GeoSci.xyz SimPEG, 28 Apr 2021

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: SimPEG Meeting April 28th

Description

Weekly SimPEG Meeting from April 28th, 2021

A

Hi and welcome everyone nice to see you all hope you had a good week the last week. um The meeting notes are posted, as always in the same location and on the slack channel. There add yourself to the quick reports if you need to, if you got anything to chat about.

A

As far as agenda items, I don't have anything particular that I'd like to talk about outside of just um friendly reminders about looking over full requests. Phone call requests things like that.

A

I'm probably going to be working a few. The next couple days on getting some of the older ones in. I put a lot of time in to do that really quickly, because there's some like definitely outstanding ones that need that should get pulled in shortly.

A

um Yeah, so probably I might be reaching out to a couple of you guys just double checking on things about like where some, what the state are some branches and if there, if there's anything else that you needed to happen um just some of the older ones that are, I know, are basically already like you're like tebow your your branch from the pgi stuff, I think, is pretty close to being ready to pull them. It's it's essentially ready.

B

Yeah since december 4th I've only been like merging mastering yeah yeah this table. It's been stable for many months now after that, so yeah.

A

I realized there.

B

Is like recently, there is like a a taste test failing on most of the pr, and I'm not sure where that comes from.

A

um Oh so I think I I addressed it on the most recent one, so every if you update to master, I think it's probably fixed that, but essentially one of the links was broken.

A

Okay, the uh so in the documentation, the we were pushing we're pointing to we're pointing windows users to a download for, like some msvc 2008 redistributable package that is no longer available, so the link was broken.

A

So I just removed that since it was no longer available- and it was mostly a issue that came from installing discretize anyway, so it didn't really make sense to have it on the synthetic documentation anymore. So I just removed it. But that should be that's uh that's likely what was doing it, but I will check as well.

B

Okay, well, I just merged main again into the pgi branch, so we'll see if that pass. Okay,.

A

Thanks joe any other kind of agenda items things, anyone has that they'd like to talk about later on after we go through quick reports,.

C

um I mean one thing is just we could come up with a a plan for em1d.

C

I took a look at it the other day and just kind of just just to see where we were at and what items need to be done, and I think we just kind of need a little, a little group of people to to kind of take responsibility and try and move that forward in the near future. I don't think there's too too much to do it's mostly organization, but just come to consensus, uh implement it and kind of get it in so uh yeah.

C

That would be something to just come up with a little bit of a plan doesn't have to be like right right now, but just kind of in the next like month or two to put some work into it.

D

Dev in terms of that I had a quick comment. I think it was just set up like some sort of hackathon or something like that uh just set a week and uh just get it done. I think it feels like or yeah I I feel sorry about that as well, because I'm not really putting energy into it. So we set a goal and set a deadline and just set a hackathon or something like that for a certain week, yeah, let's, let's just get it done because yeah.

D

I just feel like keep like feeling. Sorry about that, and I just not doing it so yeah if you just set the deadline and let's get it done. That sounds like.

C

A better that seems that seems best, because I did the same thing. I got distracted by other stuff. It just needs uh yeah. We there's some organizational stuff. Nothing seems particularly difficult to just need a couple decisions and then to fill in some some tests so that it passes the the coverage and yeah. We just need to hammer it out over an afternoon.

C

Probably so uh I don't know I'm pretty flexible with my schedule, I think so soggy I don't know if you maybe want to dictate when we do that, because it seems like your schedule's a bit more strict.

D

I'm pretty flexible in these days so uh either like this week or next week. So I think, uh maybe at the end of next week, let's get it done and uh setting the deadline by the end of next week. Yeah.

C

Well, do you want to just do this friday afternoon and then we'll see how far we get that.

D

Sounds good actually, let's, let's work on friday, aft and then set up a small hackathon cool.

C

Sure um 1 30., nope, okay, so this friday at 1 30 we'll get together, we'll figure it out.

C

um Is there anyone else that really wants to be a part of this.

