< jenkins-mlpack> Project mlpack - nightly matrix build build #457: STILL UNSTABLE in 1 hr 32 min: http://big.cc.gt.atl.ga.us:8080/job/mlpack%20-%20nightly%20matrix%20build/457/
koderok has joined #mlpack
Anand has joined #mlpack
koderok has quit [Ping timeout: 240 seconds]
< Anand> hi!
< Anand> We donot generate the "output.csv" file for the predicted labels as of now. Right?
< marcus_zoq> Hello, if we call the nbc method, we automatically generate the 'output.csv' file.
< marcus_zoq> The file should be located in the repository root.
< marcus_zoq> But keep in mind that the destructor of the nbc script deletes all files after the runtime test.
< marcus_zoq> So if you comment out the 'def __del__(self):' function you should see the 'output.csv' file.
< Anand> I am running the benchmarks through "make run LOG=Flase" in a small config file with only iris dataset
< Anand> I do not see any output.csv even after commenting that function
< marcus_zoq> Just to be clear if you run: 'make run LOG=False CONFIG=small_config.yaml' you get some runtime information right?
< Anand> I have renamed the config file
< Anand> small_config is config now
< marcus_zoq> okay, but you get the runtime information with the config file?
< Anand> yes
< Anand> but no output files
< marcus_zoq> and you delete/comment out the complete 'def __del__(self):' function in the mlpack/nbc.py file right?
< Anand> yes
< marcus_zoq> That's weird If I run 'git status' there is an untracked file that is called output.csv.
< Anand> can you see that file?
< marcus_zoq> yeah
< Anand> weird indeed!
< Anand> Ok. Let me see
< marcus_zoq> can you can the following command inside the repo root: 'nbc -t datasets/iris_train.csv -T datasets/iris_test.csv -v'
< marcus_zoq> run
< Anand> yes I got the file now
< Anand> what's wrong with make?
< Anand> Also, why does the file contain 0,1 and 2 as labels while the train file has 1,2 and 3?
< marcus_zoq> maybe a mapping bug, let's check the naive_bayes_classifier_impl.hpp file.
< marcus_zoq> Can you add the following line: 'exit()' right after 's = subprocess.check_output(cm....' in the nbc.py file.
< Anand> done
< marcus_zoq> and if you delte the output.csv file and restart the make command you can find a output.csv file?
< Anand> no
< Anand> Also the output I get on my terminal is different in the two cases when I run make and when I run nbc -t datasets/iris_train.csv -T datasets/iris_test.csv -v directly. Is that fine?
< marcus_zoq> No, can you show me your config file?
< Anand> PCA: 22 run: true 23 script: methods/mlpack/pca.py 24 format: [csv, txt] 25 datasets: 26 - files: ['datasets/iris_train.csv','datasets/iris_test.csv','datasets/iris_test_labels.csv'] 27 options: '-v'
< Anand> the numbers are line numbers. Ignore them
< marcus_zoq> ah, you are running the pca.py script.
< Anand> My God! I spent more than an hour on that! :P
< Anand> Before coming to you! :P
< Anand> NBC has got into my mind, I suppose
< Anand> Thanks, btw
< marcus_zoq> But I think, you found a bug, in the nbc labels mapping.
< marcus_zoq> No problem.
< Anand> yes, that is a bug indeed
< Anand> where to fix it?
< marcus_zoq> In the normalize_labels_impl.hpp file.
< Anand> Ok
< Anand> You need to fix it then, I guess
< Anand> What is the fix btw?
< Anand> The index 'j' need to begin from 1 ?
< marcus_zoq> naywhayare wrote the mapping code but let's see.
< marcus_zoq> Okay, I found, the bug, before we commit the changes, we have to check if the change pass the tests.
< Anand> what was the bug?
< marcus_zoq> We save the labels before mapping but we have to save the labels after the mapping.
< marcus_zoq> data::Save(outputFilename, results, true, false); -> data::Save(outputFilename, rawResults, true, false);
< Anand> is the last parameter ('false') optional?
< Anand> I have data::Save(outputFilename, results, true);
< Anand> in nbc.py
< marcus_zoq> nbc.py or nbc_main.cpp?
< Anand> sorry, nbc_main.cpp
< marcus_zoq> The last parameter indicates if we have to transpose the matrix/vector -> True transpose
< marcus_zoq> The default is true, but I then we write the labels in a single line.
< marcus_zoq> Actually the Save function provide four paramaters: filename, matrix, error message on, transpose.
< Anand> alright
< Anand> If we make it false, the labels will be all in different lines because of the transpose. right?
< marcus_zoq> data::Save(outputFilename, rawResults, true, false); -> one label per line
< Anand> Ok.
< marcus_zoq> data::Save(outputFilename, rawResults, true, true); -> the complete vector in one line
< Anand> correct
< marcus_zoq> Did you already receive the GSoC welcome package?
Anand has quit [Ping timeout: 240 seconds]
Anand has joined #mlpack
< Anand> Marcus : Yes, I received the welcome package!
< naywhayare> Anand: sorry about the bug :(
< naywhayare> but thanks for fixing it :)
< Anand> naywhayare : No problem. And btw marcus fixed it, I pointed it out! :)
< naywhayare> ah, ok. I only read the commit message and made assumptions that were apparently wron
< naywhayare> *wong
< naywhayare> **wrong
< Anand> ok. That's fine!
< Anand> :)
Anand has quit [Ping timeout: 240 seconds]
udit_s has joined #mlpack
udit_s has quit [Quit: Page closed]
udit_s has joined #mlpack
< naywhayare> udit_s: hello there! you sent me a message sometime back but then I didn't see you in IRC after that
< udit_s> naywhayare: yeah. was travelling back home. Did you get my email though ?
< naywhayare> no, I don't think I did; where did you send it?
< naywhayare> I got one on April 16th and April 28th but I don't think that's what you're referring to
< udit_s> no I sent you one on the 13 or 14 of may
< naywhayare> I don't think I got it :-(
< udit_s> at the ratml.org address. so which email id do you use ? I'll just resend it...
< naywhayare> yeah, ryan@ratml.org. I wonder why it didn't go through. if it doesn't go through this time, I'll look through the logs to see if I can figure anything out
< naywhayare> ok, that went through fine
< udit_s> hmm..
< udit_s> so basically I reached home yesterday - been setting up system slowly since then.
< naywhayare> ok. I'm helping fix a car today so I'll be in and out, but if you have any questions I will do my best to answer them quickly
< udit_s> cool.
udit_s has quit [Quit: Page closed]