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/
robb9 has joined #mlpack
< robb9> hey, i'm trying to format data for an LSTM and I am wondering if I did it correctly
< robb9> I made each "data point" a timestep (cube slice), each slice has 1 row that has 2 columns (features)
< robb9> also, in this case, (I am using an LSTM) how many inputs should the first layer have? 2 because I have 2 features?
robb9 has quit [Quit: Page closed]
seewishnew has joined #mlpack
seewishnew has quit [Ping timeout: 250 seconds]
ahmedmaher has joined #mlpack
Thong_ has joined #mlpack
Thong_ has quit [Client Quit]
Thong_ has joined #mlpack
pd09041999 has quit [Quit: Leaving]
pd09041999 has joined #mlpack
Thong_ has quit [Client Quit]
seewishnew has joined #mlpack
seewishnew has quit [Remote host closed the connection]
seewishnew has joined #mlpack
seewishnew has quit [Ping timeout: 250 seconds]
seewishnew has joined #mlpack
seewishnew has quit [Remote host closed the connection]
seewishnew has joined #mlpack
seewishnew has quit [Remote host closed the connection]
seewishnew has joined #mlpack
seewishnew has quit [Ping timeout: 258 seconds]
ahmedmaher has quit [Read error: Connection reset by peer]
ahmedmaher has joined #mlpack
Mulx10 has joined #mlpack
< Mulx10> robb9: I guess the cube structure is (n_features, n_data, n_timestep).
< Mulx10> robb9 : If the input has two features than input size should be 2, if input has two timestamps then rho should be 2
< Mulx10> robb9 : there is a typo, it is time steps not timestamps.
pd09041999 has quit [Ping timeout: 245 seconds]
Mulx10 has quit [Ping timeout: 256 seconds]
pd09041999 has joined #mlpack
heisenbug_ has joined #mlpack
< heisenbug_> Heyy, I am interested in working for MLPACK, can anyone let me know how to get started.
heisenbug_ has quit [Ping timeout: 256 seconds]
pd09041999 has quit [Ping timeout: 250 seconds]
pd09041999 has joined #mlpack
pd09041999 has quit [Read error: Connection reset by peer]
pd09041999 has joined #mlpack
seewishnew has joined #mlpack
seewishnew has quit [Remote host closed the connection]
seewishnew has joined #mlpack
seewishnew has quit [Ping timeout: 250 seconds]
pd09041999 has quit [Quit: Leaving]
rf_sust2018 has joined #mlpack
seewishnew has joined #mlpack
seewishnew has quit [Remote host closed the connection]
seewishnew has joined #mlpack
seewishnew has quit [Ping timeout: 250 seconds]
seewishnew has joined #mlpack
rf_sust2018 has quit [Quit: Leaving.]
ahmedmaher has quit [Ping timeout: 258 seconds]
ahmedmaher has joined #mlpack
i8hantanu has joined #mlpack
seewishnew has quit [Remote host closed the connection]
seewishnew has joined #mlpack
robb9 has joined #mlpack
< robb9> Mulx10: thank you! I finally understand
< robb9> also, does RNN extend a base class? I want to make it an instance variable but I can't unless I specify rho, which I don't want to do yet
seewishnew has quit [Ping timeout: 264 seconds]
seewishnew has joined #mlpack
seewishnew has quit [Remote host closed the connection]
seewishnew has joined #mlpack
bhavya01 has joined #mlpack
seewishnew has quit [Ping timeout: 264 seconds]
seewishnew has joined #mlpack
bhavya01 has quit [Quit: Ex-Chat]
bhavya01 has joined #mlpack
robb9 has quit [Quit: Page closed]
sreenik has joined #mlpack
bhavya01 has quit [Ping timeout: 268 seconds]
rf_sust2018 has joined #mlpack
seewishnew has quit [Remote host closed the connection]
seewishnew has joined #mlpack
seewishnew has quit [Remote host closed the connection]
seewishnew has joined #mlpack
seewishnew has quit [Remote host closed the connection]
seewishnew has joined #mlpack
seewishnew has quit [Ping timeout: 250 seconds]
bhavya01 has joined #mlpack
paul____ has joined #mlpack
rf_sust2018 has quit [Quit: Leaving.]
< paul____> Hi! In the context of gsoc, may I ask if there are have been any strong proposals for the mlpack-Tensorflow-pytorch translator project yet? I am very interested in the project, however if some rockstars already applied to that project I might apply for a different one to have a realistic chance.
