> 00:08:15.730 I'm not a software engineer, 97 00:08:16.110 --> 00:08:16.610 and 98 00:08:17.285 --> 00:08:17.945 I know 99 00:08:18.805 --> 00:08:19.625 how an experiment 100 00:08:20.245 --> 00:08:25.040 should run. I know how a scientist will think of their experiment. I know even how an undergraduate, 101 00:08:25.900 --> 00:08:28.400 a very junior scientist, will think about their experiment. 102 00:08:28.780 --> 00:08:35.695 So I think in terms of ease of use, that's that's where we win. That's where that's where the open source community, 103 00:08:36.075 --> 00:08:36.575 where 104 00:08:37.035 --> 00:08:37.535 enthusiasts 105 00:08:37.835 --> 00:08:43.760 and users develop the software themselves, rather than the traditional model of trying communicate that to 106 00:08:44.140 --> 00:08:49.839 a professional software engineer who's never personally run an experiment and doesn't know what that means. Right? So 107 00:08:50.540 --> 00:08:51.680 so I can 108 00:08:52.565 --> 00:09:03.740 actually build the thing the way that I want it to run and the way that I think about

my experiment. And so I think that's that's where we win in terms of features. And in most other things, in terms of sort of high end features, 109 00:09:04.200 --> 00:09:12.705 yeah, we can do we can do a lot of the fancy stuff as well. Again, because of the nature of open source software and the fact that you've got lots of enthusiasts 110 00:09:13.245 --> 00:09:15.425 who are contributing, we've we've now got 111 00:09:15.860 --> 00:09:19.640 50, 60 people who've added some level of code, 112 00:09:20.100 --> 00:09:23.640 3 or 4 of us who've contributed a lot. And 113 00:09:24.254 --> 00:09:28.754 those are various different enthusiasts who will add some high end feature of their own. So, 114 00:09:29.454 --> 00:09:38.540 and and whereas commercial software can't just sort of, like, suddenly add this new thing really quickly and and have it out there. So in terms of features, we do really well, I think. 115 00:09:38.895 --> 00:09:40.115 And in terms of usability, 116 00:09:40.575 --> 00:09:48.640 that's that's where I think we're doing extremely well, to be honest. Where we don't do so well is is in bugs and, 117 00:09:49.020 --> 00:09:51.840 you know, reacting to that sort of thing. 118 00:09:52.220 --> 00:09:57.585 We're all busy, and and so and we're not professional sci computer scientists. 119 00:09:58.045 --> 00:10:12.790 So sometimes we've we've added features, but we've not been fixing the bugs for which we've got workarounds. Right? So I know I know that there's this annoying little glitch, but I know how to avoid it, and so I just do. And and so sometimes 120 00:10:13.165 --> 00:10:16.465 that's the downside of these sorts of projects versus a commercial enterprise. 121 00:10:16.925 --> 00:10:22.865 We understand the users better, but also we don't have as much time to fix bugs, and we don't spend 122 00:10:23.199 --> 00:10:25.300 hours and hours testing things. 123 00:10:25.680 --> 00:10:28.660 So that's I guess those are the those are the 2 main differences. 124 00:10:30.079 --> 00:10:43.330 In the documentation, you mentioned that the that your project is useful for the fields of psychophysics, cognitive neuroscience, and experimental psychology. And I'm wondering if you can provide some background into how those disciplines differ and, 125 00:10:43.710 --> 00:10:46.770 what constitutes an experiment for each of those fields respectively. 126 00:10:47.815 --> 00:10:48.535 Sure. Sure. 127 00:10:49.095 --> 00:10:49.595 So 128 00:10:49.975 --> 00:10:53.435 my own work is in vision science, and that's the psychophysics 129 00:10:54.214 --> 00:10:54.855 is this, 130 00:10:55.415 --> 00:10:57.760 this area of psychology where 131 00:10:58.140 --> 00:11:02.080 we treat the brain as basically it's like an engineering a physics project, 132 00:11:02.460 --> 00:11:13.615 and we we map its inputs and its outputs and and that sort of thing. It's a very dull area of psychology to many people, but that's what I do. And that's that's where PsychoPy has its origins was was presenting 133 00:11:14.170 --> 00:11:16.990 very precisely low level stimuli, visual 134 00:11:17.290 --> 00:11:18.510 things in my case. 135 00:11:18.890 --> 00:11:19.390 So 136 00:11:19.850 --> 00:11:33.850 measuring your ability to detect something that's barely visible, for instance, that sort of experiment. Then you've got lots of so cognitive psychology would have an experiment where, say, you would test someone's memory to a set of objects, and we'd measure 137 00:11:34.230 --> 00:11:36.330 what sort of things impair your memory 138 00:11:36.790 --> 00:11:39.770 and what things make you, I don't know, respond faster. 139 00:11:40.214 --> 00:11:40.714 So 140 00:11:42.055 --> 00:11:49.035 something and and linguistics blends into this as well. So something like the Stroop task is a task that's very, 141 00:11:49.830 --> 00:11:52.490 well known in this sort of field where you present 142 00:11:53.510 --> 00:11:54.730 the word red 143 00:11:55.270 --> 00:11:56.970 written in green text 144 00:11:58.024 --> 00:12:00.204 versus the word red written in red text. 145 00:12:00.665 --> 00:12:02.925 Sub subjects participants have to report 146 00:12:03.384 --> 00:12:04.764 the color of the letters, 147 00:12:05.390 --> 00:12:21.305 but they're slower to do that if the letters are spelling out the wrong word. Okay? So you're slower to respond that you've got red letters in front of you if the if the red letters say the word green. So that would be a sort of cognitive psychology or linguistics type of study. 148 00:12:21.685 --> 00:12:22.185 And, 149 00:12:22.780 --> 00:12:28.400 you know, again, it revolves around presenter stimulus, measure response, measure your reaction time to that stimulus. 150 00:12:29.180 --> 00:12:29.680 Neuroscience 151 00:12:30.464 --> 00:12:34.324 might be similar sorts of experiments, but now you're doing that in an MR scanner. 152 00:12:34.865 --> 00:12:36.805 So you're measuring the the fMRI, 153 00:12:37.860 --> 00:12:43.720 BOLD signal of of a participant while they see different stimuli. So there are various different sort of 154 00:12:44.635 --> 00:12:45.775 fields around, 155 00:12:46.315 --> 00:12:48.815 what what overall we'd call the behavioral sciences, 156 00:12:49.435 --> 00:12:53.375 and they all tend to link to the same sort of experiment 157 00:12:53.829 --> 00:13:05.935 where you present stimulus and you get a response. There's also social psychology where maybe you present images, and then you'd ask people to rate them, or you'd ask people's opinions in some sort of a way. And and, again, you can use 158 00:13:06.395 --> 00:13:07.055 the same 159 00:13:07.595 --> 00:13:08.095 software 160 00:13:08.395 --> 00:13:11.215 to to run all of those different sorts of studies. 161 00:13:12.399 --> 00:13:19.860 So do you find that your users who have no previous formal programming training come up to speed with PsychoPy quickly? And what are some of the challenges there? 162

00:13:20.825 --> 00:13:27.245 Yeah. For some of our users, that's that's a real challenge. For some, those in more neuroscience based areas, 163 00:13:28.000 --> 00:13:40.255 they are probably fairly used to programming because they'll need to do it for their analysis anyway, and they've been used to other things like MATLAB and and that sort of stuff. So so we've got a huge spread. People in psychology and people in linguistics 164 00:13:41.275 --> 00:13:44.415 are typically less likely to be programmers beforehand. 165 00:13:44.875 --> 00:13:45.375 And 166 00:13:45.860 --> 00:13:47.400 so we've got this quite broad 167 00:13:47.780 --> 00:13:52.200 church of people from from all different backgrounds, and that's where the graphical interface 168 00:13:52.820 --> 00:13:57.595 or the scripting interface really helps. So so if you prefer to code and have absolutely 169 00:13:58.135 --> 00:14:09.510 every every bit of control over every aspect of your experiment, then you should write your script yourself. Because the builder is gonna be trying to do things generally and can't implement every feature that you can write in code. 170 00:14:10.050 --> 00:14:18.605 But but then for those less technical users, they seem to be finding the builder interface very easy to use. And I think 1 of the nice things actually is that, 171 00:14:19.225 --> 00:14:20.204 along the way, 172 00:14:20.829 --> 00:14:28.769 they learn a little bit of the logic of programming as well. And and maybe they didn't intend to move, they didn't realize they're even doing that. But the builder 173 00:14:29.565 --> 00:14:33.264 is kind of set up. You know, it has objects and loops, and 174 00:14:33.565 --> 00:14:47.490 objects have parameters, and those can be repeated and and changed at various different times. You know, they're essentially learning the basics of programming, but they're doing it in a graphical interface rather than line by line of code. They're also as they start to make those experiments 175 00:14:48.415 --> 00:14:49.235 more advanced, 176 00:14:49.695 --> 00:14:58.750 they'll want to add in little snippets of code. And it might be the odd line or 2 that just, I don't know, makes the next stimulus depend on the previous response. 177 00:14:59.290 --> 00:15:00.910 And that's something that requires 178 00:15:01.690 --> 00:15:04.269 a line or 2 of Python code. So 179 00:15:04.665 --> 00:15:09.165 they're comfortable with adding a line or 2, and they don't consider that programming. There's nothing to 180 00:15:09.785 --> 00:15:14.820 be frightened off. But, you know, I think with a lot of people, if you tell them that they're going to learn programming, 181 00:15:15.280 --> 00:15:22.745 that's that's just too terrifying a task. Or if you tell them they're going to write a script, you know, it's it's too big a thing for them to imagine. Whereas if they 182 00:15:23.125 --> 00:15:29.330 start off just writing the odd line or 2 here and there and and learning to hack

