Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Activity
    abhishek thakur
    @abhi1thakur_twitter
    where pp is a list with all the answers
    SP Mohanty
    @spMohanty
    can you send me a small sample of your predictions file ?
    say top-200 lines ?
    abhishek thakur
    @abhi1thakur_twitter
    sent
    SP Mohanty
    @spMohanty
    dont see the email yet :-?
    could be filtered out by the mailserver because of the size
    abhishek thakur
    @abhi1thakur_twitter
    image.png
    sent you in a PM @spMohanty
    SP Mohanty
    @spMohanty
    the file looks valid
    abhishek thakur
    @abhi1thakur_twitter
    what do you think the problem could be then?
    SP Mohanty
    @spMohanty
    is there any other way you can send me the whole file ?
    abhishek thakur
    @abhi1thakur_twitter
    lemme se
    SP Mohanty
    @spMohanty
    or better
    just try to
    parse your prediction file by
    f = criteo_prediction.CriteoPrediction(filepath, isGzip=False)
    while True:
        l = next(f)
        if not l:
            break
        print l
    just to ensure that the whole file can be parsed
    and theres no errors
    abhishek thakur
    @abhi1thakur_twitter
    where is criteo_prediction?
    SP Mohanty
    @spMohanty
    in the starterkit
    thats the parser for the prediction files
    abhishek thakur
    @abhi1thakur_twitter
    trying
    SP Mohanty
    @spMohanty
    remove the print statement and instead add a try..catch block to just print the link where there might be an error
    abhishek thakur
    @abhi1thakur_twitter
    TypeError                                 Traceback (most recent call last)
    <ipython-input-4-188ab8ce7022> in <module>()
          1 f = criteo_prediction.CriteoPrediction('../predictions.gz')
          2 while True:
    ----> 3     l = next(f)
          4     if not l:
          5         break
    
    ~/Workspace/criteo/crowdai-criteo-ad-placement-challenge-starter-kit/criteo_prediction.py in __next__(self)
         53 
         54     def __next__(self):
    ---> 55         return self.next()
         56 
         57     def close(self):
    
    ~/Workspace/criteo/crowdai-criteo-ad-placement-challenge-starter-kit/criteo_prediction.py in next(self)
         48         try:
         49             line = next(self.fp)
    ---> 50             return self.parse_valid_line(line)
         51         except StopIteration:
         52             raise StopIteration
    
    ~/Workspace/criteo/crowdai-criteo-ad-placement-challenge-starter-kit/criteo_prediction.py in parse_valid_line(self, line)
         24     def parse_valid_line(self, line):
         25         line = line.strip()
    ---> 26         impression_id_marker = line.index(";")
         27         impression_id = line[:impression_id_marker]
         28         assert impression_id != ""
    
    TypeError: a bytes-like object is required, not 'str'
    meh, im on python3
    lemme see
    same error
    SP Mohanty
    @spMohanty
    pass isGzip=False
    abhishek thakur
    @abhi1thakur_twitter
    but it is a gzip
    SP Mohanty
    @spMohanty
    ahh
    anyway try to find the issue, if you cannot, then send me the predictions file, and I will try to have a look in a bit
    abhishek thakur
    @abhi1thakur_twitter
    i found the issue
    SP Mohanty
    @spMohanty
    what was it ?
    abhishek thakur
    @abhi1thakur_twitter
    bunch of predictions missing for IDs in the end
    SP Mohanty
    @spMohanty
    :+1:
    anyway, do create an issue in the issue tracker.
    the error message should be better in this case
    abhishek thakur
    @abhi1thakur_twitter
    cool. in a while
    Alexey Grigorev
    @alexeygrigorev
    so what happens now? :)
    Mikhail Trofimov
    @gef0rce_twitter
    Congrats to Alexey!
    I have a few questions now.
    1) Why 55.0 was chosen as threshold?
    2) I tried a lot different methods (I mean DR, POEM, linear/nonlinear model, different) -- but all of them doesn't reform significantly better that linear model. Did you try to get 58+ on this dataset?
    Mikhail Trofimov
    @gef0rce_twitter
    I really want to understand where is problem -- error in my code, transformation of dataset or just propensity overfitting in papers' results
    SP Mohanty
    @spMohanty
    @gef0rce_twitter : Congrats to you too ;)
    the threshold was decreased to 54
    and I am sending out an email to the top-3 participants soon
    Alexey Grigorev
    @alexeygrigorev
    what happened to the leaderboard?
    wanted to brag about the results but now I'm at the 12th position :)
    SP Mohanty
    @spMohanty
    Hey @alexeygrigorev , dont worry about it.
    Its a temporary bug, @seanfcarroll is looking into is, and should respond back as soon as its fixed :D
    Alexey Grigorev
    @alexeygrigorev
    I see it's back! great! thanks
    btw looks like the switch for post-competition entries doesn't work
    these entries are always shown