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/
soonmok has joined #mlpack
soonmok_ has joined #mlpack
soonmok__ has joined #mlpack
soonmok has quit [Ping timeout: 244 seconds]
soonmok_ has quit [Ping timeout: 272 seconds]
soonmok__ has quit [Remote host closed the connection]
soonmok has joined #mlpack
soonmok has quit [Ping timeout: 245 seconds]
soonmok has joined #mlpack
soonmok has quit [Ping timeout: 272 seconds]
soonmok has joined #mlpack
petris has quit [Quit: Bye bye.]
petris has joined #mlpack
soonmok__ has joined #mlpack
soonmok has quit [Read error: Connection reset by peer]
soonmok has joined #mlpack
soonmok__ has quit [Ping timeout: 246 seconds]
pymit has joined #mlpack
pymit has quit [Remote host closed the connection]
soonmok has quit [Remote host closed the connection]
soonmok has quit [Remote host closed the connection]
soonmok has joined #mlpack
Saksham189 has joined #mlpack
< Saksham189>
The logs for 18th and 19th seem to be missing.
< zoq>
Saksham189: Thanks for the info, we will look into it.
< zoq>
rcurtin: Nice, it's definitely different.
ayesdie has joined #mlpack
< ayesdie>
rcurtin: I was writing some tests for SparseSVMFunction class and I was having some issues
< ayesdie>
which is because of the way `SparseSVMFunction::Gradient()` and `SparseSVMFunction::Evaluate()` are defined. So I wanted to ask you if my problems were actual problem with those implementation or is it just my misunderstanding?
< ayesdie>
last time, I asked you about the `parameters`(which is being passed in Evaluate() and Gradiend()), and you said it was a concatenated vector : [ w b ]
< ayesdie>
but in those functions while evaluating the dots (for evaluating hinge loss in Evaluate() and gradient in Gradient()) , we never add the `b` term. Rather we just multiply a replicated matrix of `parameters` with the dataset ( where the parameters matrix have the `b` term in it). This leads to matrix multiplication issue while testing (or maybe because of the way I'm performing the tests).
< ayesdie>
If we consider the evaluation of one dataset at a time, we are just performing `dot = 1 - y (w*x)` rather than `dot = 1 - y (w*x +b)`
< ayesdie>
I think the same goes while we are calculating dots inside `Gradient()`
< ayesdie>
the other thing that I wanna know is that while evaluating the gradient, why are we not adding the gradient term for `dots[i] < 0`?
ayesdie has quit [Ping timeout: 256 seconds]
Saksham189 has quit [Ping timeout: 256 seconds]
soonmok has quit [Remote host closed the connection]