a little bit, something that's already done, 183 00:15:29.950 --> 00:15:36.529 then they they get sort of eased in. I think we do both things. With the builder, we manage to support people 184 00:15:36.925 --> 00:15:40.545 who don't really program and who possibly never will. But also, 185 00:15:41.005 --> 00:15:45.920 we we sort of ease people in a little bit to to programming as well. 186 00:15:46.320 --> 00:15:48.500 I guess also, I mean, I now run 187 00:15:48.880 --> 00:15:54.725 quite a lot of workshops and and that sort of thing, and we spend a lot of time training people. I've taught people 188 00:15:55.025 --> 00:15:58.325 from right away from undergraduates, PhD students, and professors 189 00:15:58.705 --> 00:16:02.165 how to program in Python on to some level. And 190 00:16:02.750 --> 00:16:03.570 and I guess 191 00:16:04.190 --> 00:16:06.290 it comes back again to that thing that 192 00:16:06.750 --> 00:16:08.290 I'm not a hardcore 193 00:16:09.070 --> 00:16:11.250 computer science trained programmer. 194 00:16:11.595 --> 00:16:14.095 You know, my background, I did a degree in psychology. 195 00:16:14.555 --> 00:16:23.420 And and I can empathize a bit more with a novice programmer because I remember what it was like to be that. And and maybe it's a bit harder for a a more sophisticated 196 00:16:24.040 --> 00:16:32.865 programmer to be training people who who don't have that sort of a background. You know? They they can't imagine what it's like not to understand the difference between variable 197 00:16:33.404 --> 00:16:36.920 and a string and and and that sort of thing. So maybe 198 00:16:37.460 --> 00:16:39.080 it's my lack 199 00:16:39.620 --> 00:16:43.960 of computer science background that means we can we can do this a little bit better. 200 00:16:45.365 --> 00:16:53.465 That's great. I think it it it really speaks to 1 of Python's core strengths in that. It really seems like for whatever reason, whatever its attributes, 201 00:16:54.139 --> 00:16:55.440 the language really 202 00:16:55.820 --> 00:16:57.839 allows itself to be picked up 203 00:16:58.220 --> 00:17:02.079 comparatively easily, at least compared to other programming languages by folks 204 00:17:02.524 --> 00:17:08.145 whose, you know, primary occupation or or focus is not programming. So that's that's fantastic. 205 00:17:08.845 --> 00:17:12.649 My experience is that, you know, the very first the the starting point 206 00:17:12.950 --> 00:17:20.090 for quite a few users. It's easier when with an absolute beginner to teach them MATLAB for the first few 207 00:17:20.815 --> 00:17:21.315 hours, 208 00:17:21.855 --> 00:17:24.355 where they don't have to worry about 209 00:17:25.375 --> 00:17:35.310 things like was it an integer or a float. Right? They they it just does the right thing, and they don't have to think about different types. They don't have to import their libraries and that sort of thing. 210 00:17:35.635 --> 00:17:39.735 But very quickly, I think you're

right. That that after that first couple of hours, 211 00:17:40.115 --> 00:17:41.414 they get it. And then 212 00:17:42.035 --> 00:17:50.000 if you then try and show them about MATLAB, they get very frustrated that it can't do all of the things that they found so easy in Python. So, 213 00:17:50.300 --> 00:17:51.440 yeah, I I've 214 00:17:51.845 --> 00:17:53.705 I've had a great experience with 215 00:17:54.325 --> 00:17:55.304 teaching people 216 00:17:55.765 --> 00:17:56.265 Python. 217 00:17:56.645 --> 00:17:58.505 And if they've not learned a language before, 218 00:17:58.885 --> 00:18:01.625 they they get into Python usually fairly quickly. 219 00:18:02.789 --> 00:18:07.210 Yeah. And I think that your comment about having the ability to 220 00:18:07.590 --> 00:18:09.850 work against a known good 221 00:18:10.245 --> 00:18:32.675 piece of software and then just add some modifications here and there is definitely a great way to introduce people to some of the different concepts because they don't have to start off staring at a blank page and say, okay, well, where do I begin? Because it's already begun. They just need to continue it along, and then that will provide greater confidence when they do go back and have to start it with a blank slate. 222 00:18:33.160 --> 00:18:36.540 So the other thing that we do, all that this this within the application, 223 00:18:37.320 --> 00:18:41.900 as I say, I sort of provide this this code editor, you know, and it's not as 224 00:18:42.465 --> 00:18:42.965 sophisticated 225 00:18:43.504 --> 00:18:44.485 as, say, 226 00:18:44.865 --> 00:18:49.205 Spyder or, you know, lots of the other more professional dedicated editors. 227 00:18:50.070 --> 00:18:52.970 But it has a a menu that says demos. 228 00:18:53.429 --> 00:19:01.394 And it has a you know, they so users new to programming can look at the demos menu and and see a whole range of relatively small digestible 229 00:19:02.014 --> 00:19:04.674 demos that show 1 aspect of the software, 230 00:19:05.375 --> 00:19:09.290 in a in a unit. And there's a big green running man that says run. You know? 231 00:19:10.330 --> 00:19:10.830 So 232 00:19:11.530 --> 00:19:12.030 although 233 00:19:12.490 --> 00:19:14.750 a lot of users then will want to go on or 234 00:19:15.050 --> 00:19:18.350 or pre more experienced users from other programming language languages 235 00:19:18.735 --> 00:19:30.230 will want a more feature rich editor. To a new user, they get something that's very simple with a green running man, and it just works. That's that's the aim. So so it provides them again with a very simple 236 00:19:30.850 --> 00:19:43.485 inroad. And lots of other users, I'm sure, are just gonna use PsychoPy. Those sophisticated programmers will just use PsychoPy as a library, and they'll use their own Python installation and their own editors, and and and that's great too. 237 00:19:44.340 --> 00:19:49.000 Yeah. And I'm sure too that by having the demos, which are going

to tie into 238 00:19:49.460 --> 00:19:56.545 a discipline that they are familiar with, will make it easier for them to then understand what's actually happening in the Python code itself. 239 00:19:57.005 --> 00:20:01.505 Yeah. Yeah. So we've we've got a range of sort of from very small unitary 240 00:20:02.360 --> 00:20:06.700 things like how do I present a sound to sort of full fledged experiments. 241 00:20:07.480 --> 00:20:10.620 We've we've got quite a range of of levels there. 242 00:20:11.895 --> 00:20:18.075 And can you describe the internal architecture of PsychoPy and how you approach the design and how it's evolved over time? 243 00:20:19.560 --> 00:20:20.060 Yeah. 244 00:20:21.480 --> 00:20:24.780 So evolved over time is pretty much the key phrase there. 245 00:20:26.120 --> 00:20:35.245 Design is, there have been moments where we've really designed the architecture, but at the core, it's this library. That's that's the basic thing. So it's a library which has 246 00:20:35.705 --> 00:20:37.785 stimuli that might be visual or might be, 247 00:20:38.570 --> 00:20:44.990 there might be classes for handling your trials and saving your data and that sort of thing. 248 00:20:45.825 --> 00:20:49.845 The the key in terms of architecture, the key step was this 249 00:20:50.225 --> 00:20:56.139 development of the builder and and how that was going to work. So there was an abstraction layer 250 00:20:56.440 --> 00:21:04.139 that I started and that this was very much designed. So I started thinking about what are the key aspects of of an experiment, 251 00:21:05.275 --> 00:21:07.535 and how should that look in terms of 252 00:21:08.395 --> 00:21:10.095 a Python dictionary, essentially? 253 00:21:10.395 --> 00:21:17.330 And then how do we save that dictionary to a file, and how do we represent that dictionary visually? So what we've got is 254 00:21:17.870 --> 00:21:19.009 a an experiment 255 00:21:19.309 --> 00:21:22.605 can be built in terms of a number of components, 256 00:21:23.705 --> 00:21:30.045 like a text object or a keyboard or a image, and those get combined into routines. 257 00:21:30.760 --> 00:21:34.539 Those routines are basically just a list of components, but the component knows 258 00:21:34.840 --> 00:21:41.075 within that routine when it should start, when it should stop, and what its parameters are. Then we've got a sort of a flow 259 00:21:41.934 --> 00:21:45.554 that combines those routines in various different ways, and 260 00:21:46.500 --> 00:21:48.840 and that's essentially the whole thing. So 261 00:21:49.140 --> 00:21:52.279 so the flow is a list of routines 262 00:21:52.740 --> 00:22:03.165 that happen in a certain order, and that's all there is to it. It calls each routine and says, write your code. The routine then calls each of its components and says, write your code, 263 00:22:03.545 --> 00:22:06.420 and and then goes to the next routine 264 00:22:06.720 --> 00:22:08.500 in the flow list. 265 00:22:08.880 -->

00:22:14.740 So by having these units that are combined in various different ways, it turns out that you can 266 00:22:15.235 --> 00:22:17.095 make quite a rich structure 267 00:22:17.635 --> 00:22:23.975 of experiments. And you can have loops on the flow that cause something to repeat. And, of course, in terms of the Python code, 268 00:22:24.520 --> 00:22:25.340 that's just 269 00:22:25.640 --> 00:22:27.420 adding an indent level, essentially. 270 00:22:28.040 --> 00:22:28.540 And 271 00:22:28.920 --> 00:22:37.315 the end of the loop on your flow just de dense the code again, and we carry on to the next piece of code. So in terms of 272 00:22:37.615 --> 00:22:38.915 the graphical interface, 273 00:22:40.190 --> 00:22:46.210 writing the code, it was all relatively straightforward. There's this abstraction layer, which is a dictionary containing 274 00:22:46.669 --> 00:22:47.490 some lists. 275 00:22:48.110 --> 00:22:48.610 And 276 00:22:49.315 --> 00:22:59.429 that gets saved to a data file, and we we sort of created an an open standard not a data file. Sorry. We created an open standard sort of experiment file, which is an 277 00:22:59.809 --> 00:23:20.380 XML formatted file that essentially just stores these lists and dictionaries with their various parameter types. So that can be loaded into or potentially, I don't think anyone's written anything that has loaded these into other pieces of software. But that that's agnostic about what the software is that's using it. It could be loaded into another piece of software, and it literally just defines 278 00:23:21.235 --> 00:23:21.735 what 279 00:23:22.035 --> 00:23:24.934 sequences of stimuli and keyboard, etcetera, 280 00:23:25.235 --> 00:23:27.735 get used to create an experiment and how those 281 00:23:28.115 --> 00:23:28.615 iterate 282 00:23:28.995 --> 00:23:31.460 to create a full study. It's really a graphical interface 283 00:23:32.000 --> 00:23:35.140 that writes the code. So that that's that's the key part. Builder 284 00:23:35.840 --> 00:23:47.275 writes your script for you. And so and it literally writes a Python script. Going forwards, and we'll I think we'll come on to this later on. But going forwards, we're actually now having it output a JavaScript 285 00:23:48.054 --> 00:23:48.554 script 286 00:23:48.880 --> 00:23:49.540 instead. So 287 00:23:50.400 --> 00:24:00.664 we will be able to this this builder interface that defines your experiment doesn't need to care about what language the script is that it writes. It just has these chunks of code, 288 00:24:01.365 --> 00:24:04.904 and, those chunks of code be could be in any syntax you choose. 289 00:24:05.605 --> 00:24:09.170 So how easy is it to extend PsychoPy with new types of stimulus? 290 00:24:10.670 --> 00:24:13.410 Super easy. That's straightforward. Either in the library, 291 00:24:13.715 --> 00:24:20.535 you can add your own Python classes or subclass existing ones. That's all very straightforward. 292 00:24:20.980 --> 00:24:28.920 The rendering code is all pure OpenGL, so some people have, I don't know, done some transforms. For instance, they've got some particular 293 00:24:30.115 --> 00:24:30.615 unusual 294 00:24:31.154 --> 00:24:33.495 display, like a like a huge 295 00:24:34.034 --> 00:24:34.534 curved 296 00:24:35.154 --> 00:24:39.680 screen rather than a regular computer display. So they've had to 297 00:24:40.460 --> 00:24:51.414 subclass the window in order to add some alternative spatial transform to their window, for instance. So So all of those sorts of things are really easy to do. Or just adding a new stimulus, someone just recently contributed 298 00:24:51.794 --> 00:24:52.455 a new 299 00:24:52.835 --> 00:25:07.015 stimulus is essentially a visual button that people can click on within the regular Open GL screen. So all of those things can be added very easily in the library. But, also, in the builder, the builder is completely modular. You can add your own components, 300 00:25:07.555 --> 00:25:11.095 and all that a component needs is it needs to know what its 301 00:25:11.635 --> 00:25:23.715 pieces of code are that need to get executed at the beginning of the experiment, at the beginning of a trial, on every frame, at the end of a trial, etcetera. So it has 6 pieces of code. And if it knows what its 302 00:25:24.575 --> 00:25:26.035 initialization code is, 303 00:25:26.415 --> 00:25:29.235 then the builder will call that initialization 304 00:25:29.535 --> 00:25:32.150 code for each of the components in turn. 305 00:25:32.530 --> 00:25:34.950 And so all you've got to write for your new component 306 00:25:35.570 --> 00:25:38.550 is how how should my object be initialized, 307 00:25:39.215 --> 00:25:42.914 and and and you're done. So, yeah, it's it's really very extensible. 308 00:25:43.535 --> 00:25:46.115 And some people then go on and share those 309 00:25:46.500 --> 00:25:50.200 extensions that they've made with the rest of the community. I'm sure loads of other users 310 00:25:50.740 --> 00:25:52.760 are kind of shy to share their code, 311 00:25:53.140 --> 00:26:00.045 and they'll keep their component just in their own lab. And that's fine too. You know? So but it's being widely extended. 312 00:26:00.985 --> 00:26:05.860 That's great. So what are some interesting challenges you faced when implementing PsychoPy? 313 00:26:07.920 --> 00:26:25.520 The interesting intellectual challenge has been how do you create create this sort of abstraction of an experiment that isn't about lime so working out what that should look like in a way that's really general, and that would that would be easy enough that a 1st year undergraduate in psychology who's 314 00:26:25.900 --> 00:26:26.560 a technophobe 315 00:26:26.940 --> 00:26:29.360 can still create their study. But also 316 00:26:29.725 --> 00:26:31.345 that a professor in neuroscience 317 00:26:32.045 --> 00:26:33.025 is quite happy 318 00:26:33.325 --> 00:26:36.690 to use the software to

