I am wondering what your motivation is to join this competition. Looking for a job, a good data science challenge and/or some Kaggle points?
\n\nMy motivation is primarily the unique challenge. I have especially enjoyed the wide solution space with a large number of parameters to tweak, rather than a narrow assembly war.
", "messages": [ { "id": "126117", "postDate": "07/06/2016 13:02:54", "content": "I am wondering what your motivation is to join this competition. Looking for a job, a good data science challenge and/or some Kaggle points?
\n\nMy motivation is primarily the unique challenge. I have especially enjoyed the wide solution space with a large number of parameters to tweak, rather than a narrow assembly war.
", "rawMarkdown": "I am wondering what your motivation is to join this competition. Looking for a job, a good data science challenge and/or some Kaggle points?\r\n\r\nMy motivation is primarily the unique challenge. I have especially enjoyed the wide solution space with a large number of parameters to tweak, rather than a narrow assembly war.", "votes": null }, { "id": "126129", "postDate": "07/06/2016 16:28:08", "content": "For the challenge and kaggle points in my case - I don't have the degree FB wants for one thing, nor the time to come up with a workable Torch/NN solution the description hints that they want ;)
\n\nAside from the recruiting, this is a nice competition - the location based challenge is interesting, and hopefully will have good public->private stability. It got me into looking at KNN and KDE techniques for the first time.
\n\nI wasn't able to come up with an original solution, but I found one very simple improvement that boosts both sets of .58x Python scripts that I'll share tonight :)
", "rawMarkdown": "For the challenge and kaggle points in my case - I don't have the degree FB wants for one thing, nor the time to come up with a workable Torch/NN solution the description hints that they want ;)\r\n\r\nAside from the recruiting, this is a nice competition - the location based challenge is interesting, and hopefully will have good public->private stability. It got me into looking at KNN and KDE techniques for the first time.\r\n\r\nI wasn't able to come up with an original solution, but I found one very simple improvement that boosts both sets of .58x Python scripts that I'll share tonight :)", "votes": null }, { "id": "126132", "postDate": "07/06/2016 16:49:02", "content": "@happycube: I'm looking forward to seeing your improvement!
", "rawMarkdown": "happycube: I'm looking forward to seeing your improvement!", "votes": null }, { "id": "126135", "postDate": "07/06/2016 17:28:57", "content": "Job, if I am so lucky to get it :D . For me nothing could be better than that.
", "rawMarkdown": "Job, if I am so lucky to get it :D . For me nothing could be better than that.", "votes": null }, { "id": "126136", "postDate": "07/06/2016 18:01:20", "content": "For me it's for the kaggle points and challenge. Couldn't get a job at Facebook even if I got 100%!
", "rawMarkdown": "For me it's for the kaggle points and challenge. Couldn't get a job at Facebook even if I got 100%!", "votes": null }, { "id": "126144", "postDate": "07/06/2016 19:23:15", "content": "Definitely not for the job, I'm not in the qualifications ;)) It's just I like programming & ML, and I feel like I began to get addicted to Kaggle competitions.
", "rawMarkdown": "Definitely not for the job, I'm not in the qualifications ;)) It's just I like programming & ML, and I feel like I began to get addicted to Kaggle competitions.", "votes": null }, { "id": "126151", "postDate": "07/06/2016 20:44:53", "content": "While I would be interested in a contact from Facebook about a possible job, I mostly entered to test my data science skills against others, as this is my first Kaggle challenge. My training is in Physics, and I've always approached problems with a fairly academic perspective, where fundamental understanding was valued over all else.
\n\nThis challenge has been an eye-opener in terms of the kinds of solutions that perform well, and how to efficiently go about getting to those solutions. I've also learned a lot about how to develop a data science idea, and the importance of coding up an idea quickly, testing it on a validation set, and moving onto something else rapidly if it doesn't succeed. I think at the beginning of this competition, I was spending far too much time coding up the first iteration of an idea, and spending time making sure that the code was "good code".
