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/
xiaohong has quit [Remote host closed the connection]
xiaohong has joined #mlpack
vivekp has joined #mlpack
k3nz0 has quit [Ping timeout: 258 seconds]
xiaohong has quit [Remote host closed the connection]
xiaohong has joined #mlpack
xiaohong has quit [Remote host closed the connection]
xiaohong has joined #mlpack
xiaohong has quit [Remote host closed the connection]
xiaohong has joined #mlpack
xiaohong has quit [Remote host closed the connection]
< Toshal>
sakshamB: Sorry for the late response
< Toshal>
I think we will need a visitor for the same thing as well.
xiaohong has joined #mlpack
< Toshal>
Because for some layers there is no concept called wieght without bias. So to handle that we will need visitor.
< Toshal>
I think that the role of visitors is to handle these sort of exceptions. The basic aim to use visitors is to call getters preventing errors and exceptions.
< Toshal>
This is what I thought just let me know your thoughts regarding same.
< Toshal>
So I think we will need a visitor for calling that getter.
< Toshal>
zoq: I was thinking to add a layer similar to concat layer which will concatenate network's instead of layers. For inception layer I am needing to concatenate output of four 2 layer networks. That's why I was thinking to add a generalize layer for it.
< Toshal>
Let me know what you think.
xiaohong has quit [Remote host closed the connection]
xiaohong has joined #mlpack
xiaohong has quit [Remote host closed the connection]
xiaohong has joined #mlpack
xiaohong has quit [Remote host closed the connection]
< sakshamB>
Toshal: could you give me an example of such a layer
xiaohong has joined #mlpack
xiaohong has quit [Remote host closed the connection]
xiaohong has joined #mlpack
< sakshamB>
Toshal: hmm…I guess we would still require the visitor
xiaohong has quit [Remote host closed the connection]
xiaohong has joined #mlpack
xiaohong has quit [Remote host closed the connection]
xiaohong has joined #mlpack
yoyo has joined #mlpack
yoyo is now known as karthik1972
< karthik1972>
i have done with the installation... how to proceed further?
< sreenik[m]>
karthik1972: You can include mlpack and use it like any other c++ library. You might need to link against armadillo and boost while compiling (you can find this in the mlpack github issues section like I had mentioned before). What exactly are you aiming to do?
< karthik1972>
I want to be a contributor to MLpack
xiaohong has quit [Remote host closed the connection]
xiaohong has joined #mlpack
< karthik1972>
okay thanks a lot!!
KimSangYeon-DGU has joined #mlpack
xiaohong has quit [Remote host closed the connection]
xiaohong has joined #mlpack
< lozhnikov>
jeffin14327: I see. I'll look through the code today.
< sakshamB>
Toshal: Could you work on the visitor in a new PR and maybe I can review and see if it going to fit my needs.
< zoq>
Toshal: There is a layer called Sequential, which is basically a network as a layer. Do you think that could work? Also, I worked on an inception layer some time ago, I'll see if I can find the code.
xiaohong has quit [Read error: Connection timed out]
k3nz0 has joined #mlpack
< zoq>
favre49: You have to introduce a new template parameter which is basically an alias for the TaskType.
xiaohong has quit [Remote host closed the connection]
xiaohong has joined #mlpack
vivekp has quit [Ping timeout: 244 seconds]
vivekp has joined #mlpack
karthik1972 has quit [Ping timeout: 260 seconds]
xiaohong has quit [Remote host closed the connection]
xiaohong has joined #mlpack
k3nz0 has quit [Remote host closed the connection]
k3nz0 has joined #mlpack
yoyo has joined #mlpack
yoyo is now known as karthik1972
k3nz0_ has joined #mlpack
k3nz0 has quit [Ping timeout: 245 seconds]
< Toshal>
sakshamB: Okay, fine. I will try my best for implementing it. I hope I am able to do it. Just one thing. Are you 100 percent sure that you are needing dimension of wieghts in your work?
< Toshal>
If yes then I would start working immediately on it.
< Toshal>
zoq: You are quite correct. I think sequential layers would be helpful. Thanks for reminding me of sequential layers.
k3nz0__ has joined #mlpack
k3nz0_ has quit [Ping timeout: 248 seconds]
jeffin14397 has joined #mlpack
< jeffin14397>
lozhnikov : saw your comments on pr
< jeffin14397>
How should I pass different arguments , then without.the check
< jeffin14397>
I mean for dictionary argument , the arguments are dataset.size(), colsize and for the other types it is dataset.size(),mappings.size()
< lozhnikov>
jeffin14397: You needn't pass different arguments. The policy handles all differences.
< jeffin14397>
Umm ok , is it ok to paas arguments which are of no use inside a function?
< lozhnikov>
jeffin14397: I think Marcus has already answered: "Right, one solution would be to pass everything an only use the required parameters in the policy class."
< jeffin14397>
Yes , but I found this to be a better solution , than that and hence I opted for that , I will go with this then , passing everything and then using only required one
xiaohong has quit [Remote host closed the connection]
< sakshamB>
Toshal: yes I need the dimensions
xiaohong has joined #mlpack
< lozhnikov>
jeffin14397: I think it's hard to support. Suppose you want to add a policy. The first approach requires only the policy. But your approach requires some modification of the base class as well.