create their fancy experiment as well. So 319 00:26:37.250 --> 00:26:41.029 to to generate something that was sufficiently general that it could run 320 00:26:41.330 --> 00:26:46.765 a a real scientific study, but also sufficiently easy that a new user wasn't 321 00:26:47.385 --> 00:26:51.085 terrified. That that's been that's been the big interesting challenge, 322 00:26:51.625 --> 00:26:57.809 and and that's been a lot of fun trying to get those things right. The more frustrating challenges have been 323 00:26:58.190 --> 00:26:59.169 trying to support 324 00:27:00.285 --> 00:27:01.105 multiple platforms 325 00:27:01.885 --> 00:27:02.385 and 326 00:27:02.845 --> 00:27:07.185 trying to keep a track of all. So we have lots of dependencies. You know, I didn't want to write 327 00:27:07.649 --> 00:27:08.149 my 328 00:27:08.450 --> 00:27:09.830 own library for loading 329 00:27:10.210 --> 00:27:11.750 videos and and 330 00:27:12.529 --> 00:27:13.669 another 1 for 331 00:27:14.049 --> 00:27:20.355 presenting this, I don't know, other other sound stimulus or or something like that. So we we used a lot of other dependencies, 332 00:27:20.975 --> 00:27:25.770 and those keep changing. And that's that's quite frustrating that I spent a lot of time. 333 00:27:26.150 --> 00:27:30.570 Just each time that we provide the new distribution, the new large application, 334 00:27:31.645 --> 00:27:37.025 I have to go and dig through what's changed. And I don't know if pandas has changed its input, 335 00:27:37.805 --> 00:28:12.945 call or something like that, and then and something else has changed. It's, it's become 64 bit only or something. And so maintaining those has been quite hard, and I'm afraid we do have releases where we push out the package and and then someone comes in and says, oh, but the library has changed or there's some 1 of my dependencies has disappeared out of the package, and it's because I've had to rebuild everything from scratch at some point. That's been the hardest part and the most frustrating part as a as a developer, keeping on top of many different platforms, trying to test on all of the different platforms. So we do have a test suite, but a test suite 336 00:28:13.405 --> 00:28:17.040 only really works for the library. It's not very good for an application because 337 00:28:19.020 --> 00:28:19.520 testing 338 00:28:20.300 --> 00:28:22.240 the sequence of button clicks 339 00:28:22.540 --> 00:28:23.360 is is 340 00:28:23.980 --> 00:28:26.115 on on arbitrary display 341 00:28:26.575 --> 00:28:30.755 shapes is is a really hard thing to do. So a lot of the application 342 00:28:31.135 --> 00:28:31.615 side, 343 00:28:32.015 --> 00:28:40.650 isn't tested in the test suite. And and that means that, I'm afraid, sometimes we release things where I haven't gone and pressed all of the buttons 344 00:28:41.030 --> 00:28:41.530 on 345 00:28:42.154 --> 00:28:46.895 Linux and on Windows 7 and on Windows 10 and on Windows 10 346 00:28:47.274 --> 00:28:57.270 64 bit. You know what I mean? So sometimes we break things that way because I haven't been able to test all those things. That's that's quite frustrating for me and for the users. 347 00:28:58.130 --> 00:28:58.630 But 348 00:28:59.274 --> 00:29:03.855 but on the whole, I think I think the program the ship is gradually moving forwards, not backwards. 349 00:29:04.634 --> 00:29:11.150 That's great. I should point out that, I sometimes talk about I, but the other users and the other contributors 350 00:29:11.850 --> 00:29:15.230 are huge in this aspect. And and a lot of people 351 00:29:15.884 --> 00:29:18.304 will submit quick fixes. You know, they will 352 00:29:18.605 --> 00:29:30.220 users will notice that something's not working. And rather than saying it's broken, they will come back and say, hey. I've discovered that this doesn't work anymore, and I've I've dug around, and it's because 353 00:29:30.760 --> 00:29:32.940 pandas now needs this new argument. 354 00:29:33.405 --> 00:29:38.145 So here's the here's the change on GitHub, and and I can just do a pull 355 00:29:38.765 --> 00:29:54.495 from from their pull request on on GitHub. And and that has GitHub has made my life so much easier in terms of that sort of thing. And it means with it's really genuinely a community, so I I should stop using I altogether. It was I for about 7 years. And and now when you look at the contributions, 356 00:29:55.434 --> 00:30:10.794 some of the releases don't have any contributions from me at all. I just merely put it all together and and release it. So so, yeah, the community has grown really strong around it. Well, that that's a testament to your hard work on the project that made it a success that made people want to contribute. So 357 00:30:11.095 --> 00:30:13.115 1 thing real quick, you mentioned previously 358 00:30:13.450 --> 00:30:16.510 that in future releases, PsychoPy may emit JavaScript. 359 00:30:16.970 --> 00:30:25.375 Is that at least in part an attempt to deal with some of these difficulties into integrating third party multimedia libraries? 360 00:30:26.154 --> 00:30:28.095 That wasn't the aim of that, 361 00:30:28.400 --> 00:30:32.420 but it has crossed my mind that increasingly some of the problems, 362 00:30:32.880 --> 00:30:46.169 yeah, will get solved. And and so running a study online in a browser was something that people have been asking for that as a feature for a very long time, and I've always just written back and said, no. PsychoPy is about hardware accelerated 363 00:30:47.510 --> 00:30:52.410 graphics, and and it needs that in order that an interpreted language can do things 364 00:30:53.085 --> 00:31:03.100 in real time. And so timing precision is really important to us. We need to know, and most pieces of software wouldn't know, exactly when did your screen refresh, for instance. 365 00:31:03.400 --> 00:31:04.620 And we want to know, 366 00:31:04.920 --> 00:31:07.179 did I manage to get my stimulus drawn 367

