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/
< jeffin143[m]>
zoq (@freenode_zoq:matrix.org): saw your comment on 165 , about compute accuracy function , is there any function written as of now to compute accuracy in mlpack ???
< jeffin143[m]>
If I am not wrong the mse functions currently are a layer
< jeffin143[m]>
So if I want to find out the error of decision tree classification model ? I should write a custom one ?? And cannot use the functions in loss ???
< jeffin143[m]>
We could use the metrics/ ?? May be precision and recall ???
< jeffin143[m]>
But then that could be used for all of the model linear regression or even Ann and others
< zoq>
jeffin143[m]: Some methods already return some loss, but lr is not one of those methods, we could use the metrics of the cv namespace.
< zoq>
jeffin143[m]: In fact think that makes sense to avoid codeduplication.
< jeffin143[m]>
When we call optimize function in lr, and pass print loss , what loss does it print ?
< zoq>
jeffin143[m]: The issue is that only works in the training process
< zoq>
jeffin143[m]: since the lr method has no method yet to return objective in combination with a validation set.
< zoq>
jeffin143[m]: I just used lr as an example in the ensmallen PR, since there is no simple network we can use for testing.
< jeffin143[m]>
Zoq : So suppose , consider this case , I have trained a model , and then predicted validatin set ouput and want to see the accuracy , so I have to make a custom one , because we can't use mse ?
< jeffin143[m]>
Sorry for pestering
< jeffin143[m]>
But I am caught between metric and loss function
< jeffin143[m]>
Also zoq : which time zone are you in ?
< zoq>
jeffin143[m]: So in case of lr, you can once trained either use: .Classify() and afterwards calculate the accuracy, or you can directly use .ComputeAccuracy(). However ComputeAccuracy is not a method that is present in every method same for Classify().
< zoq>
jeffin143[m]: So you have to check which method is implemented.
< zoq>
jeffin143[m]: jeffin143[m]: another example is the NBC class, which only provides Classify().
< zoq>
jeffin143[m]: CET
< jeffin143[m]>
Ohh I see , that makes my work very difficult , so now the problem is for visual board , and we have to plot the graph then we need to know accurcy , but now the problem is not every method provide that
< zoq>
jeffin143[m]: Hm, what about we use
< zoq>
SFINAE and enable_if
< zoq>
to check if there is a .Classify() function
< jeffin143[m]>
But I guess , if we can asure all of the dl methods provide that would workout too , because we don't often plot for ml models , but do so for dl since we train it in cycles and batches and so on , so there would be multiple times loss decreasing , but for ml I guess we need not so the interim loss
< jeffin143[m]>
For fnn< loss_functin>
< jeffin143[m]>
We can specify the loss function and I guess that loss gets printed in prinloss calback , because it simply the objective
< jeffin143[m]>
Yes we can definitely do it , Thanks
< jeffin143[m]>
I will come up with something then
< zoq>
I can also sketch up a simple callback function that computes the accuracy, if that is helpful.
< zoq>
But maybe it's a good idea first to list the methods we like to visulize and see what methods are there we can use.
< zoq>
There are different aspects we have to account for, if we like to support every method, so a list/table would be helpful.
< jeffin143[m]>
Have a good day
< zoq>
jeffin143[m]: Sure, happy to help.
togo has joined #mlpack
MeganWillia has joined #mlpack
MeganWillia has quit [Client Quit]
< chopper_inbound[>
Hi, I wanted to know if we can use `field` class if we need multiple matrices as input to `Forward` function, or we need to concatenate them into single matrix ?
< zoq>
chopper_inbound[: Right now you have to vectorize the input, meaning .col() = sample of an arma::mat
< zoq>
chopper_inbound[: But we could add another method to provide such an interface.
< chopper_inbound[>
zoq: Okk! I didn't knew about this method of vectorisation! Thanks for the quick response!
< PranavReddyP16Gi>
Hey everyone I have a problem, I downloaded mlpack-3.3.1 tar file and tried building from source. I've never run into this problem before and on running make I'm getting the error make: *** [Makefile:158: all] Error 2 and I have no clue how to figure out what's wrong. Can someone help me out? Thanks.
joty has joined #mlpack
< joty>
Anyone got a link to instructions for using mlpack with vscode in win10 ?
< joty>
I've googled around but I've only been able to find instructions for visual studio
< joty>
Seems like nuget for vscode does not support c++, while it is supported in nuget for visual studio
< zoq>
joty: You can use nuget.exe to get the necessary libs.
< zoq>
joty: After that I think VSCode allows to configure the project using cmake?
< zoq>
joty: If that is the case you can follow the tutorial and set the correct path to the libs and you should be good to go.
< joty>
zoq: thanks, I'll give it a try
< zoq>
Param-29Gitter[m: Hm, can you run the command with the verbose option enabled?
ImQ009 has quit [Quit: Leaving]
joty has quit [Remote host closed the connection]
< PranavReddyP16Gi>
@zoq it's been stuck on 40% for a while guess I'll have to check again tomorrow morning :)
< PranavReddyP16Gi>
It never took this long though I wonder why this is happening
< zoq>
Param-29Gitter[m: Do you have enough space, memory?
< zoq>
Maybe only use a single core, -j1
< PranavReddyP16Gi>
And i dont
< PranavReddyP16Gi>
*And I don't know if this is required but Makefile2 line 6654 is : $(MAKE) -f src/mlpack/tests/CMakeFiles/mlpack_test.dir/build.make src/mlpack/tests/CMakeFiles/mlpack_test.dir/build