< rcurtin> paul____: I have no idea what the quality of the proposals are, but there has not been much discussion about that project
< rcurtin> there's also been basically no discussion of the 'mlpack on resource constrained devices' project
seewishnew has joined #mlpack
< paul____> Thanks for the answer rcurtin! I would definitely be more interested in the 'mlpack translator project' than the 'mlpack on resource constrained devices' project. I suppose I will put together a bit of a pitch / preliminary proposal for that project then and will eventually suggest it in this channel. Again, thanks for the quick response!
i8hantanu has quit [Quit: Connection closed for inactivity]
paul____ has quit [Quit: Page closed]
< rcurtin> sounds good
pd09041999 has joined #mlpack
rf_sust2018 has joined #mlpack
robb9 has joined #mlpack
< robb9> I'm getting some weird behavior with mlpack::data::load
< robb9> I have 256 lines in a csv file, with each line having 4 features
< robb9> however, upon loading it, the n_cols is equal to 2570 somehow, not 256.
< robb9> just using stock mlpack::data::Load("data.csv", filedata, true);
< robb9> ah, I think I found the problem
robb9 has quit [Client Quit]
ahmedmaher has quit [Read error: Connection reset by peer]
astadnik has joined #mlpack
ahmedmaher has joined #mlpack
< astadnik> Hello guys)
seewishnew has quit [Remote host closed the connection]
seewishnew has joined #mlpack
seewishnew has quit [Ping timeout: 264 seconds]
leilee has joined #mlpack
big has joined #mlpack
saksham189 has joined #mlpack
< big> hey, I have an error regarding using Predict() with my recurrent neural net
< big> it's a Mat::submat() out of bounds error
< big> I have rho set to my amount of datapoints per timestep
< big> and the amount of rows in my dataset is equal to the amount of inputs in the first layer
< big> what else could be the problem?
< big> the error specifically happens when I use Predict(). I am constructing my datapoint correctly and all members are populated
bhavya01 has quit [Remote host closed the connection]
pd09041999 has quit [Ping timeout: 250 seconds]
leilee has quit [Ping timeout: 256 seconds]
< big> I don't get a matrix multiplication error or anything, just a submat out of bounds
pd09041999 has joined #mlpack
pd09041999 has quit [Excess Flood]
pd09041999 has joined #mlpack
pd09041999 has quit [Max SendQ exceeded]
pd09041999 has joined #mlpack
saksham189 has quit [Ping timeout: 256 seconds]
< big> I think it's because I'm adding the layers wrong
pd09041999 has quit [Ping timeout: 250 seconds]
saksham189 has joined #mlpack
seewishnew has joined #mlpack
seewishnew has quit [Remote host closed the connection]
seewishnew has joined #mlpack
sreenik has quit [Ping timeout: 256 seconds]
ahmedalamir has joined #mlpack
ahmedmaher has quit [Ping timeout: 268 seconds]
pd09041999 has joined #mlpack
pd09041999 has quit [Max SendQ exceeded]
rf_sust2018 has quit [Quit: Leaving.]
saksham189 has quit [Ping timeout: 256 seconds]
saksham189 has joined #mlpack
sreenik has joined #mlpack
< sreenik> Before I knew about ensmallen I used to make use of mlpack's default SGD optimizer with AdamUpdate. Now with ens::Adam(...) however the accuracy has come down to 10% from 99% with similar parameters. Am I missing anything obvious?
seewishnew has quit [Ping timeout: 250 seconds]
seewishnew has joined #mlpack
< rcurtin> sreenik: the code didn't change, so that's quite surprising to me; are you sure nothing else in your code changed?
seewishnew has quit [Remote host closed the connection]
seewishnew has joined #mlpack
< sreenik> I don't think so. I will investigate a bit more. If the optimizer code hasn't changed, then I hope something is wrong with my code. I will open an issue in case I can't find out
seewishnew has quit [Ping timeout: 268 seconds]
seewishnew has joined #mlpack
< big> It looks like there's a very specific way to add layers to an RNN...
< big> Is it true that I have to make a separate object, say, "recurrent", add layers to that, and then add that to the model?
< sreenik> The parser is ready (ahh finally!). I am trying it out with different models. No errors so far just this thing regarding the accuracy. Also, I am not very familiar with using RNNs with mlpack, so I have left rnn features out (incorporating them later won't be much of a task though). Should I file a PR or upload it to my own repo and share the link after documenting? I am not too sure where this parser should reside.
< sreenik> Also, it needs extensive testing and maybe some structuring before it can be merged.
