00:54 <shrit[m]> rcurtin: I was thinking of resolving the arma::norm issue from the mlpack side by adding an sfinae, but it does not seems to make a lot of sense since the Evaluate function has an overloaded implementations depending on the Lmetrin configurations. I think it would be make more sense to have an if else block and call the arma::norm depending on the type of the inner element, what do you think?

00:55 <rcurtin[m]> no worries yugansharora01 (yugansharora01); we do our best, but we have to balance helping and writing code---and also all the other things we do in our lives, since for all of us this is a free time activity :)

00:56 <rcurtin[m]> shrit: if the conditional uses a compile-time value, then the compiler can optimize it, which seems fine to me

00:56 <shrit[m]> I am not sure it can be compile time.

00:56 <shrit[m]> because if else need to be constexpr

00:56 <shrit[m]> which is not available before C++18

00:56 <shrit[m]> 17*

00:56 <rcurtin[m]> `std::is_same<>` will work

00:57 <shrit[m]> I will give it a try

00:58 <rcurtin[m]> LMetric is a really core class, so before you open a PR, just run kNN in naive mode (from the command line is fine) some data to make sure the speed doesn't change

00:58 <rcurtin[m]> when you get to that stage, just let me know, I can save you a minute and provide the exact command to run

12:15 <SageKhan> Can anyone help me on using LUDWIG AI ... made by Uber AI

12:16 <SageKhan> Can anyone help me on using LUDWIG AI ... made by Uber AI

12:19 <SageKhan> I need to make a multilingual Automatic Speech recognition system. I have collected the data set of about 200 GB. I want to know how to prepare/preprocess the data for training it offline on my system. After training I want to implment it on live calls and on recorded as well. Can anyone help me on doing this on LUDWIG-AI made by UBER AI

13:08 <SageKhan> I need to make a multilingual Automatic Speech recognition system. I have collected the data set of about 200 GB. I want to know how to prepare/preprocess the data for training it offline on my system. After training I want to implment it on live calls and on recorded as well. Can anyone help me on doing this on LUDWIG-AI made by UBER AI

14:12 <RishabhGarg108Ri> @ryan:ratml.org I guess I fell in the same hole again. 😅In random forest classifier constructors, we have `numClasses` which is not required for the regressor. So, we can't simply use inheritance. I faced this earlier when implementing the `DecisionTreeRegressor`.

14:13 <RishabhGarg108Ri> What do you say, should I do it by simply adding another class for random forest regressor as I did earlier or can you think of any better approach to reduce duplication of code?

14:17 <RishabhGarg108Ri> To give you a little more info, there are 4 constructors and 4 `Train` methods in the `RandomForest` class and they can be used directly in the regressor if `numClasses` was not there.

15:23 <say4n[m]> jonpsy, zoq: I will not be able to make it to the meeting due to an unplanned event tonight, I shall go through the meeting notes later. 😅

15:48 <shrit[m]> zoq: did we drop the deterministic support for the #2777?

15:48 <shrit[m]> because the layers seem that they do not have a deterministic function, while in the master they still have

15:51 <zoq[m]1> No, but it was renamed https://github.com/zoq/mlpack/blob/ann-vtable/src/mlpack/methods/ann/layer/layer.hpp#L172-L190

15:53 <shrit[m]> okay, so each time I see Deterministic I am changing that to Training 👍️

15:53 <rcurtin[m]> yep; note that the variable is in the base layer class (and set automatically by `FFN`), so there is no need for the derived layer class to hold it

15:54 <shrit[m]> okay, I see now, in the case for the tests?

15:55 <zoq[m]1> You can take a look at the Dropout layer for an example.

15:56 <shrit[m]> Yeah, I understood what is happening

15:56 <shrit[m]> thanks

15:57 <rcurtin[m]> I got the `Convolution` layer to converge too (seemingly, at least). it turns out the bug was in `MaxPooling`

15:58 <rcurtin[m]> currently I'm playing with `mnist_cnn` to see if I can get good performance out of it; so far I have the validation accuracy up at 95.5%, but I should be able to do way better than that---I think this network structure should be able to get ~99%. but it takes a long time to tune the step size, etc. :)

15:58 <rcurtin[m]> even so, I'm fairly satisfied that things are working at least mostly correctly here

15:58 <shrit[m]> Yeah, me too

15:59 <shrit[m]> So, if the tests are passing for a specific layer, can I push the #2777 directly?

15:59 <rcurtin[m]> that's what I've been doing; zoq what do you think? it's your branch after all 😄

16:01 <zoq[m]1> I don't mind it at all, I guess before you push something rebase first to make it easier for everyone else.

16:01 <shrit[m]> Yeah, I pull and merge before working on any branch as always

16:03 <rcurtin[m]> We should see if we can identify the minimum set of layers used in the `examples/` repository, and prioritize those; personally I think it would be best if we tried to merge #2777 with minimal support (and some removed layers), then added those other layers back in one-by-one in simple, short follow-up PRs

16:03 <zoq[m]1> agreed

16:04 <rcurtin[m]> for instance I think I can avoid `TransposedConvolution` and `AtrousConvolution`, even though they would be very simple to adapt now that I've done `Convolution`---I'll do those in a follow-up PR

16:04 <shrit[m]> Yeah, so I can help finishing with the layer if I can

16:10 <jonpsy[m]> <yugansharora01yu> "but one thing it generated..." <- what, why? Once you're in WSL visual studio has nothing to do with mlpack.

16:11 <jonpsy[m]> just run the linux commands and run the tests. Be sure to use `make -j1` because it costs a lot of ram to build mlpack

17:03 <zoq[m]1> jonpsy: you send the invite with a meets room link

17:54 <shrit[m]> rcurtin: I have finished the code related to the arma::norm, if you can send me the command to run that would be perfect

17:55 <shrit[m]> I have a lot of tests that are failing on my machine from the master branch.

17:55 <shrit[m]> especially tests that are from the main_tests dir

17:56 <shrit[m]> These failing are not related to my modification, so I am not able to detect the false position from the true positive test