mlpack IRC logs, 2018-11-06

Logs for the day 2018-11-06 (starts at 0:00 UTC) are shown below.

November 2018
--- Log opened Tue Nov 06 00:00:03 2018
02:18 -!- mrohit[m] [mrohitmatr@gateway/shell/] has quit [Ping timeout: 264 seconds]
02:41 -!- mrohit[m] [mrohitmatr@gateway/shell/] has joined #mlpack
02:42 -!- cjlcarvalho [] has joined #mlpack
03:39 -!- mrohit[m] [mrohitmatr@gateway/shell/] has quit [Ping timeout: 252 seconds]
04:33 -!- mrohit[m] [mrohitmatr@gateway/shell/] has joined #mlpack
09:49 -!- cjlcarvalho [] has quit [Ping timeout: 264 seconds]
10:22 < davida> zoq: I created a GitHub project for the saving of my work on the course translation to MLPACK. I have actually managed to resolve the 'access violation' issue that I was having which I suspect was due to the sizing of some of my layers.
10:23 < davida> zoq: However I would still like you to take a look at the code since it now matches pretty much exactly (as far as I can tell) to the Coursera exercise however the MLPACK version cannot train anywhere near
10:24 < davida> as well as the Python/Tensorflow model.
10:24 < davida> My code is here:
10:25 < davida> There is only this Conv work in there so far. I am also uploading my Jupyter Notebook of the work I did during the course. You can see how well it trains and also the model used.
10:25 < davida> As I am still fairly new to MLPACK, I may well not have translated the model well so I would appreciate you input on this to set me in the right direction.
10:29 < davida> The C++ code is in Week1Main.cpp and the data required is in the ./data directory. Also, note the code uses the HDF5 and H5CPP libraries to import the data.
10:31 -!- vivekp [~vivek@unaffiliated/vivekp] has quit [Ping timeout: 268 seconds]
10:37 -!- vivekp [~vivek@unaffiliated/vivekp] has joined #mlpack
10:53 < davida> zoq: I added a file to show the training results during 100 epochs (TrainingResults.txt). Best I can achieve is ~30% and the Python/Tensorflow gets to 97% with same amount of training.
10:56 -!- cjlcarvalho [~caio@2804:d47:1d0d:a700:95d6:5d67:8b41:79ad] has joined #mlpack
11:55 -!- ayesdie [~ayesdie@] has joined #mlpack
12:04 -!- Amit__ [01ba9174@gateway/web/freenode/ip.] has joined #mlpack
12:13 < Amit__> Hello, everyone! I am Amit and I am a sophomore in college pursuing computer science, I'm actually new to opensource and I would like to start my journey with mlpack,can anyone guide me how to get started I have fair knowledge of c, c++ (STL) and python..).
12:24 -!- Amit__ [01ba9174@gateway/web/freenode/ip.] has quit [Quit: Page closed]
12:29 < zoq> davida: Thanks, I'll take a look at the code later today.
12:32 < ayesdie> Do I need to take care of anything while running `/main_tests/cf_test.cpp`, because it seems lime its failing 7 out of 13 tests.
12:36 < ayesdie> It fails all the test after `CFModelReuseTest` with the message as `[FATAL] Invalid value of specified ( ); neighborbood must be less than or equal to the number of users!`. I think I'm missing something here?
12:49 -!- ayesdie [~ayesdie@] has quit [Ping timeout: 252 seconds]
12:51 -!- ayesdie [~ayesdie@] has joined #mlpack
13:01 < rcurtin> ayesdie: how are you running the tests?
13:02 < ayesdie> i'm still trying to figure out why 7 of the default tests are failing
13:05 < ayesdie> i'll tell the details, `mlpack_test --run_test=CFMainTest --logger=HRF,all --color_output=false --report_format=HRF --show_progress=no` after i build mlpack_test
13:08 < rcurtin> try running it from the build directory, not the bin/ directory, as 'bin/mlpack_test'
13:08 < rcurtin> my guess is it's just not able to load the right datasets for testing because the working directory is different than expected
13:12 < ayesdie> It's running now :D thank you very much, I'm using CLion so I was using the default debug method from it, so perhaps that was giving the issue
13:23 < ayesdie> Rather than doing the invalid interpolation and neighbor search check separately, should I move them into one existing test of `CFAlgorithmBoundTest`?
13:28 < rcurtin> glad you got it working :)
13:28 < rcurtin> I think maybe it would be better to split the tests into different test cases
13:28 < rcurtin> that can make it easier for debugging later... instead of one big test that failed somewhere in the middle, it's much easier to see what has failed in a short test
13:29 -!- ayesdie [~ayesdie@] has quit [Quit: ayesdie]
13:29 -!- ayesdie [~ayesdie@] has joined #mlpack
13:34 -!- ayesdie [~ayesdie@] has quit [Client Quit]
13:34 -!- ayesdie [~ayesdie@] has joined #mlpack
13:42 < rcurtin> ayesdie: just in case you didn't see my response the channel is logged at :)
14:10 -!- ayesdie [~ayesdie@] has quit [Quit: ayesdie]
16:09 < ShikharJ> rcurtin: Interesting, similar for Probability and Multivariate Calculus?
16:11 < ShikharJ> davida: I can't seem to pin-point an issue within the Pooling layers. Please open an issue for this with a minor code sample?
16:12 < davida> ShikharJ: Sorry, which topic are you referring to?
16:13 < ShikharJ> davida: The one which you get while getting the convolution output as 65x65x8.
16:16 < davida> ShikharJ: In my reply to zoq earlier today I mentioned that I had found the 'allocation violation' issue with my layer values so it is not an issue any more. The code runs, but the main issue I have is that the exact same model trains to 97% in Python/Tensorflow but only to 30% accuracy on MLPACK.
16:17 < davida> I posted link to my code earlier. zoq already mentioned he will take a look at it but your input is also welcome.
16:18 < davida> My goal is to get the Coursera exercises working on MLPACK and achieving the same (or at least a similar) result.
16:55 -!- ayesdie [~ayesdie@] has joined #mlpack
17:18 < ShikharJ> davida: Ah okay, sorry I skipped the messages in between :)
17:40 -!- cjlcarvalho [~caio@2804:d47:1d0d:a700:95d6:5d67:8b41:79ad] has quit [Ping timeout: 252 seconds]
17:51 -!- ayesdie [~ayesdie@] has quit [Quit: ayesdie]
17:52 -!- ayesdie [~ayesdie@] has joined #mlpack
17:56 -!- ayesdie [~ayesdie@] has quit [Client Quit]
17:57 -!- ayesdie [~ayesdie@] has joined #mlpack
17:57 -!- ayesdie [~ayesdie@] has quit [Remote host closed the connection]
18:46 -!- ayesdie [~ayesdie@] has joined #mlpack
19:03 -!- ayesdie [~ayesdie@] has quit [Quit: ayesdie]
22:03 < zoq> davida: To make the comparison easier, do you mind to use ReLULayer instead of LeakyReLU? If we use the same model we should be able to use the same weights as well, so the output of the Forward pass should be the same.
22:07 < zoq> davida: Btw. the file is really nice; super easy to follow.
22:09 < zoq> davida: I think we can see a trend in the accuracy over the last iterations.
22:57 -!- cjlcarvalho [~caio@2804:d47:1d0d:a700:95d6:5d67:8b41:79ad] has joined #mlpack
--- Log closed Wed Nov 07 00:00:04 2018