", "rawMarkdown": "While I would be interested in a contact from Facebook about a possible job, I mostly entered to test my data science skills against others, as this is my first Kaggle challenge. My training is in Physics, and I've always approached problems with a fairly academic perspective, where fundamental understanding was valued over all else. \r\n\r\nThis challenge has been an eye-opener in terms of the kinds of solutions that perform well, and how to efficiently go about getting to those solutions. I've also learned a lot about how to develop a data science idea, and the importance of coding up an idea quickly, testing it on a validation set, and moving onto something else rapidly if it doesn't succeed. I think at the beginning of this competition, I was spending far too much time coding up the first iteration of an idea, and spending time making sure that the code was \"good code\".", "votes": null }, { "id": "126156", "postDate": "07/06/2016 21:33:24", "content": "Kaggle Points + coolness of the problem, in my case.
\n\nIf am lucky, and find time to do a proper sacrifice to the god of overfitting, this might be my first 10% finish - anyone knows what I have to offer for a big shakeup that puts me 10 positions up? - which to me is enough reward. Then it also happens that this has been the first time that I've been using KNN since my PhD, which has also brought good feelings.
\n\nRegarding the interview, I'd advise you not to take the requirements too seriously - the key point here is going to be years of experience; but I don't think Facebook would let go someone with an excellent solution + a good technical interview, regardless of whether he has or not a PhD on Neuroscience.
\n\nGL everyone!
", "rawMarkdown": "Kaggle Points + coolness of the problem, in my case.\r\n\r\nIf am lucky, and find time to do a proper sacrifice to the god of overfitting, this might be my first 10% finish - anyone knows what I have to offer for a big shakeup that puts me 10 positions up? - which to me is enough reward. Then it also happens that this has been the first time that I've been using KNN since my PhD, which has also brought good feelings.\r\n\r\nRegarding the interview, I'd advise you not to take the requirements too seriously - the key point here is going to be years of experience; but I don't think Facebook would let go someone with an excellent solution + a good technical interview, regardless of whether he has or not a PhD on Neuroscience. \r\n\r\nGL everyone!", "votes": null }, { "id": "126157", "postDate": "07/06/2016 22:02:02", "content": "My main motivation was to learn and have fun. I chose this competition because the data was interesting in having a large number of classes with a small number of features. This was also the only active competition that I could do on my laptop, though it did test my patience at times. After not getting an interview after a top 50 place in the Airbnb competition, I realized that these competitions are a really inefficient way to get interviews. With that said, I would definitely interview at Facebook if given the opportunity.
", "rawMarkdown": "My main motivation was to learn and have fun. I chose this competition because the data was interesting in having a large number of classes with a small number of features. This was also the only active competition that I could do on my laptop, though it did test my patience at times. After not getting an interview after a top 50 place in the Airbnb competition, I realized that these competitions are a really inefficient way to get interviews. With that said, I would definitely interview at Facebook if given the opportunity.", "votes": null }, { "id": "126162", "postDate": "07/06/2016 22:46:42", "content": "@Jared i think you deserve more than an interview from FB. i am very impressed by how you finish this competition on your laptop without relying on aws or super power machines.
\n\nthis competition is challenging due to its large scale. good experience working with aws, but too expensive.
", "rawMarkdown": "Jared i think you deserve more than an interview from FB. i am very impressed by how you finish this competition on your laptop without relying on aws or super power machines.\r\n\r\nthis competition is challenging due to its large scale. good experience working with aws, but too expensive.", "votes": null }, { "id": "126172", "postDate": "07/07/2016 00:07:30", "content": "Now that it's over, I used three tricks to get Abhi's .58649 script up to .59 (public+private LB):
\n\nA while ago I figured out that using log(e) for distance was better than log10. (0.58870 public LB)
Implemented circular time (as seen in the .583x scripts) and removing the duplicate train entries (0.59001)
Used KDE to filter out the outlying checkins for a particular ID 0.59018. I probably could've gotten a bit more lift by integrating it into the probability computation, but didn't have the time/oomph to do it.
Very nice happycube!
\n\nWhen using log in place of log10, did you have to change the weight?
", "rawMarkdown": "Very nice happycube! \r\n\r\nWhen using log in place of log10, did you have to change the weight?", "votes": null }, { "id": "126178", "postDate": "07/07/2016 00:14:58", "content": "@happycube, fantastic, that so simple ideas gave so much gain (especially I am talking about logarithm) !
\n\nIn addition to your ideas, I have figured out, that taking just the mean of several KNNs predictions with different numbers of neighbors (for instance taking models with neighbors = optimal_number_from_script + i for i in range(-12, 13, 3) gave a stable improvement ~0.0005 in both CV and LB scores for all scripts that used KNN. Also I used Nelder-Mead to tune weights in KNN and it also gave a small score improvement in most cases.
