rcurtin_irc changed the topic of #mlpack to: mlpack: a scalable machine learning library (https://www.mlpack.org/) -- channel logs: https://libera.irclog.whitequark.org/mlpack -- NOTE: messages sent here might not be seen by bridged users on matrix, gitter, or slack
<ABHINAVANAND[m]> zoq I found the bug. I set the weights to (2*size + 1). For some reason this was leading to fail test case in all is except windows.
<ABHINAVANAND[m]> The actual size of (2*size). When I set it to this the tests are passing.
<ABHINAVANAND[m]> Can you review the PR. It's ready to review.
<jonpsy[m]> hey, I want to apply as a member of mlpack. Where do I start ? :D
<heisenbuugGopiMT> @shrit:matrix.org
<heisenbuugGopiMT> I ran the tests there's a whopping difference. The new parser took nearly 1/4 time to complete the benchmark test.
<heisenbuugGopiMT> But something happened
<heisenbuugGopiMT> Tests for both old and new parser stopped at `row: 399`
<heisenbuugGopiMT> Can this be due to RAM?
<heisenbuugGopiMT> So should plot with that much data?
<heisenbuugGopiMT> Or should I re-run it?
<heisenbuugGopiMT> Also the work report is complete but I was waiting for speed-up results so that I can included that in mail.
<heisenbuugGopiMT> * Also the work report is complete but I was waiting for speed-up results so that I can include that in mail.
<heisenbuugGopiMT> Also there is a small bug, our new parser is not able to load when we have data with only one row.
<heisenbuugGopiMT> I am looking into it, will update once I find a solution.
<heisenbuugGopiMT> Also results will look somthing like this.
<heisenbuugGopiMT> * Also results will look something like this.
<heisenbuugGopiMT> It's 15 columns, not 5.
<shrit[m]> that looks good to me,
<shrit[m]> What do you mean that they both stopped? what happened at that moment?
<shrit[m]> It might be related to RAM issue, I am not sure, but I am really happy to see the improvement that we have done
<shrit[m]> Great work :1
<shrit[m]> * Great work 💯
<heisenbuugGopiMT> Have a look at the files which I uploaded, after 398 everything is 0.
<heisenbuugGopiMT> * Have a look at the files which I uploaded, after `line 398` everything is 0.
<heisenbuugGopiMT> This seems like `root(n)` as opposed to boost which was `linear` maybe
<say4n[m]> <jonpsy[m]> "hey, I want to apply as a member" <- https://github.com/mlpack/mlpack/blob/master/GOVERNANCE.md :)
<heisenbuugGopiMT> Thanks for the link.
<jonpsy[m]> okay, ive made myself familiar with the potential responsibilities. So, how should I structure my proposal and where do I post it? (md file said a public place) will the mailing list do?
<heisenbuugGopiMT> @shrit:matrix.org ?
<shrit[m]> I am here
<heisenbuugGopiMT> About that bug...
<shrit[m]> I will check it
<heisenbuugGopiMT> So the issue is the load_okay is setting to `false`
<shrit[m]> Which bug, the zeros bug?
<heisenbuugGopiMT> And hence it is not even entering the loop once.
<heisenbuugGopiMT> No, our parser is not able to read file with 1 row of data.
<heisenbuugGopiMT> I think it is happening coz the `ios::eof` is being set to true...
<heisenbuugGopiMT> But if I reset that before `f.good()` then we are ending up with one extra col
<shrit[m]> The function is called GetMatSize
<shrit[m]> if you are providing only one row, that means that there is only line in the file,
<shrit[m]> but I believe in you case if not getting even the first and only row of data
<shrit[m]> * but I believe in your case if not getting even the first and only row of data
<heisenbuugGopiMT> Yup..
<heisenbuugGopiMT> Shall I show you this? We can hop on a call in 10 mins if you are free.
<shrit[m]> You can true to debug this by printing all the stream status: good(), eof(), fail(), bad()
<shrit[m]> * You can try to debug this by printing all the stream status: good(), eof(), fail(), bad()
<shrit[m]> Then we will know why the stream is not good
<shrit[m]> Let me know of the results
<heisenbuugGopiMT> Okay...
<shrit[m]> I can not have a meeting today, too much things to do 😒
<heisenbuugGopiMT> Yea, it's fine...