A

I would like to show we can get it done. We can just like put it out for anyone that would like to join. That's good, okay,.

C

So then I'll schedule, uh I guess I'll schedule a meeting.

A

Sounds good, okay, okay, then! So then there's not any other agenda items and move on to some quick reports. I'll go first here. So I've started testing.

A

I had to, I realized, I had to add something for the deep prop branch to make it actually like work, exactly how we wanted it to with simpeg, because we often use it as a dictionary key.

A

We often use meshes as dictionary keys to things, so it was originally failing, and that was because I had also implemented part of this like an equality test for meshes, so it would go through like an under like a double equivalent method, so you could just compare mesh equals mesh and compare if the two meshes were equivalent to each other.

A

That is not something that currently happens, but as a side effect, because I implemented that I also had to implement a specific hash operation, so it would work as a hash key to like dictionaries.

A

So there there's two requirements for this for two objects that compare equivalent that are equal to each other. They have to have the same hash key.

A

The opposite is not true, like two objects could have the same hash key, but they could be not equal to each other, but if two objects are equal to each other, they have to have the same hash.

A

So so because I had implemented that, like equals method, equivalence method, we also have to implement the hash key, so that kind of makes the what I did done recently. Is I implemented that in a way that kind of it's it's?

A

It has the potential to be misused, and but I'm not other, because our meshes are not generally like completely immutable objects. We can do things like change their origin.

A

um You know, I think we can change their reference like their orientation. Things like that. Those, like there's certain properties that we can change once a mesh is constructed.

A

However, we don't generally change them once we start using the mesh like with all. Can we set up the mesh and then we start, and then we do things with it, so the way that it works like this hashing function works right now is it creates this identifier right? That's based off of the properties of the mesh in a controlled manner, and then it just stays like it sets that as a like, that has to be its hash. It only does it once so.

A

That's the other thing like the hash has to be has to continue has to be the same throughout the entire life of the mesh. So that's why I've done it that way. However, that does mean like by bonus. It has the potential to be misused. Let's say: okay, you do something that causes its hash value to be generated, but then you change like the origin of the mesh.

A

It could have a few unintended consequences.

A

The only the point is, though, like this hash.

A

In order to calculate it, it essentially has to make a copy of the necessary items that describe the mesh. So things like it's, the cell wits the origin, the the reference for a tree mesh. You would have to make a copy of all of the uh like the levels.

A

um All the the cell identity, like the cell indexes the things that describe that mesh, because.

E

It's essentially the.

C

A

Way, you have to do it in the same way that you test for.

C

Equivalence that that could take a while, though, that operation like, isn't it supposed to be something really simple that happens under the hood.

A

That's why I right now, I have done it where it just doesn't want and it caches it.

A

So, the first time you ask for the hatchet does that, but then every other time it's.

C

Fast, okay! Well for actually for the tree mesh in some ways, it's not that difficult, because you you have a step where you use you finalize it. Yes,.

A

Yeah, but you can still with the tree mesh you can actually still like you could yeah. You still shift the origin around. So the origin is the only thing like weird, but I don't that's the thing it's like, I'm not sure how often that comes up in practice, whether someone like because it seems to me what we create the mesh and then it's done by the time we actually want to use it.

C

You would think, but then somebody might somebody always finds a way.

A

So then either then my point is the options are just to not have an equivalent method. So that way we can just use the hash the same way. It was like because.

A

Method, it just generates a hash based off of its unique id like it's placed in memory, something like that and then the built-in equivalent thing. That's why you can use like user-defined classes as hash objects. So the question I'm basically asking is like: is it useful to have an equals method on when comparing meshes.

D

In a certain level joe, is it isn't it useful? Let's say like you, don't you don't really care like all the byproducts, the mesh, but if the core property needed to generate the mesh, if you can just check that, it's same then say that two meshes are equal.

D

Isn't it like, like a better than uh I don't know, I think, having something like that seems like a useful. uh Useful function like you've got two mesh and you just want to check whether they are equivalent. uh When would.

