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/
PileOfDirt has quit [Ping timeout: 265 seconds]
PileOfDirt has joined #mlpack
ImQ009 has joined #mlpack
< shrit[m]> Another blog updates of the last week, Hope you enjoy reading https://shrit.me/blog/
carloscarlos[m] has joined #mlpack
< sakshamb189[m]> himanshu_pathak: are you there?
< himanshu_pathak[> Yes I am here
< sakshamb189[m]> I just merged your r-value reference PR.
< himanshu_pathak[> Thanks for merging
< sakshamb189[m]> so are you working on DBN PR right now? is that correct?
< sakshamb189[m]> Also have you shared your blog?
< himanshu_pathak[> Yes I am I changed my approach a little bit to apply DBN
< himanshu_pathak[> Previously I was trying to add RBM
< himanshu_pathak[> As a layer but it was quite difficult to use RBM
< himanshu_pathak[> When added as a layer with DBN
< himanshu_pathak[> Because I need to access train()
< himanshu_pathak[> Function of RBM
< sakshamb189[m]> alright I see
< himanshu_pathak[> So, I will not be able to use visitor functions that we have in our ANN module for layer and instantiate each RBM in DBN class
< jeffin143[m]> Zoq : is there any use case where we would want to plot an image after every epoch
< jeffin143[m]> Or may be a set of images after every use case ???
< himanshu_pathak[> So, I will try to complete the in two days or so
< sakshamb189[m]> himanshu_pathak: so what is the new approach that you are planning to use?
< himanshu_pathak[> Create an instance of each RBM in DBN class
< himanshu_pathak[> And store them in a std::vector
< sakshamb189[m]> have you shared your blog?
< himanshu_pathak[> Yeah today I will do share :)
< sakshamb189[m]> alright the approach sounds good to me. Let me know if you have any troubles.
< himanshu_pathak[> Sure, I will perform some changes in RBM code I will ask it on pr if feeling suspicious about them :)
< sakshamb189[m]> Alright then we can meet next week. Have a great week ahead.
< himanshu_pathak[> Yeah have a great week :)
< zoq> jeffin143[m]: Probably not after every epoch, but after every 10th, just to see how the model is doing, YOLO is a good example which prints out the image of the train batch after some epochs.
< jeffin143[m]> Do we have a example ???zoq
< jeffin143[m]> Using Mlpack ??
< zoq> jeffin143[m]: For YOLO, kartikdutt18[m] is working on that.
< jeffin143[m]> Ok , so while training
< jeffin143[m]> Where is the image held ??
< jeffin143[m]> Can we access that using the ensmallen callback ??
< zoq> jeffin143[m]: You would have to pass the data to the callback (images in mlpack are vectorized meaning each col is one image).
< zoq> jeffin143[m]: In the callback itself you could do call model.Evaluate(image) get e.g. the boudning box apply that to the image and save it.
< zoq> jeffin143[m]: But probablly better to write a callback once YOLO is ready?
< jeffin143[m]> Yes , yesterday while pass the lamba function
< jeffin143[m]> The matrix was the one which had data
< jeffin143[m]> Now ,
< jeffin143[m]> Here the matrix would be intermediate matrix
< jeffin143[m]> After every 10 epoch
< jeffin143[m]> How should I access that matrix which has the changing matrix
< jeffin143[m]> Yeah understood
< jeffin143[m]> Model.evaluate
< jeffin143[m]> Was a good example
< zoq> jeffin143[m]: I would just pass the images or whatever the data is to the callback isntance: https://gist.github.com/zoq/2ae50073388327f93bfae74f5f53143f
< zoq> Or pass it to the lambda function, and let the user do whatever has to be done to the data.
< zoq> Including saving the image?
< jeffin143[m]> Ok
< jeffin143[m]> Understood
< jeffin143[m]> Yes it has to be saved
< jeffin143[m]> We currently don't have a way to log using in memory
< zoq> Or what about the lambda function has to return an arma::mat?
< zoq> that is an image
< zoq> if you call the lambda function int he callback you just save the result or pass it to the ml board object?
< jeffin143[m]> Yes it should return an arma::mat
< jeffin143[m]> Because I am assuming if yiu cb(image)
< jeffin143[m]> Then the image would be constant
< zoq> right
< jeffin143[m]> And everytime printing same image doesn't make any sense
< jeffin143[m]> So
< jeffin143[m]> Cb(fun)
< jeffin143[m]> Functional return arma::mat
< jeffin143[m]> That will be variable
< jeffin143[m]> Like model.bounding()
< jeffin143[m]> And similarly
< zoq> Yes, I think kartikdutt18[m] is using Evaluate, but that is just a minor detail.