I come to this competition because the features are simple but the number of labels are huge. This makes the data set interesting. My goal for this competition is to get top 10 in order to get the MASTER tier.
\n\nThank you everyone and I really enjoy this competition.
", "rawMarkdown": "I come to this competition because the features are simple but the number of labels are huge. This makes the data set interesting. My goal for this competition is to get top 10 in order to get the MASTER tier.\r\n\r\nThank you everyone and I really enjoy this competition.", "votes": null }, { "id": "126182", "postDate": "07/07/2016 00:37:53", "content": "@Jared Turkewitz, I would like to say thank you, for your sharing about "reliable < 1 submission" sometimes ago. It's fundamental for my achievement.
\n\n@happycube, great achievement for 0.59 by KNN. If you don't mind, please elaborate more about KDE, how do you implement the KDE, in R or python, what libraries to be used? Thank you.
\n\nGood luck everyone with the job interview ;)
", "rawMarkdown": "Jared Turkewitz, I would like to say thank you, for your sharing about \"reliable < 1 submission\" sometimes ago. It's fundamental for my achievement. \r\n\r\n@happycube, great achievement for 0.59 by KNN. If you don't mind, please elaborate more about KDE, how do you implement the KDE, in R or python, what libraries to be used? Thank you. \r\n\r\nGood luck everyone with the job interview ;)", "votes": null }, { "id": "126186", "postDate": "07/07/2016 00:55:06", "content": "@happycube, great ideas! Didnt expect the gain from log(e).
", "rawMarkdown": "happycube, great ideas! Didnt expect the gain from log(e).", "votes": null }, { "id": "126189", "postDate": "07/07/2016 01:00:18", "content": "It seems very few want a job with Facebook. Are they that bad?
\n\nI start every competition to try and beat my previous performance. This is my third (I exclude Santander, that was just a quick experiment), and my best result so far. Even though I had a top 50 goal, 56th place is not that far and I'm quite happy with my result.
\n\nAn interview with Facebook demands a top 10 performance, I knew I wasn't going to finish there so getting a job wasn't on my mind. Just like in the BNP challenge the $15000 prize wasn't on my mind, that doesn't mean I don't want $15000.
\n\nNow I'm on my way to Bimbo.
", "rawMarkdown": "It seems very few want a job with Facebook. Are they that bad?\r\n\r\nI start every competition to try and beat my previous performance. This is my third (I exclude Santander, that was just a quick experiment), and my best result so far. Even though I had a top 50 goal, 56th place is not that far and I'm quite happy with my result.\r\n\r\nAn interview with Facebook demands a top 10 performance, I knew I wasn't going to finish there so getting a job wasn't on my mind. Just like in the BNP challenge the $15000 prize wasn't on my mind, that doesn't mean I don't want $15000.\r\n\r\nNow I'm on my way to Bimbo.", "votes": null }, { "id": "126196", "postDate": "07/07/2016 01:26:46", "content": "EDIT: Moved this to the solution sharing thread
", "rawMarkdown": "EDIT: Moved this to the solution sharing thread", "votes": null }, { "id": "126199", "postDate": "07/07/2016 01:31:20", "content": "@Radu Does an interview require top 10? They say in the competition description that there are 10+ positions available! Surely they would be looking to interview more than 10 individuals. Although I suspect they have other hiring avenues as well.
\n\n@Jared I also think you deserve an interview here! Very early on in the comp you introduced the idea of using a grid to use xgboost. I wonder how many people's results lean heavily on your public sharing off that idea! Of course, some people would have come up with it on their own.
", "rawMarkdown": "Radu Does an interview require top 10? They say in the competition description that there are 10+ positions available! Surely they would be looking to interview more than 10 individuals. Although I suspect they have other hiring avenues as well.\r\n\r\n@Jared I also think you deserve an interview here! Very early on in the comp you introduced the idea of using a grid to use xgboost. I wonder how many people's results lean heavily on your public sharing off that idea! Of course, some people would have come up with it on their own.", "votes": null }, { "id": "126435", "postDate": "07/08/2016 12:38:44", "content": "I'll be the first honest person to say that I want an interview, and that's why I took this competition as seriously as I did, and it's also what kept me going after hitting the wall and getting stuck many times. Of course, I also wanted the badge of a top-10 finish and the Kaggle ranking points, but in my opinion a machine learning related position at Facebook is probably one of the best out there for anyone who wants a career doing this.