00:31:07.880 --> 00:31:17.085 before the screen refresh occurred? Because if not, then our timing went wrong. So I basically always push back on these users that were requesting 368 00:31:17.545 --> 00:31:30.275 a web interface because I was just like, no. The the browser can't do that. It doesn't have the the graphics hardware, pardon me, to be able to perform at that level with that level of precision. So So I always just said no. And then relatively recently, 369 00:31:30.895 --> 00:31:40.370 WebGL has has changed all of that, and even Canvas to some extent in the HTML 5. Canvas code has has changed a lot of that, and you can actually now with WebGL, 370 00:31:40.670 --> 00:31:43.170 you can use Open GL shader language 371 00:31:43.725 --> 00:31:48.545 in a browser. It's it's it's crazy what you can do. And suddenly, all of that sort of 372 00:31:49.485 --> 00:31:49.985 performance 373 00:31:50.285 --> 00:31:53.769 issue that I thought we would have is disappearing. And 374 00:31:54.070 --> 00:32:00.809 the performance will probably never be quite so good in terms of how quickly you can get responses, and we'll never be able to connect to hardware 375 00:32:01.825 --> 00:32:05.845 like parallel ports and that sort of thing from a browser. But for an awful lot of studies, 376 00:32:06.305 --> 00:32:08.005 suddenly, the web actually 377 00:32:08.305 --> 00:32:09.285 has the performance 378 00:32:10.000 --> 00:32:15.140 that we might be looking for. So that it was that change that was really driving this this 379 00:32:15.920 --> 00:32:17.860 potential of creating JavaScript. 380 00:32:18.455 --> 00:32:28.340 But as a side issue, I've I've then started looking at the way that those things are done and and, particularly a library called pixie. Js that that I've been, 381 00:32:28.880 --> 00:32:33.620 looking into a lot lately. And, wow, it makes it makes all of that media stuff really easy. 382 00:32:34.240 --> 00:32:37.795 And and a lot of the things that have sometimes caused me a lot of pain 383 00:32:38.255 --> 00:32:42.915 might get a lot easier. But it will require quite a lot of testing as well. You know, when you ask a sound 384 00:32:43.215 --> 00:32:44.915 to be played in your browser, 385 00:32:45.250 --> 00:32:48.070 how long does it actually take for noise to come out of the speakers? 386 00:32:49.170 --> 00:32:55.125 That will take a lot of testing for us to know exactly how good it is. And and it it may be that the performance 387 00:32:55.825 --> 00:32:57.205 is still not that great, 388 00:32:57.825 --> 00:33:00.005 and we'll have to look into that a lot. 389 00:33:00.705 --> 00:33:03.960 And I imagine that by being able to target the browsers 390 00:33:04.340 --> 00:33:05.240 using the 391 00:33:05.700 --> 00:33:07.560 same builder interface, 392 00:33:08.180 --> 00:33:11.024 would potentially drastically increase the size 393 00:33:11.325 --> 00:33:23.110 of experiments that can be conducted because of the fact that you can just publish them to a website and have people interact with the experiment without necessarily having to be in a lab environment. Exactly. Yeah. It's a it's a game changer for 2 reasons, actually. 1 is because 394 00:33:26.745 --> 00:33:28.125 you can send an email around 395 00:33:28.505 --> 00:33:30.125 and get people to participate 396 00:33:30.505 --> 00:33:43.490 that you've never met, that you didn't have to bring into a lab and individually get them to sign off the forms and sit them down and explain the you know, you just send them an email. And so technologies like Mechanical Turk have been quite big recently 397 00:33:43.870 --> 00:33:45.250 in getting people 398 00:33:46.355 --> 00:33:54.695 to essentially be enrolled in a study and paid by Amazon via Mechanical Turk to be a participant in a study. And some of these studies are having 399 00:33:55.160 --> 00:33:56.860 tens of thousands of participants. 400 00:33:57.240 --> 00:34:07.325 Whereas if you're going to run those participants yourself in your lab, you'd have maybe a 100 if you've got enough time. So, yeah, it's it's a game changer from that perspective. But, also, 401 00:34:07.705 --> 00:34:14.150 it means that you can if we're running a study in a web browser, we can run a study on any device. Right now, PsychoPy 402 00:34:14.450 --> 00:34:16.630 needs something that will support 403 00:34:17.089 --> 00:34:21.270 Python and Piglet and OpenGL and all of those dependent libraries. 404 00:34:21.615 --> 00:34:24.974 And that means it won't run on an iPad, and it won't run on 405 00:34:25.615 --> 00:34:27.954 well, it will run on a tablet if you 406 00:34:28.414 --> 00:34:32.619 want to go and install Linux on your tablet or Windows, I guess, or Windows Surface. 407 00:34:32.920 --> 00:34:33.420 So 408 00:34:34.200 --> 00:34:37.020 suddenly, move it to a browser, move it to a website, 409 00:34:37.695 --> 00:34:47.570 and your existing experiment that you created in Builder can now be compiled to a website, but it can be run on any device, whether it's a phone, whether it's Android, whether it's iOS. 410 00:34:47.950 --> 00:34:53.490 All it needs is something that supports JavaScript and a web browser. So, yeah, it's a it's a game changer 411 00:34:53.835 --> 00:34:57.135 in many ways, and and it'll be very exciting to see 412 00:34:57.435 --> 00:34:59.295 how that how that pans out. 413 00:35:00.555 --> 00:35:05.420 Yeah. And I think another thing worth mentioning too is that by having that greater reach and, 414 00:35:05.819 --> 00:35:10.960 potentially much larger audience, I imagine it's easier to prove statistical significance 415 00:35:11.405 --> 00:35:14.705 and by the fact that you do have such a larger corpus of data to work with. 416 00:35:15.565 --> 00:35:17.345 You can dig around and find things 417 00:35:17.885 --> 00:35:24.930 more with with very large numbers. I'm a little bit concerned actually to some extent on that score because you could also find yourself,

418 00:35:25.710 --> 00:35:27.970 reporting a lot of noise, you know, or 419 00:35:28.270 --> 00:35:29.410 tiny, tiny effects, 420 00:35:30.350 --> 00:35:44.380 that wouldn't have been significant when you look at a few hundred people are significant in in tens of thousands of people. But, actually, if it's that small that you needed tens of thousands of people to measure it, maybe it's not big enough that we should be carrying. That's that's a that's a personal query. 421 00:35:44.839 --> 00:35:46.859 Yes. That's a scientific issue. 422 00:35:47.905 --> 00:35:57.670 In looking through the documentation, I noticed that you support a number of different formats for the output data, including pickle. I'm wondering what are some of the most popular analysis tools for users of PsychoPy. 423 00:35:58.850 --> 00:36:01.990 Yeah. Sure. So pickle file is 424 00:36:02.690 --> 00:36:04.870 something that we save as a sort of 425 00:36:05.545 --> 00:36:11.964 safety net, and it it contains an awful lot of it of information. So it contains a copy of your experiment, 426 00:36:12.984 --> 00:36:45.055 which pickle is sort of, a very it's literally a copy of the study that you ran as the data were were collected, and it's and it's the Python object just pickled. So that's a pretty useful thing that that can, for instance, be used later on to save out some of the other formats as well later on. So if you accidentally ruined your CSV file, which I think is what most people would use for their analysis, you can go back to the pickle file, recreate the CSV file from it because it's just a copy of your experiment. Or if you forgot 1 of the parameters 427 00:36:45.515 --> 00:36:57.585 of your study and you accidentally didn't save it in the in the CSV file, then you can go back and and dig out what the actual parameter was from the script that has been saved within your pickle. So that's a sort of a belt and braces 428 00:36:58.125 --> 00:37:03.025 kind of a data format that I suspect most people aren't using for for real analysis. 429 00:37:03.450 --> 00:37:06.990 Although you you can load up that pickle file, and you can 430 00:37:07.610 --> 00:37:08.350 use NumPy 431 00:37:08.650 --> 00:37:14.375 or pandas to go and analyze. I mean, the the data were then saved in NumPy arrays within that file. 432 00:37:14.835 --> 00:37:20.290 And and so you you could certainly use that that way if you like. We also output a CSV file, 433 00:37:20.910 --> 00:37:26.609 so a comp comma separated value file that most people would use. It's a trial by trial, 434 00:37:27.015 --> 00:37:27.755 what happened. 435 00:37:28.055 --> 00:37:29.115 But you have to 436 00:37:29.815 --> 00:37:36.395 tell the CSV file, or you have to tell PsychoPy what to put into the CSV file. So what what variables 437 00:37:37.320 --> 00:37:44.700 are of interest to you, and it's going to load those as headers in the file. So a builder will try and do that for you 438 00:37:45.465 --> 00:37:53.085 automatically, but but maybe there's something that you wanted to study within your experiment that you need to tell Builder that it needs to do manually. 439 00:37:53.600 --> 00:37:55.860 So, occasionally, people will get that wrong, 440 00:37:56.400 --> 00:38:03.895 and then that's where the log file comes in. So so that we also output a log file, which is just an automated 441 00:38:04.355 --> 00:38:04.855 stream 442 00:38:05.475 --> 00:38:05.975 of 443 00:38:06.515 --> 00:38:13.390 events happening in your study as it occurs. So I've just changed the stimulus to this. I've just done that. We've just started a new trial. 444 00:38:13.850 --> 00:38:15.470 Subject has just pressed this button. 445 00:38:16.250 --> 00:38:18.750 Now that's not always easy to analyze 446 00:38:19.805 --> 00:38:22.065 because it's a stream of way too much information. 447 00:38:22.365 --> 00:38:26.945 Just in chronological order, and it doesn't it doesn't know the semantics of your study. It's just, 448 00:38:27.430 --> 00:38:32.010 a chronological stream. But that's very useful if you've forgotten to do something 449 00:38:32.390 --> 00:38:34.730 that you can go back and check all of the details, 450 00:38:35.194 --> 00:38:52.215 and you could write a script that passed this nasty log file and and, you know, dug out the data that were critical to you, in most cases. So so we've got various there's also an Excel file, which which we basically don't recommend anyone use. You can save your data in an Excel file if you so desire. 451 00:38:52.755 --> 00:38:58.775 In terms of the what people are using for analysis, I guess a lot of our users would be in psychology, and they would be using SPSS. 452 00:38:59.560 --> 00:39:03.260 We would naturally try to encourage people to use either Python or R. 453 00:39:03.640 --> 00:39:05.740 So they might load their data in pandas, 454 00:39:06.535 --> 00:39:09.915 or they might load their data as a as a table in R. 455 00:39:10.455 --> 00:39:12.635 And and the CSV file is designed 456 00:39:13.255 --> 00:39:14.795 so that either of those packages 457 00:39:15.300 --> 00:39:23.460 would would allow you to to load your data in a very easy step and start doing start doing analysis on that. So so that's what I would, 458 00:39:24.020 --> 00:39:24.520 naturally 459 00:39:25.204 --> 00:39:30.984 imagine most people to do. Again, I used to do my analysis on the PQL file using NumPy arrays because 460 00:39:31.285 --> 00:39:32.825 CircaPy predates pandas. 461 00:39:33.260 --> 00:39:33.760 And 462 00:39:34.060 --> 00:39:37.500 so pandas was a was a new thing to me that I, 463 00:39:37.980 --> 00:39:39.920 I subsequently learned is really useful. 464 00:39:40.505 --> 00:39:45.725 Yeah. Pandas is a somewhat recent entry on the scene, though it seems to have gained a lot of, 465 00:39:46.345 --> 00:39:51.790 following because of the fact that it's fairly easy to use and 2 dimensional arrays are a fairly 466 00:39:52.250 --> 00:40:09.770

