 Hello and welcome to Minecraft AIs, 28th of August, Developers Inc. So we've had a very busy and exciting week here. The good news, not to bury the lead, is that the SJ201s, the custom hardware hat that we've made for the Raspberry Pi 4-based Mark II dev kits, has been verified to work and this is awesome. So we've got apparently nine working boards now, and we're going to get those over to the rest of our dev team and start spinning up the enclosure and other related skills necessary to turn those from a donor board into a Mark II. So that's really, really cool. I'm very excited about that. In other news, well, I don't know, what is the other news? Chris, how about you tell us what's going on? That was kind of unfair. Sure, so I assume that means you're asking me for my status. Yeah, yeah, exactly. Okay, so I spent the last couple of days getting all the coding changes made for the schema change. That is all done. I'm running a few final tests now and I expect to call that done before I sign off for the day. I also went into Confluence and changed the schema documentation to reflect what the current schema looks like. So that's not very exciting, but that is certainly where I am. So then on Monday, I will start with the script that will copy the files over from the temporary, the temporary home on the Selene server to the DAZ storage. So it's kind of, it's kind of plugging along. Okay, great. No blockers. Great. Also, Ken is totally right that was completely unfair of me, but Chris is also going to be tackling the doing the actual skill enclosure development for the SJ201. Yeah, and on that, I know we brought up at a previous dev meeting, but, you know, do we want to talk about skills versus enclosures versus, you know, the architecture of this for this project or we just want to do what we've been doing and fix that later. So let's go on using the model that we have right now. We will, we will address this later on, but it's not broken. So let's not mess with it. Oh, yeah. Okay. So, so then my question on that would be right now there is a mark two skill. There's a mark two pie skill. There's a mark two enclosure. And there's a mark two. I think there's two different mark two enclosures because of the different guillies. Possibly so anyway, we should walk. I'm glad to make some decisions that we've kind of talked about but haven't landed on yet like we're going to use and that kind of stuff so I can bring up all the right things. Right. That's a good point. We should have a discussion about that. But I think that we should go ahead with the enclosure that is that, you know, we're most comfortable with maintaining which I think is the key one for now. And, and, you know, we should be clear with the community what our intentions are and the intention is that the, the re speaker based mark to it's going to be deprecated in the future. As soon as we can get to a one, you know, boards up and running and we, you know, we know that they're solid. So, so yeah, I just, you know, I think at some point we're going to stop maintaining that, you know, that skill enclosure, and that'll be deprecated and we'll just focus on the mark two based on the sj 201. Okay, build a completely different skill for this mark for the mark to sj 201s we're going to have buttons that we don't have, you know, on the current enclosure and that kind of stuff so maybe might make sense to build a little different thing we can just look at the things. Okay. All right. Excellent. Thanks, Chris. How about Ken. How about that. I'm just making a note real quick. Okay. So, yeah, so it's been a good couple of days I went back to the code and refactored a bit to expand the capabilities of the channel the channel being the connection between the upload server where the database data is selected and the training server where the actual training is done. It was just very straightforward here. Here's a archive of training of a training data set. Go build it. Now it's more. Here's the hyper parameters I want you to use. It could be a partial or a full partial meaning. Don't bother looking for new training data use the training data you got just use these new hyper parameter. So, so I have those two paths. You know, full train kind of partial train with the existing data just change hyper parameters so that in the future if we want to use that capability we can. I then laid out the schema for the precise model database and went to put it into the Maria DB and then learn that I didn't have the group password so I chased that down a little bit but I now have that so I should be able to get that going and so by Monday it should be saving all the runs in the database with all their hyper parameters. And in this case hyper parameters include everything so it's also going to have a reference to the manifest file and it's also going to have the value that was sent over or that the that the training data set was built with with regarding the split the ratio between training and validation. So that's not fixed so it could be 80% 75% whatever. But the point being that there's enough data in that database that you could reproduce the model. Because I don't know that I want to actually save the dot pb and dot pb program files maybe I could store them in there too as a blob I don't know if it's necessary. What's more important about this database is that the hyper parameters and the test results of a model built with those hyper parameters are stored. So later you can select all the hyper