\n\nI think overall this was a very well setup competition problem due to the many different approaches available, and that resulted in a lot of theorizing, experimentation, and new discoveries.
", "rawMarkdown": "I'll be the first honest person to say that I want an interview, and that's why I took this competition as seriously as I did, and it's also what kept me going after hitting the wall and getting stuck many times. Of course, I also wanted the badge of a top-10 finish and the Kaggle ranking points, but in my opinion a machine learning related position at Facebook is probably one of the best out there for anyone who wants a career doing this. \r\n\r\nI think overall this was a very well setup competition problem due to the many different approaches available, and that resulted in a lot of theorizing, experimentation, and new discoveries.", "votes": null }, { "id": "126613", "postDate": "07/10/2016 19:15:02", "content": "Only 6 out of first 30 submitted their resumes so far, and it would be a big exaggeration to assume that many are looking for a job at FB. My primary motivation was points. I will submit my resume just for a case, but I don't expect a job offering from FB due to visa consideration (I posted this question when the competition started, they did not respond).
", "rawMarkdown": "Only 6 out of first 30 submitted their resumes so far, and it would be a big exaggeration to assume that many are looking for a job at FB. My primary motivation was points. I will submit my resume just for a case, but I don't expect a job offering from FB due to visa consideration (I posted this question when the competition started, they did not respond).", "votes": null }, { "id": "128254", "postDate": "07/18/2016 18:51:07", "content": "Has anyone got a call from FB?
", "rawMarkdown": "Has anyone got a call from FB?", "votes": null }, { "id": "128270", "postDate": "07/18/2016 21:15:55", "content": "@Radu Stoicescu, please be patient, Facebook will start reviewing the candidates soon.
", "rawMarkdown": "Radu Stoicescu, please be patient, Facebook will start reviewing the candidates soon.", "votes": null }, { "id": "129533", "postDate": "07/30/2016 18:50:58", "content": "[quote=Wendy Kan;128270]
\n\n@Radu Stoicescu, please be patient, Facebook will start reviewing the candidates soon.
\n\n[/quote]
\n\nHas Facebook started reviewing candidates yet? Has anyone gotten contacted?
", "rawMarkdown": "[quote=Wendy Kan;128270]\r\n\r\n@Radu Stoicescu, please be patient, Facebook will start reviewing the candidates soon. \r\n\r\n[/quote]\r\n\r\nHas Facebook started reviewing candidates yet? Has anyone gotten contacted?", "votes": null }, { "id": "130776", "postDate": "08/10/2016 08:48:04", "content": "[quote=Jared Turkewitz;129533]
\n\n[quote=Wendy Kan;128270]
\n\n@Radu Stoicescu, please be patient, Facebook will start reviewing the candidates soon.
\n\n[/quote]
\n\nHas Facebook started reviewing candidates yet? Has anyone gotten contacted?
\n\n[/quote]\nWhile I did not apply, I was contacted today. (#28 in on LB)
", "rawMarkdown": "[quote=Jared Turkewitz;129533]\r\n\r\n[quote=Wendy Kan;128270]\r\n\r\n@Radu Stoicescu, please be patient, Facebook will start reviewing the candidates soon. \r\n\r\n[/quote]\r\n\r\nHas Facebook started reviewing candidates yet? Has anyone gotten contacted?\r\n\r\n[/quote]\r\nWhile I did not apply, I was contacted today. (#28 in on LB)", "votes": null }, { "id": "130795", "postDate": "08/10/2016 11:12:14", "content": "[quote=Frederik;130776]
\n\nWhile I did not apply, I was contacted today. (#28 in on LB)
\n\n[/quote]
\n\nGreat to hear that.
\n\nWhile I haven't applied either (I didn't have a great result nor do I have enough experience), after an unpleasant experience with airbnb I was curious to find out how serious companies really are about hiring from a kaggle challenge. I wish you all good luck, and please share your experience with us.