A

You seem kind of useful yeah. Actually I don't know, I'm not sure, that's why I asked if it's something like if it's something that's useful, because if not I'll, just remove it and then it'll be just like it was.

A

I was more so thinking of trying it like there's a with properties. You can test if two things are equals to each other. So that's why I put it in there originally, when simply removing removing the properties branch. I was trying to mimic behavior that it did before.

E

My instinct is is that if I'm comparing message meshes, I'm probably just going to be looking at a handful of specific properties, but that that's going to be on a case by case basis, um I don't think I've ever used an equals method on a mesh. I don't know if others have me neither yeah.

A

So then, if that's that's the case, then I'll just like get rid of it, it'll be fine. There's a very long-winded way of me asking that question.

D

You don't necessarily need to deprecate it. I guess, but uh if there's something like that and then, if that's simple enough to check the base like checking the base properties.

A

If you implement like a double underscore equal in order to use it as a key, you also have to implement the hash like the hash function.

F

D

I see that's my point so yeah yeah yeah. I don't really see the major kind of.

A

Benefit of having it, what I will do is I'll just change it to like dot equals or like it'll. Something like that would equal equals won't work, but you could also you could always just like okay mesh dot equals mesh to have that return, something or at least leave us there for that.

C

um Okay, yeah, I I think I ran into one it's probably related to this, but I tried to generate the tutorials with the d prop branch and discretize and found that I guess it's. The projection matrix for each receiver said like object, is not hashable, so I'm guessing. That has something to do with what you're yeah you're working.

A

You got it so it'll. It should work now, though, because I pushed one that has hashing working so the next time I'll just erase it and it'll still work like that.

C

Okay, oh I see yeah.

A

um So the other thing I finalized that boundary condition branch on discretize I've got the testing um testing 100 of the difference. So if you look at the test, you should see how it works. If you like to cure like like see you, if you look at the test, you should see how to use it to do things um if you're curious at how some of those little internal things are done, you can check on the inside, but the tests should show exactly like how you would want to like how I would post use it.

A

So there's I've got uh convergence testings for like a poisson for sound like problem um as well as like an em problem, so curl, curl e plus e equals some right hand, side function.

A

And as well as like, yeah,.

D

So joe, is it general enough? Let's say what we typically do: you do a weak form and then you separate the boundary condition that set it to zero. Usually like that natural boundary cases.

E

Something is zero, yeah.

D

uh But the like, uh can I actually set the boundary condition in an unnatural place? Let's say, divergence of e divergence of j is equal to zero, for instance right, so you can still actually put set up a boundary condition on j on such a case right so which is not natural, but it is possible.

D

Is it general enough that I could do that types of stuff, or it is somewhat limited to that kind of that that that separated boundary condition that you uh separated from that in the wig.

A

Form it is limited to that weak form formulation for the boundary conditions. That's the only place like that's the only place that we can really apply the boundary conditions to it, because otherwise, that implicit assumption that, like okay, you say that natural boundary conditions are going to thing. So when you do that weak form that you know transfer that higher order integration by parts thing, if you disregard that boundary term, that means that you are assuming that that value on the boundary is zero, but that that is a part of the assumption.

A

So, even if you like- okay, if you think of the nodal dc problem- and you say, okay, j is equal to zero on the boundary for the nodal. That's the natural boundary condition you can't like if you, even if you like, uh set the value of potential at one of the boundary nodes.

A

It still implies that j is zero on the boundary. That's true.

D

That's true yeah, so the reason why I asked uh so I was actually I'm gonna talk about that, but I was working on the theater's code recently bringing into simpac and he actually used a slightly different boundary condition, so what he does what's called perfectly conducting boundary conditions, so he set the electrical like. He used the eb formulation, but he actually set the electrical field on the edges to be zero and also that the outward component of the magnetic field at the boundary.

D

So it's like a face mag to be zero and it seems like it works a bit better uh compared to our code so and that's actually unnatural boundary condition see. I think that basically using the finite difference, that's somewhat intuitive. In that case, like doesn't really matter about the weak form. You just set the boundary condition. So I yeah, I was just kind of curious. What's yeah.

