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/
< akfluffy>
hey, how do I set a specific number of filters for a convolutional layer?
< akfluffy>
nevermind that, I have another question. Someone here recently mentioned the ability to pass an image with 3 color channels to a CNN. How do I do that if the FFN only accepts matrices and not cubes?
< rcurtin>
gmanlan: sorry for the slow response---this was ICML week---I think I see what needs to be done, I'll try and handle it now
< lozhnikov>
jeffin143: Maybe "SplitByAnyOf" or "SplitByAnyOfTokenizer"? Actually, I don't have good ideas right now. I'll try to think about the name.
KimSangYeon-DGU has joined #mlpack
KimSangYeon-DGU has quit [Quit: Page closed]
vivekp has quit [Read error: Connection reset by peer]
vivekp has joined #mlpack
vivekp has quit [Ping timeout: 244 seconds]
vivekp has joined #mlpack
< zoq>
akfluffy: You have to vectorise the image so, one column is a one image.
< akfluffy>
zoq: thanks, and each row is a color channel?
< zoq>
The channels are embedded in the column as well: col -> R G B
< akfluffy>
zoq: thank you very much :)
< zoq>
Once the Loading Images PR we will provide some examples.
travis-ci has joined #mlpack
< travis-ci>
robertohueso/mlpack#14 (mc_kde_error_bounds - 3c17e98 : Roberto Hueso Gomez): The build was fixed.
< akfluffy>
zoq: in the case of the first convolutional layer (input) when vectorising an image, would the input width and input height be the image dimensions? or would it represent all pixels as one dimension and the 3 channels as another?
< akfluffy>
for example, if I had a 448x448 image, would the convolutional layer dimensions be 448x448 or 200704x3 ?
< zoq>
448x448 and inSize would be 3
< zoq>
I think that it should work, but I guess I have to test it out.
< akfluffy>
zoq: insize as in number of input activation maps?
< zoq>
right
< akfluffy>
zoq: thanks :)
< zoq>
I'll write a simple example to test it out tomorrow.
< akfluffy>
how far is Bandicoot? would it accelerate my model? even with nvBlas I'm still getting about 10s per evaluation with 9 convolutional layers