parameters cluster it and have a half a chance at figuring out what the relationships between the hyper parameters need to be. I have discovered that the hyper parameters tend to perform better when their prime numbers. So here's that. That seems random. Yeah, no it's actually common that primes tend to work a lot better in machine learning algorithms, because it doesn't give it the chance to take cut corners down low. And it like you know say well let me find the closest you know that's divisible by two and kind of an that's not going to happen. It's just something I've learned over the years anyway. So yeah so by Monday we should be having that database getting populated. Hopefully, I will get a mark two on Sunday. Thank you Derek, and I will get right on this yes no bug. Because it seems like it should have been addressed a long time ago, but we probably didn't have a lot of use of it. But I assume as we start scaling we will and so it's going to become more important. So I'll look at that now I've also been helping out on the matter most of the forums trying to pick up the slack wall guesses out until Monday so I've helped two or three people now with wake word issues and I'm not sure if that's a result of the 20 the 20 point X release. Because some of these people are just trying these things with really strange stuff like the guy today was using a webcam microphone. That wasn't working too well for him and then the other fellow I was talking to was running on a pine phone, which is one of its near and dear to my heart I know the rock chip very well. The other phone, unlike the pine 64 is not the system on a chip and it's actually underpowered so doesn't surprise me and some of the other community members actually pointed out that size is consuming over 60% CPU on the pine phone. And then they were like, Well, we really should get that full request for TensorFlow light and I sympathize with them and that is on the radar but a little bit couple of bands out. Awesome. Thanks for that update. Derek. Good. Just a little history can we did explore a rock chip based design. Before we before the pie for came out because the pie three were like that's not quite powerful. My four is better than the rock chip but the rock chip system on chip is not bad. I've actually, you know, forwarded Android to it around slide. Yeah, yeah, it was interesting like there was like a four plus two kind of thing where they had a quad core plus two other was like a, not quite a six cores but it was interesting architecture. One is really good because it has everything built in HDMI to usb is it has all the audio and video processing built into it. It's pretty powerful and then you don't have to worry about the whole interfacing the components because it's a system on a chip and they do that for you. And then it has, you know, the specialty CPUs. There's the low power one for you know background stuff and the high power ones before that stuff. Yeah, I think the pie four is a better choice just you know. Yeah, plus everyone, you know, it's got the broad community support because it's so so popular. Yeah, so what I've been up to I've got so I got that device off the kin. I also sent off an enclosure to Kevin that had an audio chamber in it. I think he can it's a late the laser cut laser cut enclosure that we're going to start off with doesn't like take like maybe an hour and a half to put together all together, whereas 3d printed version can take several days of printing to get all the parts. So it sounds like what they success we've had that I'm going to need to make quite a few more of those. Thankfully, I've already got to done. So we'll get one assembled and put together for Chris, as soon as I get boards from Kevin so he's going to ship those. Pretty soon I think, maybe even tomorrow. So when I get those in. Yeah, I'll get one packaged up and send over to Chris as soon as I can. I've done a little bit work on what will be the 3d printed design. And a lot of my time in the last two days has been dedicated towards some slide deck stuff for the fundraising effort. So yeah, no blockers for me I think I've got everything updated. I think I did update that I only just did it today can I updated the the audio services issue on first boot up to attack you and I moved it into the market to So, yeah, yeah, I saw that. Yeah, I got both. Cool. Hey, Derek, we're going to have to talk a little bit about screen design as well as we'll have the horizontal screens and the vertical screen. Right. The existing screens will have to change. Yep. Yeah, it's going to be a bit of work. I don't know if in the short term we can do something like just stretch it or, you know, something quick and dirty. It'll make my eyes bleed. But, but yeah, yeah, we need to fix it for real eventually. That's it for me guys. Okay, great. Well, this seems like a lot of good stuff happening this week. So we're right now in the middle of our sprint we're going to continue next week with the same sprint. Right. But for Monday, let's go through and make sure we've updated all those tickets just so that we can check our progress and we'll review that on Monday. Is there anything else that people want to add before we sign off here? No. All right. Well, when things are going well, I guess the meetings are short. So, thanks very much. Enjoy your weekends. You know, whether you're working on the mark two or not. And I'll talk to you all on Monday. Have a good weekend. Have a good weekend everybody. Bye.