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
caiojcarvalho has joined #mlpack
kevinmathew has quit [Ping timeout: 250 seconds]
wannabeOG has joined #mlpack
< wannabeOG> zoq: Yes I was talking about them. Just had a quick look through them
kevinmathew has joined #mlpack
< wannabeOG> zoq: Any specific pointers regarding them or anything to get started?
wannabeOG has quit [Quit: Page closed]
< kevinmathew> Anything. Also I have a question. Does mlpack offer projects for GSoC which are more machine learning specific and less c++ specific?
< jenkins-mlpack2> Project docker mlpack weekly build build #24: STILL UNSTABLE in 6 hr 49 min:
kevinmathew has quit [Ping timeout: 250 seconds]
kevinmathew has joined #mlpack
kevinmathew has quit [Quit: Leaving]
kevinmathew has joined #mlpack
kevinmathew has quit [Client Quit]
caiojcarvalho has quit [Ping timeout: 250 seconds]
< jenkins-mlpack2> Project docker mlpack nightly build build #161: STILL UNSTABLE in 2 hr 56 min:
ayesdie has joined #mlpack
ayesdie has quit [Client Quit]
ayesdie has joined #mlpack
< ayesdie> I looked at cf_model and its implementation. Since I was not familiar with the concepts of boost::variant, I also looked at how it works and now the code seems well understandable.
< ayesdie> Since variant is new to me, I don't have a good plan on how to implement rating normalization other than hard coding all the types and modifying member functions of CFModel which might be unnecessary.
< ayesdie> Any lead that might help?
ayesdie has quit [Quit: ayesdie]
vivekp has quit [Ping timeout: 245 seconds]
vivekp has joined #mlpack
mlpackApprentice has joined #mlpack
< mlpackApprentice> Hello. I'm trying to call the Prediction() method on 1 sample of time series with an RNN, and I expect to receive a prediction for the next time step.
< mlpackApprentice> I’m running an RNN with a FastLSTM (12 inputs, 5 outputs). The way I trained it is by dividing a dataset of 24000 values into 1000 time series of 24 timesteps, so with a 12x1000x24 cube. The “predictors” input in Train() was this cube, and the “responses” input was the same cube, but where every value was shifted one timestep forward (because you train the RNN based on what the next value is).
< mlpackApprentice> Then I tried to make a prediction by calling Predict() on 1 sample with 24 time steps (so a 12x1x24 cube), and a Response cube without any specified measurements to be filled with values by the Predict method itself.
< mlpackApprentice> The problem is, this crashes my program. I also tried specifying the dimensions of the Response cube (5x1x1), because my RNN has 5 outputs. But after calling the Predict method, my Response cube has turned into a 5x0x0 cube. Should I pass in larger cubes?
< mlpackApprentice> Thank you in advance for reading.
< mlpackApprentice> Never mind, it appears an earlier std::move() emptied the Predict method's input.... but looking at the output I have another question :)
< mlpackApprentice> I was actually expecting a prediction for the next hour, but it gives 24 values as well, just like the input. Are these predictions for each of the 24 input values? And so if I want the prediction of the 25th value, I should just take the 24'th slice of the Prediction output cube?
witness has quit [Quit: Connection closed for inactivity]
< rcurtin> ayesdie: you might consider taking a look at the use of boost::variant in, e.g., src/mlpack/methods/range_search/rs_model.hpp
< rcurtin> maybe this could help give you an idea of how to do it
mlpackApprentice has quit [Ping timeout: 256 seconds]
< zoq> wannabeOG: I think a good step to get familiar with the codebase is to fix some issue or improve/extend an existing; so the github issue page should be helpful.
< zoq> kevinmathew: We are open for ideas, so if you have something in mind, let's chat; also some projects could contain a research component like the RL project.
< zoq> mlpackApprentice: You can just look at the 25th value, however it would be nice to just get the next value, which is possible for the FFN but not for the RNN class; I'll see if I can open a PR in the next days.
ImQ009 has joined #mlpack
ImQ009 has quit [Read error: Connection reset by peer]