< jeffin14397>
Ohh , I didn't think that way, you are right , fixing the base class mean , manipulating the policy classes always
< jeffin14397>
But if in future, a policy demands for new parameters then we need to pass that to , and then needs to amend base class in that case
xiaohong has quit [Ping timeout: 246 seconds]
< lozhnikov>
jeffin14397: It depends on the quality of the API. Sometimes it's possible to think about all future needs in the beginning:)
< lozhnikov>
jeffin14397: Have you received the message about the blog post?
< jeffin14397>
Yes sry about that , I will write it by today , these day I was all into learning bits and pieces of policy designing and finding out codes
< lozhnikov>
ok
jeffin14397 has quit [Ping timeout: 260 seconds]
xiaohong has joined #mlpack
karthik1972 has quit [Ping timeout: 260 seconds]
< ShikharJ>
sakshamB: Toshal: I don't think I'll be available today for much long. Can we start if you guys are here?
xiaohong has quit [Remote host closed the connection]
< sakshamB>
ShikharJ: yes I am here
< ShikharJ>
sakshamB: I've started on the final reviews for MiniBatch Discrimination, but got side-swapped with some other work, I'll probably finish that today.
< ShikharJ>
Looks good mostly though
< sakshamB>
ShikharJ: ok. also should I continue with implementing the Padding layer or do we fix the convolution layers first?
< ShikharJ>
sakshamB: Which convolution layers? Atrous?
< ShikharJ>
sakshamB: I don't think transposed needs that big of a fixing, it fails mostly on OSX. Atrous is another thing though. What about the visitors that we discussed last time?
< sakshamB>
Toshal: will be opeing a new PR for it. I discussed with him on IRC.
< ShikharJ>
sakshamB: You can do either of those two, but be aware that Transposed convolution would take sometime getting used to. So I would suggest working on padding layer first.
< Toshal>
ShikharJ: I am here
< sakshamB>
ShikharJ: yes I’ll open a PR for the padding layer
< ShikharJ>
sakshamB: I'm saying this because given the current testing infrastructure, it will be better to add padding layer and check against existing tests. After adding the changes Atharva's PR is making, we'll have to look if the problem persists and then look for the changes, which would be time-consuming.
< ShikharJ>
Toshal: Great, what's the progress on FID?
< ShikharJ>
As I recall you were last working on that?
xiaohong has quit [Remote host closed the connection]
< Toshal>
ShikharJ: I am working on inception layer which is needed for it's testing.
< Toshal>
The implementation of the the basic frechet distance is quite complete.
< Toshal>
I just need to add the parameters of the inception layer.
< ShikharJ>
Hmm, okay, normally I'd like to know more, but I'm kinda in a hurry. I'll review your work shortly.
< Toshal>
Okay fine. No problem
< Toshal>
Also I am quite sorry for not attending the previous meeting.
< ShikharJ>
Toshal: And I didn't say this last time, but congratulations on your job. Hope you have a great time there!
< ShikharJ>
Toshal: No worries there.
< Toshal>
ShikharJ: Thanks.
< ShikharJ>
sakshamB: Toshal: I'll be off for now, but I'll be back later during the day (morning/afternoon for you guys). Have a fun week!
< Toshal>
Okay.
< Toshal>
Have a good day.
KimSangYeon-DGU has quit [Remote host closed the connection]
vivekp has quit [Ping timeout: 245 seconds]
vivekp has joined #mlpack
vivekp has quit [Ping timeout: 245 seconds]
vivekp has joined #mlpack
vivekp has quit [Read error: Connection reset by peer]
< jeffin143>
Moved For loop inside the Encode Function // Since I need it for TF-IDF Function
< jeffin143>
Also used enable_if for the vector<vector<size_t>> : Added a check since, all the other encoding doesn't provide this interface and hence added just a check for dictionary Encoding
< lozhnikov>
jeffin143: I added some comments.
< jeffin143>
lozhnikov : I suggest to encode only one token here. In that case you needn't mappings here. --> Do you mean remove for loop out of the function ?
< lozhnikov>
jeffin143: I see. Thanks for the clarification. However I suggest to add a function that encodes only one token for one pass encoding.
< jeffin143>
ok, I will do it, When we will arrive at that PR, We will probably find out a way to do it :)
< jeffin143>
lozhnikov : I suggest to pass the code instead of mappings, dataset and originalStrings --> not sure i understand this , pass the code means ?
< lozhnikov>
jeffin143: It's for one pass encoding. It's enough to pass a value that encodes the token.
< lozhnikov>
jeffin143: Something like mappings.at(dataset[i][j])
< jeffin143>
Ok , got it
< jeffin143>
Last doubt, introduce some kind of traits means, what should i do, i thought is_same is type checking traits
< lozhnikov>
jeffin143: It's a struct that describes the properties of the policy.
< jeffin143>
Also , can i make a folder in methods section , which as a folder string_utility and then everything we build related can be structured inside that folder ?
< lozhnikov>
jeffin143: I guess the methods folder usually contains algorithms with CLI.