A

It's kind of effect it kind of comes from the way that the boundary conditions work with this finite, like weak form of the finite volume kind of thing, and that we that's kind of the only real place that we can put them at is that on that boundary integral like we can only you can use some of those functions to approximate the values at the boundary.

D

Right, I see I see I see so we can actually still find a way through that kind of injected information into that kind of weak form, boundary, yeah, yeah, okay,.

A

Cool, so the point is like this: this, the branch adds all the necessary items that you can use to probably build up those boundary conditions, so the the fun mental parts is that it I had to implement everything averaging to faces. Okay, so.

E

A

Define things like you know, boundary outward: normals are necessary uh right boundary locations uh right right.

D

Like it's all built.

A

Off of those operator matrices.

D

Right and that's right, I can ask one more question, so I'm actually like starting to work on the flow problem in these days and that the boundary condition is driving force and one of the tricky boundary condition is like your boundary is actually at the internal mesh. So it's not actually at the boundary. But let's say you got a river, then you want to. You want to put a dirichlet at a reverse location.

D

So, like a is it kind of possible, the current structure is possible to kind of boundary conditions in.

A

um It should be the kind of structure like you should be able to follow the general structure there. So what's our formulation, you'd also have to write you'd have to you'd, have to be able to go in and put in, like active cells, we have to be able to have an active cells pass to like the mass matrices that we make as well, because, like we're talking about for, like the river flow of the river channel stuff right, you're kind of almost excluding values from the pde solution. Right, that's true! That's true!.

D

Yeah, that's actually how that the mod flow and that types of yeah.

F

It's actually simple.

D

In that case, because they're not using the staggered grid um but yeah.

A

D

A

I from what I understand like we would have to go through and put that part of thing into the those magic mass matrices have like an active cell index. That's where, because that way like it does, that volume integral, but then the the boundaries should be pretty similar like the the structure is the same right. We just need to find what faces around the boundary.

A

Like what nodes are on the boundary and.

F

Their outward normals and then kind.

A

Of structure is the same.

F

Gotcha, that's great. It's exciting.

A

And I like it, I I did tests on the test convert like there's convergence tests for those pdes, the simple pd's for on tree mesh, tensor, mesh um curve and linear mesh.

A

So I hit lindsay and I have to get that get together to see about the cylindrical meshes and see how they how it translates to that. But I think it's in a good spot, at least for right now to put it in, so we can use these building blocks and the other thing is if we find ourselves having to like create the same matrix. A bunch of times like I have a few automated, like oh cell, gradient, rob and boundary condition thing that we talked about before one for like the edge divergence robin thing.

A

So if we find ourselves having to like create those type of matrices repeatedly like in simpeg, like oh I'd like to do this boundary condition a bunch. So we should probably think about like okay. Well, let's make that a specialized function and discretize on that mesh. Instead of having to call it 20.

E

Or write it 20 different times.

A

But it's all, I think the parts are there to construct it. So please take a look. We can get it in.

D

Yeah I'll absolutely take a look uh yeah, it's very exciting, joe because yeah. I think that could make a lot of improvements in our both dc and en code. So.

A

It makes it really easy to implement, like those robin type boundary conditions on the nodal problem for dc like it's very, it's actually really simple. Like it's a conceptually simple thing: it works out pretty well, um you know it's just hard to implement dirichlet conditions on the noble problems or you're not going to do that. Well,.

D

You know, can you do uh robin on the uh maxwell's equation, so, like an in mt problem, I thought I think that the similar idea could be nicely used because uh like what they call impedance boundary conditions. If you.

E

D

Imposing electrical field at the edges, we can just put the the like: a ratio between heat field and uh magnetic field. I guess so.

