Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Repo info
Activity
Eyad Sibai
@eyadsibai
hello
Alex Rogozhnikov
@arogozhnikov
hi :)
Eyad Sibai
@eyadsibai
how r u
Tatiana Likhomanenko
@tlikhomanenko
Fine!
Eyad Sibai
@eyadsibai
I was wondering about the plan and the future of rep
Alex Rogozhnikov
@arogozhnikov
Hi, Eyad. Andrey told me about your thoughts.
I remember you suggested to split current rep to core and separate libraries to somehow make life easier to people who want to try REP.
I believe this is harder to keep splitted parts updated and tested than the whole thing,
but we can really simplify things (and we started to):
pip install rep now works, it installs rep and minimal pythonic dependencies,
but no libraries. So the core things now may be tried without much efforts.
When user wants to try some other library, he installs it. We will have to reorganize documentation according to it.
Arseny Kravchenko
@arsenyinfo

Hi @arogozhnikov,
Thanks for the new approach - as soon as pip install rep works it much easier to start during leisure time.

REP is really good, thanks for your job!

Arseny Kravchenko
@arsenyinfo

several questions if you don't mind:

thanks

Alex Rogozhnikov
@arogozhnikov
Hi, @arsenyinfo, thanks for comments.
Alex Rogozhnikov
@arogozhnikov

  • Why pandas. Well, actually, this is the most widespread format with named columns. Initially we also planned to use it's eval (but turned out it's quite useless). You're right that there are currently problems with sparse data (I don't know how to resolve it, because AFAIR numpy doesn't have sparse array with named columns). Ideas welcome.

  • RandomizedGridSearch is not implemented yet, but I have in plans to add one more ParameterGenerator to cover this.

Arseny Kravchenko
@arsenyinfo
Got it - thanks again
Semen Budenkov
@Semen52
Hi, can you help me visualize report? How I have to call show() function?
Alex Rogozhnikov
@arogozhnikov
@Semen52 , report itself is not visualizable.
You need to visualize output of it's methods like
roc_curves = report.roc() # this method returns plotting object, which can draw itself. You can pickle it.
roc_curves.plot() # this means - draw the output
Semen Budenkov
@Semen52
@arogozhnikov thanks, plt.show() works :-)
Alex Rogozhnikov
@arogozhnikov
Ah, this means you aren't using ipython's inline plotting :)
Arseny Kravchenko
@arsenyinfo
hey folks,
would you advise how do I get auth for http://root.cern.ch/git/root.git?
Alex Rogozhnikov
@arogozhnikov
Do you need auth?
I tried git clone http://root.cern.ch/git/root.git and it seems to work.
Arseny Kravchenko
@arsenyinfo
sorry - may be twas a temporary issue
Max Schlupp
@mschlupp

hey all, I am currently playing around with rep and until now like it a lot!
however I think there is a bug with the NeuroLabBase classe.
If you try to train a LVQ it gets an argument 'tansf' from 'BASIC_PARAMS'. That makes perfect sense for MLP default class, but not for the LVQ, as the neurolab function newlvq() doesn't allow for it.

you should be able to reproduce the error with:
```from rep.estimators import NeurolabClassifier
import pandas as pd

clf = NeurolabClassifier(net_type='learning-vector')

ds = pd.DataFrame()
ds['feature1']=[0,1,2,3,4,5]
ds['feature2']=[5,7,2,4,7,9]
ds['y']=[0,0,0,1,1,1]

clf.fit(ds[['feature1','feature2']],ds['y'])```

not sure, if this channel is still active. if not, I'll open an issue on github

Andrey Ustyuzhanin
@anaderi
@mschlupp , this channel is active, but to keep discussion focused it is better to create an issue plus till 4th of May there are national holidays in Russia, not everybody is around )
Alex Rogozhnikov
@arogozhnikov
Hi, thanks for reporting, I've opened an issue for this (imail seems to send gitter notifications to spam, sorry)
Andrey Ustyuzhanin
@anaderi
@arogozhnikov , have a look at #83, please
Alex Rogozhnikov
@arogozhnikov
@anaderi ok, I've added notebook, please try my branch build_improvements
if you have some trivial setup to test workability of jupyterhub, I suggest adding it to circle.yml
Andrey Ustyuzhanin
@anaderi
@arogozhnikov , build_improvements works fine now, but it doesn't run git clone command (and ends up with empty /notebooks folder). it should be executed by start.sh just before running jupterhub-singleuser in start.sh)
Alex Rogozhnikov
@arogozhnikov

@anaderi It is executed:
https://github.com/arogozhnikov/rep/blob/1611caca54ec5d23c82617e94bec9ea252c0e860/ci/start_jupyter.sh#L30

Possible problems: folder rights, github authorization.

Andrey Ustyuzhanin
@anaderi
it didn't clone since git was asked to clone in /notebooks which exists. there are several ways to avoid it: http://stackoverflow.com/questions/5377960/whats-the-best-practice-to-git-clone-into-an-existing-folder
Alex Rogozhnikov
@arogozhnikov
ok, I've added a more reasonable hack (pushd / popd, this changes behavior of git), please try again
Furiously Curious
@FuriouslyCurious
Hey guys, is anyone from REP team at the NIPS conference?
Andrey Ustyuzhanin
@anaderi
@FuriouslyCurious hey, yup, I'm here. will have poster at Challenges in Machine Learning workshop this Friday. Otherwise feel free to find me anytime)
Furiously Curious
@FuriouslyCurious
@anaderi Couldn't find you on Whova conference app. Can you please add me on WhatsApp or Whova? +14128973700
Andrey Ustyuzhanin
@anaderi
Just sent whova request for joining nips. Don't use whatsapp. Viber and telegram) sent regular text though)
@FuriouslyCurious see above)
Furiously Curious
@FuriouslyCurious
👍
Kevin
@kevinkraft
Hi. I have a question about yandex/rep. I'm using the "efficiencies" function in rep/rep/report/classification.py. I want to get the matplotlib objects from this function and change the axis and do various other things with it. But I can't figure out how to access it from the classes in rep/rep/plotting.py
Ankit Jha
@ankitjha97
Hello Everyone.
I want to contribute to rep.Can anyone tell where should I start from?