< jeffin143[m]> Oh I see
< kartikdutt18[m]> <zoq "Yes, I think kartikdutt18 is usi"> Right, using Evaluate the model will output a feature map which will have encoded bounding boxes which will be decoded using a post processing function.
< jeffin143[m]> Oh
< rcurtin> shrit[m]: sorry I have been slow with debugging the cereal issues... all of yesterday was fighting with mlpack.org and setting things up so future downtime will be less of a problem (and I'm still doing some of that now)
< rcurtin> today I am going kayaking, but I am hoping I can still find some time to debug the issue you mentioned
< shrit[m]> rcurtin That is good to hear, it has been a long time since I did go kakyaking
< shrit[m]> *kayaking
< shrit[m]> Even swimming,
< shrit[m]> This years have never been more chaotic
< rcurtin> yeah, kayaking is a thing one can do while staying far apart from other people... that and riding a bike are like my only outdoor activities these days
< rcurtin> last week I successfully rolled the kayak, so I am hoping I can improve my technique for that a little bit this week :)
< shrit[m]> Do you rent the Kayak, because that what we do here
< shrit[m]> You know we do not have a lot of spaces in small apartement, so mostly we rent kayak
< shrit[m]> even we rent bikes if you want to do some long distances
< rcurtin> space is cheap in america, so we bought some kayaks and store them in our crawl space under the house :)
< rcurtin> even though I live basically walking distance (<1mi) from midtown Atlanta with lots of highrises and everything, most housing is detached single-family homes, even that close to a city center
< rcurtin> I am not sure how common that is in the world, but it certainly seems like every other city I travel to in another country is way more dense
PileOfDirt has quit [Ping timeout: 240 seconds]
PileOfDirt has joined #mlpack
< rcurtin> shrit[m]: I am not done debugging yet, but I think the problem is that we are using cereal::make_array() inside of the serialize() implementation for armadillo objects in serialize_armadillo.hpp
< rcurtin> specifically, we make the call `mat.set_size(n_rows, n_cols)`, which allocates memory for the matrix
< rcurtin> but then we call `cereal::make_array()` using mat.mem, which actually allocates new memory for `mat.mem`, but probably does not set mat.mem's pointer right
< rcurtin> however, since we've already allocated memory for the matrix with `set_size()`, we can simply iterate and serialize each element, instead of needing to use the array wrapper to allocate memory during load
< rcurtin> I am walking out the door now, so I'll be back later, but here is the diff I am trying: https://pastebin.com/9W2xB39R
< walragatver[m]> jeffin143: birm: Are you here?
< jeffin143[m]> Hi walragatver :)
< walragatver[m]> jeffin143: Do you want to discuss anything?
< jeffin143[m]> walragatver: yes I had a discussion with zoq about some callback regarding , images
< jeffin143[m]> I will add that to call back pr
< jeffin143[m]> Rest everything is fine
< walragatver[m]> Okay
< jeffin143[m]> I resolved the comments you made on embedding
< jeffin143[m]> And histogram
< walragatver[m]> jeffin143: Okay
< walragatver[m]> I wil give you final reviews on them soon
< walragatver[m]> I don't have anything to discuss currently.
< jeffin143[m]> Ok
< jeffin143[m]> How is your job going :) walragatver
< jeffin143[m]> Where are you currently
< walragatver[m]> jeffin143: It's going good.
< walragatver[m]> I am currently working from home
< walragatver[m]> So currently I am at my hometown. It's name is Malegaon, situated near Nashik
< walragatver[m]> jeffin143: What about you?
< walragatver[m]> How's your job going
< jeffin143[m]> It's good :)
< walragatver[m]> jeffin143: I would be working in Bangalore. Where would you be working?
< jeffin143[m]> walragatver: same place
< jeffin143[m]> :)
< walragatver[m]> jeffin143: Okay great
< jeffin143[m]> walragatver: :)
< walragatver[m]> jeffin143: I am signing off now. Let me know if you need any help
< jeffin143[m]> Sure :)
< walragatver[m]> Let's meet on Wednesday. Till then take care
< birm[m]> Sorry for missing jeffin143 im floating around random spots in the us. I'll respond to any emails as I can :)
< jeffin143[m]> Yes
< jeffin143[m]> birm: how are you ?
< birm[m]> Doing ok. Just visiting my brother, and wandering around the country. how are you?
ImQ009 has quit [Quit: Leaving]