large portion of the problem space that people were using NumPy for. And it adds a lot of niceties on top of that in terms of the data frame format. And I was gonna ask if you've looked into the use of the feather library for being able to add that interoperability between Python and R for people who want to do their analysis 467 00:40:10.230 --> 00:40:11.210 in in R? 468 00:40:12.069 --> 00:40:14.329 I haven't, actually. I don't I haven't 469 00:40:14.685 --> 00:40:16.385 personally worked with the Feather Library 470 00:40:16.845 --> 00:40:22.305 at all. I haven't really used R myself. I'm I'm still using Python 471 00:40:22.829 --> 00:40:26.210 because I've now got all of my scripts that will do various 472 00:40:26.589 --> 00:40:29.809 scientific analyses, like bootstrapping or whatever and simulations 473 00:40:30.535 --> 00:40:31.275 in Python. 474 00:40:31.575 --> 00:40:34.795 I've never had the need then to go and learn R. So 475 00:40:35.255 --> 00:40:36.635 but it's something that 476 00:40:37.175 --> 00:40:38.715 a number of our other users 477 00:40:39.210 --> 00:40:46.190 are certainly using R a great deal, and and I suspect they will be digging around in Feather as well, but but I haven't got that far yet myself. 478 00:40:47.055 --> 00:40:59.589 And how is data input typically managed? I'm wondering if PsychoPy supports automated readings from test equipment or if it's something that's the responsibility of the people conducting the experiment to control the data input. 479 00:41:00.290 --> 00:41:02.470 Yeah. No. Typically, it comes from 480 00:41:03.175 --> 00:41:03.995 the experimenter 481 00:41:04.615 --> 00:41:07.515 writing the code to handle controls or whatever. 482 00:41:07.895 --> 00:41:09.435 A lot of it obviously is very 483 00:41:10.220 --> 00:41:10.720 straightforward. 484 00:41:11.100 --> 00:41:14.240 I mean, most of most of the inputs are coming from things like, 485 00:41:14.780 --> 00:41:17.520 keyboards and and microphones and and 486 00:41:18.055 --> 00:41:20.075 mice. So that's that's all 487 00:41:20.775 --> 00:41:21.515 easy enough. 488 00:41:22.215 --> 00:41:24.875 They might if if you've got some hardware like, 489 00:41:25.180 --> 00:41:28.400 I mentioned MRI before. So if you're doing an fMRI study, 490 00:41:28.940 --> 00:41:29.440 then 491 00:41:29.900 --> 00:41:34.079 you need to monitor probably the trigger pulses that the 492 00:41:35.025 --> 00:41:39.765 device is sending out. So a scanner will will send out a pulse every, say, 2 seconds 493 00:41:40.145 --> 00:41:41.445 to announce that it's 494 00:41:41.760 --> 00:41:45.620 just about to start creating a new image of the brain. And 495 00:41:46.080 --> 00:41:51.380 so you need to detect when those things occur. But those are relatively straightforward things. They're either 496 00:41:51.715 --> 00:41:55.015 a serial input or a parallel port import or 497 00:41:55.555 --> 00:41:58.215 a a keyboard sometimes that they'll they'll simulate 498 00:41:58.730 --> 00:41:59.950 pressing a key. 499 00:42:00.250 --> 00:42:00.750 So 500 00:42:01.290 --> 00:42:02.830 those are all relatively straightforward, 501 00:42:03.450 --> 00:42:06.910 and it does come down to the user writing writing their code 502 00:42:07.475 --> 00:42:09.335 to to interface with them. 503 00:42:09.875 --> 00:42:17.799 Depending on what sort of hardware, we might have them built in, and and for some, you might need to add your own. So eye tracking is another good example where 504 00:42:18.339 --> 00:42:23.160 we've got support for, I don't know, 6 or 7 different eye trackers. 505 00:42:24.019 --> 00:42:24.519 And 506 00:42:25.765 --> 00:42:30.825 if if you're using 1 of those, then you can plug into a sort of a unified API 507 00:42:31.445 --> 00:42:47.455 and, say, on each on each frame or on each trial, find out what the position of the eyes was at this point using 1 set of function calls. So we've got built in support for quite a few. But if you've got your own eye tracker that's different to 1 of those, then you might need to use your own custom Python code. 508 00:42:48.075 --> 00:42:50.015 Hopefully, your your hardware 509 00:42:50.315 --> 00:42:53.375 supplier will will support Python as well. So 510 00:42:53.730 --> 00:42:56.950 it it depends on what particular devices you're using. 511 00:42:59.330 --> 00:43:06.845 And I imagine that because of the fact that timing is of such critical importance in these experiments that it is certainly necessary 512 00:43:07.145 --> 00:43:10.525 to make sure that those data inputs are being fed through the PsychoPy 513 00:43:11.260 --> 00:43:15.200 program so that you can correlate the input times with the stimulus times? 514 00:43:15.660 --> 00:43:17.760 Yeah. So we've got 515 00:43:18.665 --> 00:43:19.165 essentially 516 00:43:20.265 --> 00:43:21.005 2 different 517 00:43:21.545 --> 00:43:25.565 ways of doing this, and and and 1 1 is just using the traditional 518 00:43:26.025 --> 00:43:26.525 Piglet 519 00:43:27.150 --> 00:43:30.450 sort of, which is the the graphical library that we use. 520 00:43:31.069 --> 00:43:41.434 It's like pygame for those who don't know. So we might be just, say, recording key presses using that, which is really easy to do, but isn't actually super precise. So so, typically, 521 00:43:41.734 --> 00:43:44.395 you'd be updating your stimulus on every frame, 522 00:43:45.010 --> 00:43:53.910 and that means you're tied to a refresh rate of, say, 60 hertz. So if you're checking the keyboard once per frame in in part of that 523 00:43:54.815 --> 00:43:56.675 flow of code, then, 524 00:43:57.055 --> 00:43:59.875 of course, you're only checking the keyboard every 16 milliseconds. 525 00:44:00.975 --> 00:44:01.475 And 526 00:44:02.210 --> 00:44:15.214 so there's there's an alternative system that we have, which is which is a a library called IO Hub, which is now really very much integrated in PsychoPy. It's it's really a part of PsychoPy. And that was

written by a contributor 527 00:44:15.515 --> 00:44:18.015 who comes from the eye tracking world, Sol Simpson. 528 00:44:18.395 --> 00:44:18.895 And 529 00:44:19.329 --> 00:44:22.390 he has written that to essentially be monitoring 530 00:44:22.690 --> 00:44:24.549 hardware in a separate process 531 00:44:25.010 --> 00:44:27.510 and passing g events back to 532 00:44:27.890 --> 00:44:28.710 the Python. 533 00:44:29.715 --> 00:44:33.895 Well, it is a Python process as well, but it's it's passing events 534 00:44:34.355 --> 00:44:35.495 back to the PsychoPy 535 00:44:36.275 --> 00:44:37.095 main process. 536 00:44:37.650 --> 00:44:39.270 And this avoids the the, 537 00:44:39.730 --> 00:44:46.310 global interpreter lock and allows you to to run your hardware acquisition, your data acquisition, 538 00:44:46.965 --> 00:44:48.505 and and outputs at times 539 00:44:49.285 --> 00:44:50.985 on a separate process, on a separate, 540 00:44:52.085 --> 00:44:59.370 core if you assuming that you've got a multi core machine. Now that has methods for synchronizing in both directions. Either 541 00:45:00.070 --> 00:45:09.645 either it can save its own data file, so it it also supports a a data format of its own based on HDF 5. And so it can stream data 542 00:45:10.105 --> 00:45:15.640 at a kilohertz, say, which is what some eye trackers will be using. It'll stream data at a kilohertz 543 00:45:16.099 --> 00:45:21.640 down to an HDF 5 file, but also you can pull it from the main PsychoPy process 544 00:45:22.005 --> 00:45:24.985 to get some snapshot of what's going on, for instance. 545 00:45:25.365 --> 00:45:25.865 So 546 00:45:26.165 --> 00:45:28.745 we have ways of of doing that. But using IOhub, 547 00:45:29.720 --> 00:45:31.820 you know, it's a lot more intensive, 548 00:45:32.760 --> 00:45:38.380 in terms of the the user has to work a little bit harder to to get that all working. It's it's 549 00:45:38.785 --> 00:45:39.525 more complicated. 550 00:45:39.825 --> 00:45:41.365 It's running a separate process 551 00:45:41.665 --> 00:45:44.245 and sending events backwards and forwards. So 552 00:45:44.625 --> 00:45:50.710 for simple studies, I think people still tend to use the the the basic get keys kind of calls, 553 00:45:51.250 --> 00:45:53.750 even though those aren't quite as, 554 00:45:54.130 --> 00:45:56.950 temporarily precise as as using IOhub. 555 00:45:58.244 --> 00:46:00.025 It, it strikes me that 556 00:46:00.405 --> 00:46:07.840 this kind of framework with all this eye tracking and millisecond scale input and rich visual display capabilities 557 00:46:08.460 --> 00:46:15.760 could also be really useful outside the field of psychology for things like UX researchers or things like that to 558 00:46:16.285 --> 00:46:19.345 test new input paradigms and the like? 559 00:46:20.444 --> 00:46:22.625 Yeah. I don't have statistics 560 00:46:23.005 --> 00:46:28.980 on where this is being used outside psychology, which I'd really like to know about, actually. Anyone who's 561 00:46:29.280 --> 00:46:30.180 using PsychoPy 562 00:46:30.720 --> 00:46:32.100 not in the world of academia, 563 00:46:32.565 --> 00:46:37.065 I'd really like to hear from you how you're using it because I it helps me to persuade 564 00:46:37.925 --> 00:46:39.385 people that this is a, 565 00:46:39.845 --> 00:46:41.945 a useful endeavor outside academia. 566 00:46:42.550 --> 00:46:46.250 Certainly, some people are using it in other things like in in clinical research 567 00:46:46.630 --> 00:46:47.130 and, 568 00:46:47.910 --> 00:46:48.970 in in developing 569 00:46:49.975 --> 00:46:52.555 hardware, I believe. But but my 570 00:46:53.335 --> 00:46:54.475 I get to track 571 00:46:54.775 --> 00:46:56.475 how many people are using PsychoPy, 572 00:46:57.080 --> 00:47:10.695 but not in what way. So so I I just don't have very much information about that. Certainly, I've been amazed at the variety of experiments and the variety of ways that people have have used it. I had in mind a very limited 573 00:47:10.995 --> 00:47:16.760 set of use, particularly for the builder. And people have gone way beyond what I ever imagined 574 00:47:17.220 --> 00:47:19.079 and creating quite elaborate studies 575 00:47:19.540 --> 00:47:21.960 that, I hadn't really pictured being done. 576 00:47:22.295 --> 00:47:26.955 What are some of the most interesting experiments that you're aware of having been conducted using cycle pi? 577 00:47:27.735 --> 00:47:30.075 Oh, goodness. That's that's a really hard question. 578 00:47:30.440 --> 00:47:41.895 Thousands of studies have been run, but I don't follow what they all are. So so that's a it's a very hard question to answer. 1 of the demos that we actually supply that I think is really fun is called the balloon analog risk task, 579 00:47:42.515 --> 00:47:45.975 and that's a a way of measuring your risk taking behavior 580 00:47:47.109 --> 00:47:52.010 by allowing you to blow up a a virtual visual balloon, and 581 00:47:52.550 --> 00:47:57.215 it will pop at some point. And the question is, how far do you keep on blowing it 582 00:47:57.515 --> 00:48:05.600 before it pops? You get more score the bigger the balloon gets. But if it pops, you get nothing. And so it's a way of measuring your risk taking 583 00:48:05.900 --> 00:48:06.880 by investigating 584 00:48:07.660 --> 00:48:16.464 how far you blow up this balloon. So that's actually just 1 of the demos in PsychoPy, and that's actually actually kind of, a neat way of showing the level of flexibility 585 00:48:16.765 --> 00:48:23.180 that that we have and the you know, it's an interactive study. It's not a traditional sort of I present a stimulus, I collect a response. 586 00:48:23.480 --> 00:48:30.059 You're you're growing this balloon. So that sort of demos a little bit the the range of things that can be done and and 587 00:48:30.455 --> 00:48:32.635 maybe 1 of the more interesting types of studies. 588 00:48:32.935 --> 00:48:42.309 There are also