", "rawMarkdown": "[quote=Frederik;130776]\r\n\r\nWhile I did not apply, I was contacted today. (#28 in on LB)\r\n\r\n[/quote]\r\n\r\n\r\nGreat to hear that.\r\n\r\nWhile I haven't applied either (I didn't have a great result nor do I have enough experience), after an unpleasant experience with [airbnb][1] I was curious to find out how serious companies really are about hiring from a kaggle challenge. I wish you all good luck, and please share your experience with us.\r\n\r\n[1]: https://www.kaggle.com/c/airbnb-recruiting-new-user-bookings/forums/t/19268/did-anyone-get-contacted-by-airbnb-at-this-point", "votes": null } ], "comments": [ { "id": 126129, "author_name": "happycube", "author_url": "", "post_date": "07/06/2016 16:28:08", "content": "For the challenge and kaggle points in my case - I don't have the degree FB wants for one thing, nor the time to come up with a workable Torch/NN solution the description hints that they want ;)
\n\nAside from the recruiting, this is a nice competition - the location based challenge is interesting, and hopefully will have good public->private stability. It got me into looking at KNN and KDE techniques for the first time.
\n\nI wasn't able to come up with an original solution, but I found one very simple improvement that boosts both sets of .58x Python scripts that I'll share tonight :)
", "votes": null, "replies": [] }, { "id": 126132, "author_name": "zeroblue", "author_url": "", "post_date": "07/06/2016 16:49:02", "content": "@happycube: I'm looking forward to seeing your improvement!
", "votes": null, "replies": [] }, { "id": 126135, "author_name": "chopra", "author_url": "", "post_date": "07/06/2016 17:28:57", "content": "Job, if I am so lucky to get it :D . For me nothing could be better than that.
", "votes": null, "replies": [] }, { "id": 126136, "author_name": "anokas", "author_url": "", "post_date": "07/06/2016 18:01:20", "content": "For me it's for the kaggle points and challenge. Couldn't get a job at Facebook even if I got 100%!
", "votes": null, "replies": [] }, { "id": 126144, "author_name": "aldente", "author_url": "", "post_date": "07/06/2016 19:23:15", "content": "Definitely not for the job, I'm not in the qualifications ;)) It's just I like programming & ML, and I feel like I began to get addicted to Kaggle competitions.
", "votes": null, "replies": [] }, { "id": 126151, "author_name": "rshekhar2", "author_url": "", "post_date": "07/06/2016 20:44:53", "content": "While I would be interested in a contact from Facebook about a possible job, I mostly entered to test my data science skills against others, as this is my first Kaggle challenge. My training is in Physics, and I've always approached problems with a fairly academic perspective, where fundamental understanding was valued over all else.
\n\nThis challenge has been an eye-opener in terms of the kinds of solutions that perform well, and how to efficiently go about getting to those solutions. I've also learned a lot about how to develop a data science idea, and the importance of coding up an idea quickly, testing it on a validation set, and moving onto something else rapidly if it doesn't succeed. I think at the beginning of this competition, I was spending far too much time coding up the first iteration of an idea, and spending time making sure that the code was "good code".
", "votes": null, "replies": [] }, { "id": 126156, "author_name": "carrdelling", "author_url": "", "post_date": "07/06/2016 21:33:24", "content": "Kaggle Points + coolness of the problem, in my case.
\n\nIf am lucky, and find time to do a proper sacrifice to the god of overfitting, this might be my first 10% finish - anyone knows what I have to offer for a big shakeup that puts me 10 positions up? - which to me is enough reward. Then it also happens that this has been the first time that I've been using KNN since my PhD, which has also brought good feelings.
\n\nRegarding the interview, I'd advise you not to take the requirements too seriously - the key point here is going to be years of experience; but I don't think Facebook would let go someone with an excellent solution + a good technical interview, regardless of whether he has or not a PhD on Neuroscience.
\n\nGL everyone!
", "votes": null, "replies": [] }, { "id": 126157, "author_name": "jturkewitz", "author_url": "", "post_date": "07/06/2016 22:02:02", "content": "My main motivation was to learn and have fun. I chose this competition because the data was interesting in having a large number of classes with a small number of features. This was also the only active competition that I could do on my laptop, though it did test my patience at times. After not getting an interview after a top 50 place in the Airbnb competition, I realized that these competitions are a really inefficient way to get interviews. With that said, I would definitely interview at Facebook if given the opportunity.