A

I think of that kind of plane. It could be easy. It comes with how you formulate it like if some some of the boundary conditions are easier to formulate on like if it's discretized in a certain way right, yeah dirichlet conditions on cell centered or self-centered solutions to poissons are easier to implement dirichlet conditions on, uh let's say, electric field. If it's solved, that faces is easier to implement um okay.

A

So it's just a thought about like it makes a difference like where the nodal where the values are depending on what boundary condition you want to implement all right.

D

A

So, okay, the other things really quickly, sorry uh edited and pulled in that static, utils branch on pr I was able to get the plot pseudo section function to kind of match and extend like a little bit of amalgamation, so it works devon. The way that you had it before, like you, can still call it the same orders. um I just made certain things, keyword, arguments and then put cases in so it works the same way you did before and it works the way that you're trying to do it.

A

Now, I've added a few of those tests back and there's a few more errors that I taught by adding some of those like tests that have been commented out so and then lastly, they're going to be giving a presentation at this pearls, computational food dynamics, thing.

A

In may, 17th, I think, is a presentation date, but I have to have it in by next friday, so I might be reaching out to a few people. It's it's mostly detailing uh desk and up and parallel operations and setback that we've been working with, so I'd be reaching out to a couple people, okay. Moving on from me devon. I see you're next.

C

uh Yeah I mean we, so we just went over the em-1d stuff. um I think I should start continuing with the doing the api docstring stuff for discretize, um and then it sounds like so you're making a lot of changes to to discretize in terms of the the deep prop stuff and this robin boundary conditions, branch, and uh so one thing I thought would be useful- is to regenerate all of the tutorials on simpeg and ensure that it kicks out all the stuff that it's supposed to. So uh I'm rebuilding that from scratch.

C

Right now on the robin boundary conditions, branch- and it sounds like do you want me to also do that- for the the d prop branch yeah.

A

So the robin one just adds on functionality. It doesn't change anything it just.

C

Adds more functionality?

C

Okay, so that's not going to be a problem, but the the d prop one is the one we really want to test yeah yeah and you fix that hashable thing so I'll see if that that fixes it. uh If I do run into any issues, do you want them listed in in the port like in a do? You have like an open pr for dprop or something and then I'll open, pull requests for that there on discretize, so yeah. If I find any errors I'll uh I'll put them in there.

C

um That that's basically it just do some not fancy stuff to to test things and add doc strings that that's about it.

D

That I, like that's, extremely important. I really appreciate what you've done, because that I actually worked on the transform and I basically started from your tutorials and that was actually pretty straightforward to generate a new material like basically same but different examples. So it's that's extremely useful.

C

That's that's kind of the point. I I'm finding a lot of people just go through the tutorials. They they copy it out and then they just change stuff for their own. So most people do anyway. So thanks.

A

Soggy and dieter who's not here, got some interest yeah.

D

I participated. The transform uh conference uh gave a tutorial, I thought actually it went well and that it checks the number of views. uh Hours actually was the so like so far the best and that that that transformed so I felt like it was actually useful title. I I think it kind of worked, I guess uh yeah and then, on top of that, I sort of like guys.

D

I didn't really code in these days, and I was mostly writing and running the codes. I was just kind of tired of that, so I was looking for something technical and new, so I kind of thought that so d is bringing this emg 3d into simpac so, but I think it wasn't really ready to invert and then like he actually got the jt back in a sense, but that wasn't sort of like there were quite a few things that needed to be adjusted to be used in syntax.

D

So I I worked on that problem as a hackathon. I didn't really tell anything about that in that the transform hackathon stuff I just set it up. The local hackathon is motivated so.

F

I actually spent.

D

Quite a bit of time- and actually you know- was busy so actually I was sorely working on it and then just asking a few questions to either uh and that kind of worked because we got robotic. I don't know like 12 hours, two hour windows that I can ask questions to either.

D

So I what I ended up doing so I uh developed the jvac function in in their side, so theater's side so like uh so what we typically do, this order test agilent test, sometimes actually people don't know, steeler, actually not recognize how that thing is working. So I kind of developed that I tested his gradient function and developed a new jpeg following what he has done, because it's basically a transpose or adjoint operation of his gradient or jt bag.