things like the implicit association test or IAT. So PsychoPy is used a fair bit for that in in various different labs, which is 589 00:48:42.770 --> 00:48:44.549 a sort of test where you 590 00:48:45.010 --> 00:48:46.130 you're trying to understand 591 00:48:46.995 --> 00:48:48.195 someone has a, 592 00:48:48.515 --> 00:49:00.290 an overt opinion of a particular stimulus. They say that they like this thing or they dislike this thing, and you're trying to work out if that's true. So you can present images of various different stimuli, and you ask people to respond 593 00:49:00.830 --> 00:49:03.650 about likes and dislikes. And you measure essentially 594 00:49:04.215 --> 00:49:16.309 if you know that they're wanting to respond that they like this thing that what that they actually don't. You can measure the reaction time to their response in a in a speeded task, in a in a rapid stream of stimuli. 595 00:49:16.849 --> 00:49:34.960 When you're intending to respond 1 way, but actually you feel something different, then it delays your response. So that would be an implicit association test, and that's sort of 1 of the myriad things that PsychoPy is using. But, you know, it's it's being it's being used in so many different brain scanning 596 00:49:35.625 --> 00:49:38.765 studies or brain computer interface studies as well. 597 00:49:39.145 --> 00:49:40.125 So so people 598 00:49:40.825 --> 00:49:41.724 that are using 599 00:49:43.270 --> 00:49:49.050 while measuring EEG, so electrical signals on the scalp and trying to control the stimulus 600 00:49:49.785 --> 00:49:52.684 by using the electrical stimulus by using the electrical 601 00:49:53.224 --> 00:49:57.700 activity on the scalp. So that's called a brain computer interface and or BCI. 602 00:49:58.160 --> 00:49:59.700 Some people are certainly using Sakupai 603 00:50:00.000 --> 00:50:07.125 for that sort of thing as well because it can present stimuli very rapidly based on inputs. So there there are quite a range of different things. 604 00:50:08.145 --> 00:50:24.135 While I was reading through the docs, I found the page describing the integration with the open science framework for sharing and validating an experiment and collected data with other members of the field. I'm wondering if you can explain why that's beneficial to the researchers and compare it with other options such as GitHub for use within the sciences? 605 00:50:25.395 --> 00:50:26.615 Sure. So 606 00:50:26.915 --> 00:50:28.995 this is a new development for us, 607 00:50:29.555 --> 00:50:30.775 and it's an interesting, 608 00:50:31.880 --> 00:50:32.619 an interesting 609 00:50:33.000 --> 00:50:36.380 new direction. The idea is that when I run my study, 610 00:50:36.760 --> 00:50:44.665 a, I want to share it with my collaborators in other departments, say, b, I want to share it with other computers in my lab. 611 00:50:44.965 --> 00:50:45.465 And 612 00:50:46.005 --> 00:50:48.025 and, c, afterwards, I 613 00:50:48.559 --> 00:50:50.819 potentially want to share that with the rest of the world 614 00:50:51.119 --> 00:51:01.915 in a convenient way that other people can go and download my data. Now the question about whether so open science framework allows those sorts of facilities. It allows you to store data 615 00:51:02.375 --> 00:51:05.175 that either can be private to you and and 616 00:51:05.910 --> 00:51:07.290 but also can be 617 00:51:07.829 --> 00:51:11.530 exposed to the rest of the world if you want to share it and make it public. Now 618 00:51:12.069 --> 00:51:14.635 that I didn't find their interface 619 00:51:15.015 --> 00:51:21.035 all that intuitive, although that's what it has over GitHub. So the idea is that for a lot of scientists, 620 00:51:21.335 --> 00:51:23.115 GitHub is really written for coders, 621 00:51:23.560 --> 00:51:24.780 and they don't 622 00:51:25.080 --> 00:51:27.340 they don't know what Git is. So 623 00:51:27.800 --> 00:51:31.020 the idea for them of going getting a GitHub account 624 00:51:31.445 --> 00:51:32.905 and uploading something 625 00:51:33.205 --> 00:51:33.705 using 626 00:51:34.085 --> 00:51:44.550 a terminal tool, although I guess now there are graphical interfaces for that as well, that that would put off quite a few scientists that are less technically competent. So open science framework is sort of a way in 627 00:51:44.930 --> 00:51:47.510 particularly, it's come from from the field of psychology 628 00:51:48.130 --> 00:51:49.750 to allow people to 629 00:51:50.395 --> 00:51:50.895 upload, 630 00:51:51.275 --> 00:51:52.015 make public, 631 00:51:52.555 --> 00:51:53.535 share with collaborators 632 00:51:54.395 --> 00:51:56.335 their their experiments and their data. 633 00:51:57.250 --> 00:52:00.950 Now I say, I didn't actually find their their user interface 634 00:52:02.210 --> 00:52:03.270 super easy 635 00:52:03.650 --> 00:52:07.275 to to for a new big a new user. So 636 00:52:07.575 --> 00:52:09.035 they actually had provided 637 00:52:09.575 --> 00:52:13.674 some funding for us from the Center For Open Science, who wrote open science framework, 638 00:52:13.974 --> 00:52:15.674 to allow us to synchronize 639 00:52:16.260 --> 00:52:17.080 files directly 640 00:52:17.620 --> 00:52:18.680 from within PsychoPy. 641 00:52:19.300 --> 00:52:20.440 And so 642 00:52:20.820 --> 00:52:26.035 you can now go to the projects menu as as of the new version that was just released, yesterday, 643 00:52:26.494 --> 00:52:26.994 1.841. 644 00:52:27.615 --> 00:52:29.714 You can go to the projects menu. 645 00:52:30.174 --> 00:52:30.914 You can 646 00:52:31.220 --> 00:52:33.480 take a folder from your local machine, 647 00:52:33.860 --> 00:52:41.355 and you can create a project on Open Science Framework based on that folder, and you can just press sync, and it will keep those 648 00:52:41.895 --> 00:52:45.515 projects in sync. You can also search for it from another computer 649 00:52:46.135 --> 00:52:46.635 and 650 00:52:47.010

--> 00:52:49.910 download that project. And, again, you can just do syncing 651 00:52:50.370 --> 00:52:50.870 from 652 00:52:51.330 --> 00:53:00.954 1 machine to another within your lab. And then at some point, if you want to, you can go to open science framework and and click a button to make that public. So it's about trying to 653 00:53:01.494 --> 00:53:02.555 enable users 654 00:53:03.340 --> 00:53:04.800 who are a bit less 655 00:53:05.180 --> 00:53:06.800 tech aware or or 656 00:53:07.420 --> 00:53:11.520 or don't want to have to learn lots of different interfaces and lots of different technologies. 657 00:53:11.975 --> 00:53:22.370 They've got the hang of PsychoPy. There's a menu that says projects. They can go to that menu and either search for an existing experiment. That's what I'm hoping in the longer term is that people will make their experiments 658 00:53:22.830 --> 00:53:23.810 available online. 659 00:53:24.350 --> 00:53:26.850 Other users will be able to go and search for, 660 00:53:27.230 --> 00:53:28.290 show me an existing 661 00:53:29.055 --> 00:53:29.555 implementation 662 00:53:30.015 --> 00:53:32.035 of the implicit association test. 663 00:53:33.295 --> 00:53:39.180 And then I can download that and build my work on that, and it will speed up science. That's the idea. 664 00:53:40.280 --> 00:53:49.075 And I imagine too that it will increase the transparency and repeatability of some of those experiments too, while at the same time creating 665 00:53:49.375 --> 00:53:53.640 the provenance for the data so that it's easier to go back and understand 666 00:53:53.940 --> 00:53:58.840 where the statistics come from when you're publishing the results of your experiment. 667 00:53:59.655 --> 00:54:08.155 That's that's definitely a great goal scientifically for Center For Open Science. And for me, I would love to see that happen. I don't think that's what's driving the individual 668 00:54:08.880 --> 00:54:11.060 scientist who's wanting to upload their data. 669 00:54:11.360 --> 00:54:13.940 So so right now, that isn't having people 670 00:54:14.320 --> 00:54:17.220 examine your data and look for flaws isn't an incentive 671 00:54:17.595 --> 00:54:18.495 to most scientists. 672 00:54:19.115 --> 00:54:23.215 But but being able to share it and encourage so in reality, 673 00:54:23.595 --> 00:54:25.055 getting other people to 674 00:54:25.435 --> 00:54:25.935 see 675 00:54:26.520 --> 00:54:30.780 your experiment and your original code and build their study on it, 676 00:54:31.080 --> 00:54:40.684 that improves the visibility of your work and gets more people to base their study on your work, which means you get more citations for your paper. So that is an incentive for a scientist 677 00:54:41.065 --> 00:54:44.125 that they should be aware of. They they should be trying to increase 678 00:54:44.570 --> 00:54:45.550 spin off projects 679 00:54:45.930 --> 00:54:46.910 from their work. 680 00:54:47.290 --> 00:54:51.390 And and this allow publicly sharing your code, as well as your data, 681 00:54:51.755 --> 00:54:58.655 allows allows people to create spin off projects more easily. And that that is something that that scientists should care about, 682 00:54:59.059 --> 00:55:01.319 whether or not they do it for the altruistic 683 00:55:01.700 --> 00:55:08.735 reason that they want science to progress well. And and, you know, it's all very good of you to to feel that you want, 684 00:55:09.195 --> 00:55:12.655 people to be able to examine your data and look for flaws. People 685 00:55:13.275 --> 00:55:17.510 in reality, it's a lot of effort, often to to make your data available 686 00:55:17.810 --> 00:55:21.670 in a format that other people like and and all that sort of thing. So 687 00:55:22.130 --> 00:55:26.285 people are often quite shy of doing that, unless there's 688 00:55:26.585 --> 00:55:31.085 a a clear incentive. So I try to focus a bit more on why it's good for the scientist, 689 00:55:31.625 --> 00:55:38.020 on the individual scientist, as well as the fact that I, of of course, I do believe that it's good for science as a field as well. 690 00:55:38.720 --> 00:55:43.335 Do you have a road map of features that you'd like to add to PsychoPy, or is it largely driven 691 00:55:43.635 --> 00:55:47.175 by contributions from practitioners who are extending it to suit their needs? 692 00:55:48.355 --> 00:55:51.815 Both. Very much both. There are lots of things that I would love to add 693 00:55:52.309 --> 00:55:54.569 and haven't had time to do. So 694 00:55:54.869 --> 00:55:56.970 PsychoPy's development has been 695 00:55:57.510 --> 00:55:58.569 driven by 696 00:55:59.645 --> 00:56:01.105 either things that my lab 697 00:56:01.485 --> 00:56:06.785 currently needs, and and I want to get in there, and all that a contributor's lab needs. 698 00:56:07.250 --> 00:56:14.630 And and so they're going to go and add that feature because they need it for their study right now. And that's the best way, 699 00:56:15.015 --> 00:56:20.875 in many cases for things to be added. There are lots of other things that I would love to add, and if I had time, I would. 700 00:56:21.415 --> 00:56:21.915 But 701 00:56:22.535 --> 00:56:24.155 my priority has to be with, 702 00:56:24.630 --> 00:56:33.210 you know, my PhD student and and and and that sort of thing. So it's there's a there's a mixture going on. There's also an issue that although 703 00:56:33.645 --> 00:56:39.745 I would love to add a certain feature so right now, I'm in discussion with someone who's wanting to contribute 704 00:56:40.685 --> 00:56:41.185 stereo 705 00:56:42.010 --> 00:56:42.510 displays 706 00:56:42.970 --> 00:56:43.470 and 707 00:56:44.170 --> 00:56:44.750 and enable 708 00:56:45.050 --> 00:56:45.550 PsychoPy 709 00:56:46.410 --> 00:56:56.434 to support, you know, shutter goggles and that sort of thing better without without the user having to write particularly hard code. So how do we make that easy for the user,

