01:34 < shrit[m]> rcurtin, is there any mlpack svg logo available?

01:37 < zoq> shrit[m]: see https://github.com/mlpack/mlpack.org/tree/master/img there is one in white and one in black.

01:38 < shrit[m]> Perfect thanks,

02:57 < rcurtin> pradkrish: yeah, I use irssi

04:27 < RishabhGarg108Gi> Hey @rcurtin , can you please clarify this - https://github.com/mlpack/mlpack/issues/2005#issuecomment-724130027. It would be great help. Thanks :)

08:42 < ka_shGitter[m]> Hi everyone, I'm trying to solve sparse decomposition problem using mlpack. Since now I've been trying to solve it using mlpack::sparse_coding::SparseCoding but unable to get desired results, I want to solve X ~= coeff * D equation which X dim(1 x n) is input data, D dim(m x n) another input data on which coeff will be solved and coeff dim(1 x m) will be the output of the solver.

08:43 < ka_shGitter[m]> I've been using sklearns decomposition.SparseCoder in python and now need some help with mlpack to do similar in c++. https://scikit-learn.org/stable/modules/generated/sklearn.decomposition.SparseCoder.html

08:43 < ka_shGitter[m]> Any help would be greatly appreciated, thank you

13:28 < rcurtin> ka_shGitter[m]: yeah, SparseCoding should be able to solve that problem for you; what problem are you having?

14:07 < ka_shGitter[m]> (edited) ... ... nAtomsTimes 0.0] ... => ... ... nAtomsTime 0.0] ...

14:09 < ka_shGitter[m]> The output codes I got is a vector of dimension `nAtoms` which is expected and with values ranging btw (-1,1), however it gave 0.5 as first element and rest values are not at all close to zero

14:39 < rcurtin> ka_shGitter[m]: I don't follow why you are doing step 3; shouldn't you just be able to pass the data in directly?

14:40 < rcurtin> also, with sparse coding, the dictionary is not necessarily sparse---instead, each data point is encoded as a sparse linear combination of elements in the dictionary

14:43 < ka_shGitter[m]> Step 3 is like a testing thing for me as in my application I will be passing more unknown Y(nPoints,1) matrices to get the codes(nAtoms,1) for the given dictionary

14:43 < rishishounakGitt> could someone suggest issues for a newcomer

14:43 < rcurtin> ka_shGitter[m]: also, maybe you need a larger lambda1 parameter?

14:44 < ka_shGitter[m]> I've tried with as large as 100.0 but results aren't expected, with 0.1 I'm getting 0.65700 and with 100.0 I'm getting 0.66011

14:45 < rcurtin> what do the 0.65700 and 0.66011 represent?

14:45 < ka_shGitter[m]> (edited) ... getting 0.66011 => ... getting 0.66011 for first atom

14:46 < rcurtin> ok, I see, and you are saying that the rest of the elements in the encoded vector also have all nonzero values?

14:46 < ka_shGitter[m]> Yup

14:46 < rcurtin> can you provide a minimal reproducible example?

14:46 < rcurtin> like some code that I can run on my end to see what happens

14:47 < ka_shGitter[m]> Sure

14:55 < ka_shGitter[m]> (edited) ... `

14:55 < ka_shGitter[m]> double ... => ... `

14:55 < ka_shGitter[m]> double ...

15:33 < rcurtin> ka_shGitter[m]: thanks, let me try it; I ran the sparse coding tests and ensured that at least for those test cases, the result given was sparse

15:38 < rcurtin> ka_shGitter[m]: I got these results:

15:38 < rcurtin> 0.999999 0 0 0 0 0 0 0 0 2.0761e-07 0 0 0 0 0 0 0 0 0 0 7.00827e-07 0 0 0 0 0 0 0 0 0 0

15:39 < rcurtin> I didn't tune any parameters; I just ensured that the number of rows in `dataset` was the dimensionality of the data (31), and the number of columns was the number of points (136)

15:39 < rcurtin> sorry I got that backwards! the dimensionality is 136, and the number of points is 31

15:40 < rcurtin> so `dataset` has size `136x31` to give those results, and `Y` has size `136x1`

15:40 < rcurtin> did I overlook something in the example? that seems like it is performing correctly

15:58 < ka_shGitter[m]> Codes looks perfect to me, I'll give it a try and get back in an hour

17:18 < ka_shGitter[m]> Thank you very much for help, it worked for me as well :)

17:18 < rcurtin> awesome, I guess just the matrices were transposed or something?

17:20 < ka_shGitter[m]> True, also I was directly loading it from the application interface which was a bit messed up so didn't notice the correctness of dimensions

17:37 < rcurtin> ok, great that you got it worked out though

17:38 < rcurtin> but even though it had the wrong dimensions, it still ran correctly?

17:46 < rcurtin> (I'm just trying to figure out if there is anything to fix here)

17:46 < rcurtin> (if SparseCoding let you run on invalid data and still gave a result, it may be that we can issue a warning in such a situation to have saved you the debugging time)

18:53 < NippunSharmaGitt> Hi all, if anyone is free can you please review #2704 ?

21:13 < abernauer[m]> Technical Python Assessment with EY went well today.

22:06 < rcurtin> abernauer[m]: awesome, hoping for the best

22:52 < zoq> abernauer[m]: fingers crossed, whiteboard python interview?