D

So I developed that on his part and then I developed the assimilation class in simpex side, so yeah. If I put a link about the curve note that I I wrote a small summary about what I have done and there's a repo that you can check what that looked like and uh yeah.

D

So I kind of did that and uh what's that, what I thought actually cool is that so I it's a recipe, so I know once I developed that jt back and jeep back if that's test is working, I have pretty high confidence that that inversion will work. So I was actually able to just run the inversion and got an interesting like a result. I can actually show quickly.

D

Oh uh can you let me share.

F

D

Yeah, so this is the summary article that I kind of wrote and yeah if you want to read that feel free to read, but at the end it kind of shows. This is the miranium example, so that this is the like a c and that's the c sediment. I think there are quite a few sources and receivers.

D

That's the true and that's the estimated so yeah. I think it worked and there's a couple of uh links that, if you're interested take a look- and uh I thought that's quite exciting- because what theater coast does like it is really designed to handle large solving a large system.

D

So, whereas, like uh sort of our trick in these days, is just making the mesh small enough by using octri code and whatever tricks, we have to solve a relatively small system. So I think this could kind of open up a little bit different avenue uh of uh actually solving a large system like millions of like millions of degrees of freedom uh in relatively quick time and uh what's kind of nice is it's easily paralyzable. So every like each solve is different.

D

So like the current parallelization in his code is by source and frequency which could be uh kind of nicely paralyzable so uh anyway, so that's uh that's what I have done and uh yeah I think uh theater is coming back in a couple of weeks, is in mexico now, but uh yeah we'll.

F

Kind of revisit and sort of finalize uh that development anyway, so yeah, that's, but that's what I have done.

A

Cool I had a few questions about. Like the you're talking about I read through earlier, um you had complex data types, it's just kind of like it's definitely interesting, and I mean we. Obviously we handle them differently, normally in synthetic, so I'm just kind of curious what you've been what you've noticed, because it's also I'm not as well.

A

um It's not as intuitive to me how you handle noise on a complex data, value.

D

Yeah, so it's so actually like if you joe, if you take a look at that repo, I generated as like a dummy, complex data type, because the current data object assumes that data should be afloat so, but I think the other than that simpac is pretty general to handle the complex data so that inversion module is general enough to handle the complex data and the tricky part is the the setting the uncertainty.

D

So if you take a look at that document, I wrote a small kind of section about how you can set the uncertainty for the complex data and, what's actually nice about that, you can at least save like a one solve, because what we're doing we break apart, real and imaginary and then solve or like a real and.

F

D

uh When we're solving a inverse of some sort right hand side, so our like our definition of data is real and imaginary. So to do that, we actually solve twice, although you factorize once, but that is still cost so well, I mean not quite.

A

The field itself is complex, so when we do the forward operation, we just take the like. Sometimes we take the imaginary, sometimes big real part of it right right right for exactly like it's like not as and that same kind of thing, I think happens in jvec, um but jt back, it's not really as efficient as it could be right now, because we it's there's a bunch of lazy things in there. It's kind of like almost reverse like it needs to.

A

We need to accumulate a bunch of stuff and then do one solve before we and that doesn't often happen in our like em problems. Right.

D

Now right, so I think that's where I think well, that's. That was my point. The four problem doesn't really matter, but the when you're solving jt bag, you actually solve twice for real and imaginary. So that's that's what I meant you solved twice.

A

Yeah, we don't need to be doing that even right. Now, like we there's, no reason, we need to be doing it like that. It's just been how it has been implemented. It's the same reason that we solve it.

A

We solve it once for each like different type of receivers like if you think about the mt right now we solve it once for, like oh, the e, the x x receiver, the x y receiver y x, and we do that kind of j t vector each time, but really we only need to do it once we need to accumulate all those values into one vector and then do it.

D

Right, I I think, if you are actually kind of decomposing, that your data as a real and imaginary, I, like I, I thought about it quite a few times- there's no way you can actually fade to solve twice. uh That was my kind of feeling I don't know like there's, probably a.