< big> is there a debug mode that I can make armadillo use? I need verbosity to see where in Predict() I'm getting the out of bounds error
seewishnew has quit [Remote host closed the connection]
seewishnew has joined #mlpack
seewishnew has quit [Ping timeout: 268 seconds]
< sreenik> rcurtin: Sorry it was my mistake. It's alright now. Ensmallen is working exactly as it is supposed to. The dataset had been corrupted somehow :)
vivekp has quit [Ping timeout: 244 seconds]
Suryo has joined #mlpack
< Suryo> zoq, rcurtin: I had submitted my proposal last week (I've been busy with my studies since) and was wondering if you guys got a chance to look at it. I just wanted to know what to add to it, or if there's anything I need to remove from it...
Suryo has quit [Client Quit]
< zoq> big: model.Add<Recurrent<> >(Add<>(), ...); works as well
< zoq> big: About debugging what about using gdb to step through the code?
< zoq> Suryo: hm, I think I have missed the application, let's see.
< big> Well I have been getting a Matrix::submat out of bounds() error when I try to Train or Predict using my model
< big> I've tried using gdb but I haven't gotten anything useful
< big> I have 2 timesteps (rho=2, slices of cube is 2), 4 features (4 rows in cube) and 16 columns per step (16 datapoints per step)
< zoq> big: hm, it should be possible to get the line that caused the error using gdb, but I agree that might be time consuming
< zoq> big: The shape sounds resaonable to me
< big> all of my data is populated but when I try to Train or Predict it gives the same error
< zoq> right at the first iteration?
< big> I'm not sure, how would I go about checking which iteration it's on? i'm assuming more gdb
< big> I'll see if I can tell and I'll let you know
< zoq> You could get some more output if you define ENS_PRINT_INFO and ENS_PRINT_WARN
< big> Alright, I'll try that. Making armadillo verbose didn't give any helpful output
< big> Would it be helpful if I told you that it printed absolutely nothing with both enabled?
< big> I'm going to try reinstalling armadillo and mlpack
< zoq> Kinda, I guess it returns before the first iteration.
< zoq> Not sure, perhaps you can create a simple model including the data that can be used to reproduce the issue? In case you don't get anything reasonable from the gdb output.
< big> Would having only the following layers maybe cause a problem somehow? Linear(4,10), GRU(10,10), Linear(10,3)
< big> Yeah I've taken everything out and have the simplest possible model right now for testing purposes
< zoq> the structure looks alright to me, probably something strange with the data encoding
< zoq> if you like you can post the link here and I can take a look at it once I have a chance, probably not before tomorrow
< big> Alright, if this reinstall doesn't work then I'll do that. Thanks
saksham189 has quit [Quit: Page closed]
< rcurtin> Suryo: I have a bit of time set out tonight to look at proposals, I'll be sure to look at yours also
< zoq> Currently looking into the application.
ahmedmaheralamir has joined #mlpack
ahmedalamir has quit [Read error: Connection reset by peer]
ahmedmaher has joined #mlpack
ahmedalamir has joined #mlpack
ahmedmaheralamir has quit [Ping timeout: 246 seconds]
ahmedmaher has quit [Ping timeout: 268 seconds]
ahmedmaher has joined #mlpack
ahmedalamir has quit [Ping timeout: 246 seconds]
big has quit [Ping timeout: 256 seconds]
big has joined #mlpack
sreenik has quit [Quit: Page closed]
< big> I'll do some additional gdb work but I can't seem to figure out what the problem is
anidh1997 has joined #mlpack
< big> zoq: Also it's not my installation, I just tried on a different machine with the same error
ahmedmaher has quit [Ping timeout: 245 seconds]
big has quit [Ping timeout: 256 seconds]
anidh1997 has quit [Quit: Page closed]
Suryo has joined #mlpack
< Suryo> zoq: thanks for taking a look at my application and commenting. I think that a lot of your comments are actually addressed in different parts of the application, but I'll review it to make sure that all the points you've raised are clear in the application.
Suryo has quit [Client Quit]
ahmedmaher has joined #mlpack
big has joined #mlpack
< big> zoq: tracing with GDB and I may have found the culprit. for some reason, it thinks the batchSize is 256 instead of rho which is what I gave it
< big> mlpack::ann::RNN<...>, ...>::Predict (this=0x7fffffffd3c0, predictors=..., results=..., batchSize=256)
< big> is the default value 256 or something?
ahmedalamir has joined #mlpack
ahmedmaher has quit [Read error: Connection reset by peer]
< big> Nevermind, that's not the error
big has quit [Quit: Page closed]
ahmedmaheralamir has joined #mlpack
ahmedalamir has quit [Ping timeout: 255 seconds]