<heisenbuugGopiMT> If you are busy we can talk about this tomorrow...
<shrit[m]> no worries, I can up in the chat
<heisenbuugGopiMT> f.good(): 0
<heisenbuugGopiMT> f.fail(): 0
<heisenbuugGopiMT> f.eof(): 1
<heisenbuugGopiMT> f.bad(): 0
<heisenbuugGopiMT> As I thought, `eof` bit is being set to 1
<heisenbuugGopiMT> I will look into you, we can discuss further tomorrow.
<shrit[m]> heisenbuug (Gopi M Tatiraju): I wondering if the bug can be observed in the master version as well, do you have an idea?
<shrit[m]> If we can see it in the master, it would be easier to open an issue and handle it after merging you pull request
<heisenbuugGopiMT> Nope, no issue in master version.
<heisenbuugGopiMT> I think it got messed up when we divided the code into fucntions.
<heisenbuugGopiMT> Let me confirm once again.
<heisenbuugGopiMT> Yup, no issue in the master.
<shrit[m]> ?
<shrit[m]> I mean we are checking twice if the stream is good, why?
<heisenbuugGopiMT> I will check, give me some time.
aakashi2001 has joined #mlpack
aakashi2001 has quit [Changing host]
aakashi2001 has joined #mlpack
aakashi2001 has quit [Ping timeout: 250 seconds]
<zoq[m]> > <@_slack_mlpack_U01HVTEQU2X:matrix.org> zoq I found the bug. I set the weights to (2*size + 1). For some reason this was leading to fail test case in all is except windows.... (full message at https://libera.ems.host/_matrix/media/r0/download/libera.chat/c16df78612a04f6771cefafa0ae5b5d9160058ab)
<zoq[m]> <jonpsy[m]> "okay, ive made myself familiar w" <- We usually ask mentors at the end of GSoC for feedback and suggestions to invite people to the org and do a vote afterwards. So in case you are a GSoC applicant (you are one) you don't have to submit an extra application.
aakashi2001 has joined #mlpack
aakashi2001 has quit [Changing host]
aakashi2001 has joined #mlpack
aakashi2001 has quit [Ping timeout: 258 seconds]
aakashi2001 has joined #mlpack
<jonpsy[m]> Great. so I've made some progress on the loss thingy
<jonpsy[m]> would you like to discuss?
aakashi2001 has quit [Quit: Leaving]
aakashi2001 has joined #mlpack
aakashi2001 has quit [Ping timeout: 252 seconds]
<heisenbuugGopiMT> Although I agree that maybe we can remove the check from `GetMatrixSize()`
<heisenbuugGopiMT> We have the same in armadillo actually, have a [look](https://gitlab.com/conradsnicta/armadillo-code/-/blob/10.4.x/include/armadillo_bits/diskio_meat.hpp#L1636)
<shrit[m]> heisenbuug (Gopi M Tatiraju): is this one that is causing the issue or the numerical one?
<heisenbuugGopiMT> This one, categorical one.
<shrit[m]> Okay
<heisenbuugGopiMT> Found a solution.
<heisenbuugGopiMT> And also add `trim(line)`
<heisenbuugGopiMT> This was we won't be touching the stream before it goes to `GetMatSize<>()`
<heisenbuugGopiMT> I also tested with more than 1 row seems like working. Do you think its a good fix?
<heisenbuugGopiMT> * This way we won't be touching the stream before it goes to `GetMatSize<>()`
<jonpsy[m]> zoq kartikdutt18 (kartikdutt18) I'm having difficulty designing the backward function, can you help? thanks
<shrit[m]> heisenbuug (Gopi M Tatiraju): That seems as a good plan.
<jonpsy[m]> want gradient w.r.t to HQ and Q
<heisenbuugGopiMT> > @heisenbuug: That seems as a good plan.
<heisenbuugGopiMT> okay, I will push this modification.
<heisenbuugGopiMT> @shrit:matrix.org somehow making that change is resulting in failing test cases. The quoted text cases are failing after that change.
<heisenbuugGopiMT> I am looking into it.
<say4n[m]> Hi Sai, are you planning to have another mid week meeting this week? I can’t make it to the meeting today due to some engagements.