B

D

But uh uh yeah, but if you actually set up a complex data and then it simplifies that problem, you just solve it once.

A

And I guess that's still technically a double solve because it's like.

A

Two vectors right like on the back end, but yeah.

D

Right anyway, that's uh very gory.

A

Details um well, then, I think we've got lindsey up next.

E

Thanks um yeah just a couple quick points, so I started uh or soggy started the pull request um for the governance documents in the simpeg community repository um and so take a look, feel free to add your thoughts as like a pull request review. If you're happy with it, it would be great to see like the approve.

E

If there's things you think should be changed um like feel free to withhold your your sort of approve until you until you think it's in shape, and we can sort of chat through that if, if it's worth sort of having a another conversation, a lot of it's based on the jupiter uh governance document, the scipy governance document are sort of the two that I looked at to draw some of the language because some of it, you know a lot of people, have already thought about these things.

E

um They're all also um licensed in the public domain. So we are free to reuse those thoughts, um but I did tailor some things to be a bit more specific to snpeg, like we've talked about in the in the governance meetings in the past. So take a look um feel free to add your thoughts. uh Questions concerns what I've tentatively suggested and I'd appreciate.

E

Folks, thoughts here is like a two week: ish timeline to request input from and if everyone within that two weeks has approved or if there's minor comments that we go in and address, we go ahead and merge. If, but if there is sort of any um big concerns we can, we can definitely hold off on that. But I was thinking you know having a little bit having a bit of a timeline to say this. This should be done unless there's any major concerns um within two weeks.

E

So the only reason I think that we would do well. um If, if folks feel like you need longer than that, then we can plan for a longer timeline, um but does that seem sort of reasonable okay, because I think dieter will be back by then too so he'll have a chance hopefully to to take a look too.

E

Cool, um I put a link into the poster that uh joe john and I gave at the eige meeting yesterday morning um and I dropped. There were a couple links that stood out to me um from conversations and things like that. So I dropped those in this seminars channel. I think um on simpeg some cool projects, there's one for an open source hardware for a resistivity meter which is pretty cool, um and then there was another one that stood out to me.

E

That was um collecting data for agro, geophysics applications and sort of trying to provide a website where you can locate. All of these different uh types of data sets so they're not hosting the data right now, but it's basically like a metadata um search, which is pretty cool um and they're continuing to grow that so those those were two that stood out to me and I think all of the posters will be available online afterwards, is what um sarah sarah gary was the one who who invited us.

E

So it's good to catch up with her a bit um joe. Was there anything else you wanted to raise with that workshop yesterday,.

A

um I I don't know just that, it was kind of I enjoyed it. It was fun they used gather town. I just went back in. I was like oh, I can still get into the poster session. If you guys are curious. What it looks like.

E

Oh cool: well, if you yeah, maybe you could drop a link in if, if people want to check that out.

A

I'll drop a link to it really quick. If anybody wants to jump in and jump in, there should be open to everyone but see. Look, you can go. Look at the posters.

A

It's kind of weird.

E

um Yeah, so it's good and then the other one that I have uh just on the list is that the iaga um conference there is an abstract deadline on the 30th, which is very soon, um but there is an em session and they're short abstracts they're short abstracts. uh There is an em session, uh so if you're interested in submitting something, um I'm sure there's other sessions, it might be worth taking a look. um If anyone wants to do something simple related feel free to um to share that, and you can solicit some co-authors so.

E

D

D

Lindsay you're organizing that session right.

E

Yes uh with your other folks, so it should be a good one. I'm looking forward to it um it'd be great to see some simple stuff in there.

A

Exciting, that's good, um so I don't quite have everything else or anything else to bring up um anyone else. Does it'd be you're, welcome to chat through it right now, if you'd like, um if not I'll, see some of you guys on friday afternoon at least a couple of you guys so yeah hearing no objections.

A

Nice to see you guys I'll, stop the recording.

B

A