ChanServ changed the topic of #mlpack to: "mlpack: a fast, flexible machine learning library :: We don't always respond instantly, but we will respond; please be patient :: Logs at http://www.mlpack.org/irc/
< robertohueso>
rcurtin: Exactly
< robertohueso>
Also tried the m > |R| idea, but if I set a high probability of getting a prediction within the error tolerance, let's say 95% (i.e. MCProb=0.05) then all Score(...) calls recurse (because m > |R|)
jeffin143 has joined #mlpack
< jeffin143>
lozhnikov : u there, I am not sure what functionality do we want achieve using copy constructor*
< robertohueso>
But still, anyway, whenever a Monte Carlo prediction is calculated (I set a high MCProb value), the error gets out of the tolerance limit. I understand this is a probabilistic method but it fails too many times to be right (it didn't work even once).
< robertohueso>
I'm sure I'm making some mistake either with the sampling or the sumation so I'll check it tomorrow (it's a bit late here now)
< robertohueso>
I'm really sorry this is getting behind schedule :(
< rcurtin>
no, no problem at all :) we already found an error in the paper, it's ok
< rcurtin>
I'll try and think about other things to test. perhaps one test would be to generate points from a one-dimensional gaussian, and make sure that the sample mean and variance are converging to what they are expected to
< rcurtin>
that could help rule out a bug or something like this
< robertohueso>
Thank you :) I'll try that
< rcurtin>
and if you like, I'm happy to dig deeper if you want to send me some instructions on how to replicate what you're seeing, what you expect to see, etc. :)
jeffin143 has quit [Ping timeout: 248 seconds]
jeffin143 has joined #mlpack
< jeffin143>
lozhnikov : also We we're providing , getmappings() so that user can modify, but since it doesn't function properly ,what we can do is provide an overload function where user can pass a map and then we can just handle that internally.?
< jeffin143>
And if this idea doesn't suit, how should we redefine our data member mappings.? So that we can do what is required
< lozhnikov>
jeffin143: The standard copy-constructor doesn't handle string_view correctly.
< lozhnikov>
jeffin143: The mappings variable is defined perfectly well. But we shouldn't provide the ability to modify the mappings explicitly.
vivekp has quit [Ping timeout: 248 seconds]
vivekp has joined #mlpack
< ShikharJ>
sakshamB: toshal: Guys, let me know when you guys want another review? Either on the PR or here, works fine.
jeffin143 has joined #mlpack
< jeffin143>
lozhnikov , how should we rectify it.?
< sakshamB>
ShikharJ: hey, I have made the requested changes to highways network and MiniBatchDiscrimation PRs. Also Inception Score PR should be ready for review.
< lozhnikov>
jeffin143: I think the easiest way is to return a pair. The pair should contain the map and the deque of the strings.
< jeffin143>
You are talking about copy constructor.
< jeffin143>
?
< lozhnikov>
No, I am talking about the Mappings() function.
< jeffin143>
Ok
vivekp has quit [Ping timeout: 272 seconds]
< jeffin143>
So mappings function will return a pair and if user wants to modify it then
< jeffin143>
He must modify that pair
< lozhnikov>
Not exactly. But if the user modifies the map, the user has to modify the deque as well. But you didn't provide the interface.
< lozhnikov>
So you can just add an interface to modify the deque.
< jeffin143>
Ok ,
< jeffin143>
Also one more doubt*
< jeffin143>
Dictionary_encoding ob1,ob2;. You are saying Ob1=ob2 won't work..? Right
< lozhnikov>
Yes, sure.
< lozhnikov>
Suppose you have class C which contains string S and string_view V. V is linked to S.
< jeffin143>
Ohk so now i have to copy all those strings , and then create new V.? For a new object
< lozhnikov>
Let NC = C. Then NC.V is equal to C.V. And NC.V is linked to C.S (instead of NC.S!!!)
< lozhnikov>
That's why the standard copy-constructor won't work.
< lozhnikov>
You have to manually link string views to the new objects.
< jeffin143>
Yes , so I have to override the constructor , by creating the strings and the Linkin the string_vjew
< jeffin143>
Ok
< lozhnikov>
Yes, that's right.
< jeffin143>
Thanks , would work it out and ping you thanks 😀
< lozhnikov>
ok)
< lozhnikov>
jeffin143: I forgot to say: we have to redefine the operator= as well.