", "votes": null, "replies": [] }, { "id": 126162, "author_name": "lishubin", "author_url": "", "post_date": "07/06/2016 22:46:42", "content": "@Jared i think you deserve more than an interview from FB. i am very impressed by how you finish this competition on your laptop without relying on aws or super power machines.
\n\nthis competition is challenging due to its large scale. good experience working with aws, but too expensive.
", "votes": null, "replies": [] }, { "id": 126172, "author_name": "happycube", "author_url": "", "post_date": "07/07/2016 00:07:30", "content": "Now that it's over, I used three tricks to get Abhi's .58649 script up to .59 (public+private LB):
\n\nA while ago I figured out that using log(e) for distance was better than log10. (0.58870 public LB)
Implemented circular time (as seen in the .583x scripts) and removing the duplicate train entries (0.59001)
Used KDE to filter out the outlying checkins for a particular ID 0.59018. I probably could've gotten a bit more lift by integrating it into the probability computation, but didn't have the time/oomph to do it.
Very nice happycube!
\n\nWhen using log in place of log10, did you have to change the weight?
", "votes": null, "replies": [] }, { "id": 126178, "author_name": "arcady27", "author_url": "", "post_date": "07/07/2016 00:14:58", "content": "@happycube, fantastic, that so simple ideas gave so much gain (especially I am talking about logarithm) !
\n\nIn addition to your ideas, I have figured out, that taking just the mean of several KNNs predictions with different numbers of neighbors (for instance taking models with neighbors = optimal_number_from_script + i for i in range(-12, 13, 3) gave a stable improvement ~0.0005 in both CV and LB scores for all scripts that used KNN. Also I used Nelder-Mead to tune weights in KNN and it also gave a small score improvement in most cases.
I come to this competition because the features are simple but the number of labels are huge. This makes the data set interesting. My goal for this competition is to get top 10 in order to get the MASTER tier.
\n\nThank you everyone and I really enjoy this competition.
", "votes": null, "replies": [] }, { "id": 126182, "author_name": "aldente", "author_url": "", "post_date": "07/07/2016 00:37:53", "content": "@Jared Turkewitz, I would like to say thank you, for your sharing about "reliable < 1 submission" sometimes ago. It's fundamental for my achievement.
\n\n@happycube, great achievement for 0.59 by KNN. If you don't mind, please elaborate more about KDE, how do you implement the KDE, in R or python, what libraries to be used? Thank you.
\n\nGood luck everyone with the job interview ;)
", "votes": null, "replies": [] }, { "id": 126186, "author_name": "suryavan", "author_url": "", "post_date": "07/07/2016 00:55:06", "content": "@happycube, great ideas! Didnt expect the gain from log(e).
", "votes": null, "replies": [] }, { "id": 126189, "author_name": "ma350365879", "author_url": "", "post_date": "07/07/2016 01:00:18", "content": "It seems very few want a job with Facebook. Are they that bad?
\n\nI start every competition to try and beat my previous performance. This is my third (I exclude Santander, that was just a quick experiment), and my best result so far. Even though I had a top 50 goal, 56th place is not that far and I'm quite happy with my result.
\n\nAn interview with Facebook demands a top 10 performance, I knew I wasn't going to finish there so getting a job wasn't on my mind. Just like in the BNP challenge the $15000 prize wasn't on my mind, that doesn't mean I don't want $15000.
\n\nNow I'm on my way to Bimbo.
", "votes": null, "replies": [] }, { "id": 126196, "author_name": "rshekhar2", "author_url": "", "post_date": "07/07/2016 01:26:46", "content": "EDIT: Moved this to the solution sharing thread
", "votes": null, "replies": [] }, { "id": 126199, "author_name": "cjcoursera", "author_url": "", "post_date": "07/07/2016 01:31:20", "content": "@Radu Does an interview require top 10? They say in the competition description that there are 10+ positions available! Surely they would be looking to interview more than 10 individuals. Although I suspect they have other hiring avenues as well.
\n\n@Jared I also think you deserve an interview here! Very early on in the comp you introduced the idea of using a grid to use xgboost. I wonder how many people's results lean heavily on your public sharing off that idea! Of course, some people would have come up with it on their own.
