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
texasmusicinstru has quit [Remote host closed the connection]
<heisenbuugGopiMT> Boost is basically testing on 2000 values which are in case of `digamma` are divided into 4 parts.
<heisenbuugGopiMT> Each type has nearly 500 values.
<heisenbuugGopiMT> Large values, negative values, roots near 0, near the positive roots.
<heisenbuugGopiMT> Shall I store them in a different these values, with their respective answers in a different file other than test file?
<heisenbuugGopiMT> If so, boost is using `.ipp`, should we use simple `csv` or use the same as `boost`?
<rcurtin[m]> personally I think a CSV is just fine
<shrit[m]> I think CSV is fine as well
<heisenbuugGopiMT> okay...
<heisenbuugGopiMT> Should I create a new directory in tests named data to store these files?
<shrit[m]> I think it should be added to the place where mlpack datasets used in the `build/` dire
<shrit[m]> s/dire/directory/
<heisenbuugGopiMT> OKay, so there is already data directory in test directory
<heisenbuugGopiMT> I think I can add those here only?
<heisenbuugGopiMT> My bad I somehow missed it earlier...
<shrit[m]> Yeah, you can added it
<heisenbuugGopiMT> Okay, I think this is last, whats with the epsilon?
<heisenbuugGopiMT> How to decide it's parameter?
<heisenbuugGopiMT> Some places we are using `1e-7` and some `1e-5`
<heisenbuugGopiMT> I believe it's the difference between the values which we are comparing, right?
<shrit[m]> for the tests?
<heisenbuugGopiMT> Yupp
<shrit[m]> Yeah I think it is the marge of difference because no two doubles are the same
<heisenbuugGopiMT> `REQUIRE(output(0, 0) == Approx(1.0).epsilon(1e-7));`
<heisenbuugGopiMT> Like this...
<shrit[m]> s/marge/limite/
<shrit[m]> Yeah, it is just the difference between these two is smaller than epsilon
<shrit[m]> d1- d2 < epsilon
<shrit[m]> if epsilon is too small, then we can say d1~ d2
<shrit[m]> or =
<heisenbuugGopiMT> Oh okay, so what should I keep this?
<heisenbuugGopiMT> Should I test some values myself?
<say4n[m]> if I am not mistaken, generally epsilon relates to the measure of precision of the prediction (see: https://stats.stackexchange.com/a/204560)
<say4n[m]> so a sufficiently small should be okay :)
<say4n[m]> *small value
<heisenbuugGopiMT> Yea, I guess I will try out with some values and see what happens.
braceletboy has joined #mlpack
<jonpsy[m]> rcurtin: I know you aren't the biggest fan of Windows. But have you tried 11? After the initial freeze, its working good enough
<rcurtin[m]> I don't have any reason to try it 😃
<rcurtin[m]> I don't imagine it's bad, but me and the Windows workflow have some fundamental incompatibilities :)
<RishabhGarg108Ri> I did install it on my father's laptop. Haven't used it much, but the UI feels nice. :)
<shrit[m]> RishabhGarg108 (RishabhGarg108): Yeah, I agree that most propriety OS tries to make it look nice rather than make it work well
<shrit[m]> when I install Linux, I never install any kind of of UI or GUI, if I can it would be fine for me to have a terminal and a browser
<shrit[m]> Still waiting for the day when someone will come with a way to provide a great browsing on a framebuffer