and how do we support 710 00:56:56.974 --> 00:56:57.795 8 different 711 00:56:58.095 --> 00:56:59.154 forms of stereo 712 00:57:00.000 --> 00:57:17.045 outputs from from a graphics card. Now I'm not the right guy to write that code because I don't have any of that stereo kit. I don't have a TV with shutter goggles that allows me to to work on them. And my lab isn't doing that sort of study. So it's never going to be the top of my priority list, 713 00:57:17.960 --> 00:57:20.540 and it would be very hard for me to go and test that rigorously. 714 00:57:21.080 --> 00:57:26.540 What we need is a user who runs those sorts of studies, who knows more about that technology, 715 00:57:27.545 --> 00:57:31.085 and who needs it in their next study. That's the way that the 716 00:57:31.465 --> 00:57:37.005 the code gets best written. And that's exactly what we've got going on right now is that there's someone coming to us. 717 00:57:37.480 --> 00:57:40.220 They've been using this sort of technology for a long time, 718 00:57:40.520 --> 00:57:41.020 and 719 00:57:41.400 --> 00:57:48.525 they know how to implement it. Well, you know, maybe with a little bit of support from me about how does the PsychoPy engine 720 00:57:48.825 --> 00:57:53.965 currently render things so that we we need to talk a little bit about how to interface 721 00:57:54.720 --> 00:57:57.060 his module with mine. That's fine. 722 00:57:57.680 --> 00:58:01.700 But that's that's the way that most of the code has sort of evolved. 723 00:58:02.434 --> 00:58:06.694 Or take for instance, the rating scale. You know, that came from Jeremy Gray, 724 00:58:07.234 --> 00:58:10.454 and I've never done a study that involved a rating scale, 725 00:58:11.450 --> 00:58:13.310 but he does them all the time. So 726 00:58:14.330 --> 00:58:19.790 he he wanted to have this stimulus available to make it easier for him to run his studies. And that's how it got contributed. 727 00:58:20.555 --> 00:58:23.535 There is a road map as well, though. And there are things that I 728 00:58:23.915 --> 00:58:25.535 do very much intend to, 729 00:58:26.795 --> 00:58:28.735 get done, either me or 730 00:58:29.080 --> 00:58:29.580 someone. 731 00:58:30.600 --> 00:58:32.380 Or, ideally, as we get bigger, 732 00:58:33.000 --> 00:58:34.620 we will hopefully have 733 00:58:35.320 --> 00:58:38.540 funding options, and we will have ways that we can pay 734 00:58:38.905 --> 00:58:41.085 people. So far, we've had very, 735 00:58:41.545 --> 00:58:44.204 very small amounts of money from a few places 736 00:58:44.585 --> 00:58:47.724 that are willing to contribute to to our open science drive. 737 00:58:48.069 --> 00:58:52.650 But 1 of those things is is absolutely the the development of web based experiments 738 00:58:52.950 --> 00:58:57.450 and the building of the JavaScript library. Again, I'm not the right guy to write that code. 739 00:58:57.785 --> 00:58:59.245 That's not something I've 740 00:58:59.705 --> 00:59:06.525 known much

about. I've not worked very much on that. So I would have to learn a great deal about JavaScript to write that myself. 741 00:59:07.250 --> 00:59:13.990 But we can we can get other people, probably hire other people. So we've been working with a company called elixir.com, 742 00:59:15.684 --> 00:59:19.145 and they've been writing the the sort of JavaScript library, 743 00:59:20.005 --> 00:59:22.730 that's the equivalent of the PsychoPy 744 00:59:23.030 --> 00:59:24.089 Python library. 745 00:59:24.390 --> 00:59:24.890 So 746 00:59:25.270 --> 00:59:26.569 that is something that will, 747 00:59:27.109 --> 00:59:28.490 for sure, be happening, 748 00:59:28.869 --> 00:59:30.410 and, hopefully, that will be 749 00:59:30.955 --> 00:59:36.015 over the next year or so. We'll we we've we've got that in a proof of concept right now. 750 00:59:36.395 --> 00:59:36.895 And, 751 00:59:37.355 --> 00:59:40.760 over the next year, I'm sure we'll we'll be supporting 752 00:59:41.700 --> 00:59:43.800 full experiments in a browser 753 00:59:44.420 --> 00:59:47.000 using open science framework to save your data 754 00:59:47.725 --> 00:59:49.185 and and using builder 755 00:59:49.645 --> 00:59:56.305 to generate the the web files that you need to do that. 1 of the things that we that I'm really keen on is 756 00:59:57.210 --> 00:59:57.710 making 757 00:59:58.810 --> 01:00:03.550 builder better at detecting when a user is about to do something that's not wise. 758 01:00:04.570 --> 01:00:05.470 So right now, 759 01:00:06.295 --> 01:00:08.875 you could ask builder to draw your stimulus 760 01:00:09.575 --> 01:00:10.715 smaller than a pixel 761 01:00:11.655 --> 01:00:14.155 or 4 meters to the right of your screen 762 01:00:14.615 --> 01:00:15.015 or, 763 01:00:16.320 --> 01:00:19.380 with an invisible texture, and PsychoPy won't complain. 764 01:00:19.760 --> 01:00:20.580 It will just 765 01:00:21.200 --> 01:00:22.820 fail to draw anything. Right? 766 01:00:23.120 --> 01:00:23.620 And 767 01:00:23.935 --> 01:00:32.675 to the user who gets no stimulus appear on their screen, that's confusing. So 1 of the things that PsychoPy needs to do is detect 768 01:00:33.309 --> 01:00:40.369 and do an integrity check of your experiment and say, you know, it looks like your stimulus is about to be drawn smaller than a pixel. 769 01:00:41.305 --> 01:00:45.484 Do is that really what you wanted to do? And just provide little warnings like that. 770 01:00:46.185 --> 01:00:47.724 That that sort of code, 771 01:00:48.190 --> 01:00:51.570 again, it's the sort of thing that because my lab doesn't need it because 772 01:00:51.950 --> 01:00:55.490 experienced users know what's going wrong and they won't make those mistakes, 773 01:00:55.870 --> 01:00:57.250 it's hard to make a priority, 774 01:00:57.605 --> 01:01:02.505 but it's something that I feel is really important that will prevent people from making silly errors, 775 01:01:03.125 --> 01:01:14.170 including me. I make silly errors in my own

experiments as well. And and I would like to warn me, oh, you're about to do that thing again that you've done a 1000000 times before, and you still haven't learned that that's wrong. 776 01:01:14.645 --> 01:01:16.805 So that's that's 1 step that's that's, 777 01:01:17.365 --> 01:01:19.705 another thing that's important for the future. 778 01:01:20.325 --> 01:01:25.780 Improving precision is always 1 more thing that we we're always trying to work to get 779 01:01:26.080 --> 01:01:27.780 faster playing of the sounds, 780 01:01:28.080 --> 01:01:32.340 faster rendering of of large movies, especially as we go to HD 781 01:01:32.785 --> 01:01:33.605 movies increasingly, 782 01:01:34.385 --> 01:01:38.005 we've got to improve the precision and performance of a lot of aspects of PsychoPy. 783 01:01:38.305 --> 01:01:40.244 And and that's something that's just a constant, 784 01:01:41.200 --> 01:01:43.380 aspect that we're always trying to improve. 785 01:01:45.359 --> 01:01:51.885 So are there any questions that we didn't ask that you think we should have or any other topics that you'd like to cover before we close out? 786 01:01:54.425 --> 01:01:55.325 No. I think 787 01:01:55.789 --> 01:01:59.329 1 of the things that I find interesting is this question about 788 01:01:59.869 --> 01:02:03.809 open source development versus the traditional commercial world and and 789 01:02:04.415 --> 01:02:08.035 software engineers versus academic engineers, but we ended up talking about that anyway. 790 01:02:09.615 --> 01:02:15.230 Do you want to talk any more about things like so 1 of the things that PsychoPy does that maybe 791 01:02:15.610 --> 01:02:18.510 is unusual is this tracking of users, and it's 792 01:02:18.810 --> 01:02:20.910 it's hard for an open source project 793 01:02:21.465 --> 01:02:22.845 to persuade people 794 01:02:23.465 --> 01:02:23.965 to 795 01:02:24.505 --> 01:02:26.365 to fund it and to persuade 796 01:02:26.745 --> 01:02:29.165 you know, in my case, I've got a job as an academic, 797 01:02:29.480 --> 01:02:31.740 but I have to persuade people that 798 01:02:32.040 --> 01:02:34.140 that's a useful use of my time. 799 01:02:34.440 --> 01:02:38.940 I'm not always making discoveries, but I'm making software tools that other people 800 01:02:39.245 --> 01:02:40.625 can use to make discoveries. 801 01:02:41.085 --> 01:02:49.850 And and so sometimes it can be hard to express and and and and persuade people that these projects are important. 802 01:02:50.550 --> 01:02:54.650 And that's something that's that's fairly close to my heart as well. 803 01:02:55.664 --> 01:02:58.565 Sometimes getting people to fund a project, they're like, 804 01:02:58.865 --> 01:03:05.710 but why would we fund it? It's not going to make any money. Or or academics, you know, my seniors would be saying, 805 01:03:06.090 --> 01:03:10.945 but why are you doing this and not just paying a software engineer to write the software for you? 806 01:03:11.825 --> 01:03:14.325 So or why are you just not using something off the shelf? 807 01:03:15.025 --> 01:03:19.925 So those are sort of topics that I that are close to my heart, and then that's why I ended up writing. 808 01:03:20.650 --> 01:03:21.790 So as PsychoPy 809 01:03:22.410 --> 01:03:23.310 gets launched, 810 01:03:24.010 --> 01:03:27.710 it pings a website, and that's what allows us to collect usage data. 811 01:03:28.095 --> 01:03:31.155 So we track how many unique IP addresses 812 01:03:31.695 --> 01:03:33.555 launch the application every month. 813 01:03:33.935 --> 01:03:40.090 And and that helps us a little bit to to go to potential funders and to go to 814 01:03:40.790 --> 01:03:44.490 my superiors and say, look. This is an important thing. We've got 815 01:03:44.875 --> 01:03:46.015 12, 000 users. 816 01:03:46.474 --> 01:03:49.135 And and so that's kind of an interesting 817 01:03:50.234 --> 01:03:51.535 topic in a way, 818 01:03:52.119 --> 01:03:57.099 But I don't know if that's something that that I don't know how to frame that as a question that you would want to include. 819 01:03:57.800 --> 01:04:02.195 You did a perfect job there. And I think that too that brings out a good point that 820 01:04:02.575 --> 01:04:05.715 as software becomes a more integral piece of 821 01:04:06.095 --> 01:04:14.440 academic endeavors, that I think that we are probably gonna start to see the paradigm shift a little bit away from the publish or perish 822 01:04:14.740 --> 01:04:18.759 or at least recognizing that developing software is a form of publishing 823 01:04:19.345 --> 01:04:21.205 as is evidenced by the fact that, 824 01:04:21.665 --> 01:04:33.210 a lot of these scientific open source projects are actually cited in academic works. I know that you've got a citation reference in the documentation for PsychoPy and similarly for things like SunPy and AstroPy. 825 01:04:34.635 --> 01:04:39.135 Yeah. That's but that's 1 of the interesting things is trying to find track those users where 826 01:04:39.435 --> 01:04:48.260 the citation count on that paper is, I don't know, something like 6 5, 600 people, something like that. But I know that monthly, there are 12, 000 users. 827 01:04:49.200 --> 01:04:52.945 So so, you know, there's a there's an order of magnitude discrepancy there 828 01:04:53.825 --> 01:04:54.625 that that is, 829 01:04:55.665 --> 01:04:59.365 that is interesting. And and presumably, it's that citations will follow on 830 01:04:59.745 --> 01:05:03.869 gradually later or people are using the software for all sorts of other endeavors. 831 01:05:04.170 --> 01:05:06.589 And when I mentioned earlier on about wanting to know about 832 01:05:07.130 --> 01:05:07.630 nonacademic 833 01:05:08.009 --> 01:05:09.950 uses, it's because in the UK, 834 01:05:10.655 --> 01:05:12.435 this, by many classifications, 835 01:05:12.974 --> 01:05:15.555 doesn't actually count as having any impact. 836 01:05:15.855 --> 01:05:22.650 So the UK counts impact of a scientific of a