", "votes": null, "replies": [] }, { "id": 126435, "author_name": "qwang88", "author_url": "", "post_date": "07/08/2016 12:38:44", "content": "I'll be the first honest person to say that I want an interview, and that's why I took this competition as seriously as I did, and it's also what kept me going after hitting the wall and getting stuck many times. Of course, I also wanted the badge of a top-10 finish and the Kaggle ranking points, but in my opinion a machine learning related position at Facebook is probably one of the best out there for anyone who wants a career doing this.
\n\nI think overall this was a very well setup competition problem due to the many different approaches available, and that resulted in a lot of theorizing, experimentation, and new discoveries.
", "votes": null, "replies": [] }, { "id": 126613, "author_name": "rakhlin", "author_url": "", "post_date": "07/10/2016 19:15:02", "content": "Only 6 out of first 30 submitted their resumes so far, and it would be a big exaggeration to assume that many are looking for a job at FB. My primary motivation was points. I will submit my resume just for a case, but I don't expect a job offering from FB due to visa consideration (I posted this question when the competition started, they did not respond).
", "votes": null, "replies": [] }, { "id": 128254, "author_name": "radustoicescu", "author_url": "", "post_date": "07/18/2016 18:51:07", "content": "Has anyone got a call from FB?
", "votes": null, "replies": [] }, { "id": 128270, "author_name": "wendykan", "author_url": "", "post_date": "07/18/2016 21:15:55", "content": "@Radu Stoicescu, please be patient, Facebook will start reviewing the candidates soon.
", "votes": null, "replies": [] }, { "id": 129533, "author_name": "jturkewitz", "author_url": "", "post_date": "07/30/2016 18:50:58", "content": "[quote=Wendy Kan;128270]
\n\n@Radu Stoicescu, please be patient, Facebook will start reviewing the candidates soon.
\n\n[/quote]
\n\nHas Facebook started reviewing candidates yet? Has anyone gotten contacted?
", "votes": null, "replies": [] }, { "id": 130776, "author_name": "frederik", "author_url": "", "post_date": "08/10/2016 08:48:04", "content": "[quote=Jared Turkewitz;129533]
\n\n[quote=Wendy Kan;128270]
\n\n@Radu Stoicescu, please be patient, Facebook will start reviewing the candidates soon.
\n\n[/quote]
\n\nHas Facebook started reviewing candidates yet? Has anyone gotten contacted?
\n\n[/quote]\nWhile I did not apply, I was contacted today. (#28 in on LB)
", "votes": null, "replies": [] }, { "id": 130795, "author_name": "radustoicescu", "author_url": "", "post_date": "08/10/2016 11:12:14", "content": "[quote=Frederik;130776]
\n\nWhile I did not apply, I was contacted today. (#28 in on LB)
\n\n[/quote]
\n\nGreat to hear that.
\n\nWhile I haven't applied either (I didn't have a great result nor do I have enough experience), after an unpleasant experience with airbnb I was curious to find out how serious companies really are about hiring from a kaggle challenge. I wish you all good luck, and please share your experience with us.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "126117": "I am wondering what your motivation is to join this competition. Looking for a job, a good data science challenge and/or some Kaggle points?\r\n\r\nMy motivation is primarily the unique challenge. I have especially enjoyed the wide solution space with a large number of parameters to tweak, rather than a narrow assembly war.", "126129": "For the challenge and kaggle points in my case - I don't have the degree FB wants for one thing, nor the time to come up with a workable Torch/NN solution the description hints that they want ;)\r\n\r\nAside from the recruiting, this is a nice competition - the location based challenge is interesting, and hopefully will have good public->private stability. It got me into looking at KNN and KDE techniques for the first time.\r\n\r\nI wasn't able to come up with an original solution, but I found one very simple improvement that boosts both sets of .58x Python scripts that I'll share tonight :)", "126132": "happycube: I'm looking forward to seeing your improvement!", "126135": "Job, if I am so lucky to get it :D . For me nothing could be better than that.", "126136": "For me it's for the kaggle points and challenge. Couldn't get a job at Facebook even if I got 100%!", "126144": "Definitely not for the job, I'm not in the qualifications ;)) It's just I like programming & ML, and I feel like I began to get addicted to Kaggle competitions.", "126151": "While I would be interested in a contact from Facebook about a possible job, I mostly entered to test my data science skills against others, as this is my first Kaggle challenge. My training is in Physics, and I've always approached problems with a fairly academic perspective, where fundamental understanding was valued over all else. \r\n\r\nThis challenge has been an eye-opener in terms of the kinds of solutions that perform well, and how to efficiently go about getting to those solutions. I've also learned a lot about how to develop a data science idea, and the importance of coding up an idea quickly, testing it on a validation set, and moving onto something else rapidly if it doesn't succeed. I think at the beginning of this competition, I was spending far too much time coding up the first iteration of an idea, and spending time making sure that the code was \"good code\".", "126156": "Kaggle Points + coolness of the problem, in my case.