of an academic endeavor as when it affects society 837 01:05:22.950 --> 01:05:27.210 outside academia or when it raises money. And because I don't sell it 838 01:05:27.785 --> 01:05:30.285 and because the people using it are academics, 839 01:05:31.385 --> 01:05:47.525 tens of thousands of users don't count as any impact, which is a so having 1 person using it in the clinic for for nonacademic use but to find patients counts a lot more, it turns out, than tens of thousands of people teaching their students. 840 01:05:47.984 --> 01:05:49.585 So so we've got these, 841 01:05:50.305 --> 01:05:50.805 broken 842 01:05:51.299 --> 01:05:52.279 kind of incentives 843 01:05:52.819 --> 01:05:54.519 in many ways, and and 844 01:05:54.900 --> 01:06:03.645 and those those are frustrating for me. Those are those are topics that I find interesting. Let's put it that way. This this comes back to a recurring theme that we've seen on our podcast. 845 01:06:04.585 --> 01:06:11.839 If you're using a piece of open source and you love it and you, you know, it's making your life a better place and you're enhancing your job performance, 846 01:06:12.539 --> 01:06:27.505 tell the right you know, the people who are in charge of that project about it. Drop them an email. It, you know, it it takes 5 minutes to say, hey. Thank you very much. I'm using it for blah blah blah. And it means so much to people like like Jonathan here slaving away, 847 01:06:28.029 --> 01:06:29.890 building this this incredible stuff. 848 01:06:31.069 --> 01:06:32.930 Yeah. Yeah. Very much. And and certainly, 849 01:06:34.109 --> 01:06:35.970 it while writing papers, 850 01:06:36.345 --> 01:06:39.964 like journal articles versus writing software, I I found it interesting 851 01:06:40.505 --> 01:06:54.305 when I go to a conference that I do get people come up and say, oh, you're the guy who writes sci fi. That's great. I love that. That's brilliant. And it's really nice. And no one's ever come up to me and said, hey. You're the guy who wrote that paper on adaptation to PLAD visual stimuli. 852 01:06:55.005 --> 01:06:56.385 And I really love that paper. 853 01:06:56.685 --> 01:06:58.785 So, yeah, it is it is motivating 854 01:06:59.165 --> 01:06:59.665 having 855 01:07:00.210 --> 01:07:01.349 individuals come up 856 01:07:01.730 --> 01:07:03.750 and and say that they like your work. And 857 01:07:04.450 --> 01:07:07.109 that actually, for a lot of open source developers, 858 01:07:07.605 --> 01:07:09.465 is I think what drives them. You know, 859 01:07:10.005 --> 01:07:17.385 they they like to feel that even though they're not being recognized financially or or in terms of, other 860 01:07:17.750 --> 01:07:21.770 other more tangible methods. Hearing people say they like your software, 861 01:07:22.230 --> 01:07:24.329 is definitely what gets a lot of us out of bed. 862 01:07:25.135 --> 01:07:34.435 And on that note, what are some of the best ways for people to get in touch with you and let them know about their use of PsychoPy or follow what you're up to and how things are going with the project? 863 01:07:35.170 --> 01:07:37.670 We've just recently switched actually to 864 01:07:37.970 --> 01:07:40.790 from from a Google group to Discourse. 865 01:07:41.170 --> 01:07:42.550 So Discourse is 866 01:07:42.875 --> 01:07:43.694 an open 867 01:07:44.075 --> 01:07:48.015 source forum product, and and we've now got a a forum 868 01:07:48.714 --> 01:07:49.775 on discourse.psychopi.org. 869 01:07:51.115 --> 01:07:52.015 That's been brilliant. 870 01:07:53.089 --> 01:08:00.150 That's that's something that I I hadn't realized actually how much we needed a forum rather than just a Google email list. 871 01:08:01.035 --> 01:08:03.695 So, actually, that's the best place to get in touch with the group. 872 01:08:04.715 --> 01:08:05.115 And, 873 01:08:05.515 --> 01:08:06.255 yeah, that's 874 01:08:06.715 --> 01:08:10.860 if you're writing your own open source projects and or or just wanting 875 01:08:11.240 --> 01:08:14.140 an online discussion, I would also recommend having a look at Discourse 876 01:08:14.520 --> 01:08:16.860 because it's a it's a pretty awesome package. 877 01:08:17.325 --> 01:08:20.145 Yeah. That's what we're using for our forum for the podcast. 878 01:08:20.525 --> 01:08:22.125 Okay. Yeah. It's it's, 879 01:08:22.605 --> 01:08:24.785 it's really neat. It does everything right. 880 01:08:25.350 --> 01:08:25.850 Yeah. 881 01:08:26.230 --> 01:08:30.330 And and from our perspective, it does things like if a user 882 01:08:30.710 --> 01:08:31.610 starts typing, 883 01:08:32.469 --> 01:08:49.610 a new post, but it looks kind of similar to a question that's already been answered, then they get a very clear flag saying, have you looked at this post? Because it's quite similar to what you've just written, which hopefully means that that we are having to answer fewer questions that people will find existing answers instead. So things like that is really neat. 884 01:08:50.425 --> 01:08:57.725 So with that, I'll move us on into the picks. And my pick today is the book, hackers, heroes of the computer revolution by Steven Levy. 885 01:08:58.110 --> 01:09:08.845 I listened to it a week or 2 back and thoroughly enjoyed it. It's a interesting way to get some of the history of early computing and how it, 886 01:09:09.305 --> 01:09:14.525 how people who were working outside of the intended frameworks of the hardware, 887 01:09:15.370 --> 01:09:20.750 sort of pushed the pushed the industry forward and introduced a lot of new innovations 888 01:09:21.130 --> 01:09:24.795 that might not otherwise have happened. So definitely worth a look for people 889 01:09:25.594 --> 01:09:28.895 who are not who who who who weren't there to experience it firsthand. 890 01:09:29.755 --> 01:09:34.175 So definitely recommend giving that a read or a listen. And with that, I'll pass it to you, Chris. 891 01:09:34.670 --> 01:09:38.050 That's a great book. I read it decades ago. Man, I'm old. 892 01:09:39.870 --> 01:09:42.190 My pick this week is a, 893 01:09:42.510 --> 01:09:43.970 an iOS application. 894 01:09:44.655 --> 01:09:47.715 Unfortunately, there's no Android version right now. It's called Castro, 895 01:09:48.335 --> 01:10:01.960 and it has really just made my podcast listening life a much much better place rather than using the model of you subscribe to these podcasts and therefore you will listen to every episode that they produce in order. 896 01:10:02.340 --> 01:10:15.830 It treats it almost like an RSS reader like Feedly or something like that, where you have an inbox and new episodes are presented to you, and you can tell it this podcast is just so good just just queue it don't even, you know, don't even bother. 897 01:10:16.210 --> 01:10:17.110 But for others 898 01:10:17.410 --> 01:10:31.010 you can say, you know, put it into my inbox and let me choose and you can say, oh, that that episode is interesting, add it to the front of the queue or the back of the queue or archive it and episodes that you archive don't just go away forever. You can go back and leaf through your archives 899 01:10:31.550 --> 01:10:42.575 to see if there's something you actually do wanna listen to. It's just great. The interface is phenomenal, and, and I really love it. And that's my pick. John, what do you have for us or picks today, if anything? 900 01:10:43.080 --> 01:10:47.099 I I think mine would would I'm just gonna go back to discourse. 901 01:10:47.880 --> 01:10:50.300 My my pick of the week would be discourse 902 01:10:50.835 --> 01:10:52.855 for for forums, and and 903 01:10:53.395 --> 01:10:56.055 discourse is a is a a a great, 904 01:10:56.435 --> 01:10:58.775 forum package, open source, 905 01:10:59.860 --> 01:11:04.360 really well written. It does everything right, and they're also very responsive developers. 906 01:11:05.060 --> 01:11:05.960 When we found 907 01:11:06.385 --> 01:11:12.885 some issues and had questions, they were they were really good at getting back to us as well. So for any other open source developers 908 01:11:13.800 --> 01:11:17.740 or people just wanting a forum in general, I would really recommend you check out Discourse. 909 01:11:18.520 --> 01:11:20.860 Yeah. And they've also made it really easy to 910 01:11:21.355 --> 01:11:30.360 get it installed and upgraded and self host it, as well as having a hosted offering as if if you are not inclined to run your own server for it. 911 01:11:31.320 --> 01:11:41.804 So we really appreciate you taking the time out of your day to tell us about your work on PsychoPy. It's definitely sounds like a really interesting project and I hope you enjoy the rest of your day. Thank you very much.