\r\n\r\nIf am lucky, and find time to do a proper sacrifice to the god of overfitting, this might be my first 10% finish - anyone knows what I have to offer for a big shakeup that puts me 10 positions up? - which to me is enough reward. Then it also happens that this has been the first time that I've been using KNN since my PhD, which has also brought good feelings.\r\n\r\nRegarding the interview, I'd advise you not to take the requirements too seriously - the key point here is going to be years of experience; but I don't think Facebook would let go someone with an excellent solution + a good technical interview, regardless of whether he has or not a PhD on Neuroscience. \r\n\r\nGL everyone!", "126157": "My main motivation was to learn and have fun. I chose this competition because the data was interesting in having a large number of classes with a small number of features. This was also the only active competition that I could do on my laptop, though it did test my patience at times. After not getting an interview after a top 50 place in the Airbnb competition, I realized that these competitions are a really inefficient way to get interviews. With that said, I would definitely interview at Facebook if given the opportunity.", "126162": "Jared i think you deserve more than an interview from FB. i am very impressed by how you finish this competition on your laptop without relying on aws or super power machines.\r\n\r\nthis competition is challenging due to its large scale. good experience working with aws, but too expensive.", "126172": "Now that it's over, I used three tricks to get Abhi's .58649 script up to .59 (public+private LB):\r\n\r\n- A while ago I figured out that using log(e) for distance was better than log10. (0.58870 public LB)\r\n\r\n- Implemented circular time (as seen in the .583x scripts) and removing the duplicate train entries (0.59001)\r\n\r\n- Used KDE to filter out the outlying checkins for a particular ID 0.59018. I probably could've gotten a bit more lift by integrating it into the probability computation, but didn't have the time/oomph to do it.", "126176": "Very nice happycube! \r\n\r\nWhen using log in place of log10, did you have to change the weight?", "126178": "happycube, fantastic, that so simple ideas gave so much gain (especially I am talking about logarithm) !\r\n\r\nIn addition to your ideas, I have figured out, that taking just the mean of several KNNs predictions with different numbers of neighbors (for instance taking models with `neighbors = optimal_number_from_script + i for i in range(-12, 13, 3) ` gave a stable improvement ~0.0005 in both CV and LB scores for all scripts that used KNN. Also I used Nelder-Mead to tune weights in KNN and it also gave a small score improvement in most cases.", "126181": "I come to this competition because the features are simple but the number of labels are huge. This makes the data set interesting. My goal for this competition is to get top 10 in order to get the MASTER tier.\r\n\r\nThank you everyone and I really enjoy this competition.", "126182": "Jared Turkewitz, I would like to say thank you, for your sharing about \"reliable < 1 submission\" sometimes ago. It's fundamental for my achievement. \r\n\r\n@happycube, great achievement for 0.59 by KNN. If you don't mind, please elaborate more about KDE, how do you implement the KDE, in R or python, what libraries to be used? Thank you. \r\n\r\nGood luck everyone with the job interview ;)", "126186": "happycube, great ideas! Didnt expect the gain from log(e).", "126189": "Job of course, but almost give up after seeing so many awesome solutions, especially when having no idea why so many higher ranks over there.\r\n - Deeper insights into how KNN works and glade to give some contribution on improving the algorithm. I guess this is one of the rare place where we can get such a huge data to work with KNN.\r\n - Learn how ensembling works and how to improve it.\r\n - Deal with big, simple and multi-classes dataset.", "126193": "It seems very few want a job with Facebook. Are they that bad?\r\n\r\nI start every competition to try and beat my previous performance. This is my third (I exclude Santander, that was just a quick experiment), and my best result so far. Even though I had a top 50 goal, 56th place is not that far and I'm quite happy with my result.\r\n\r\nAn interview with Facebook demands a top 10 performance, I knew I wasn't going to finish there so getting a job wasn't on my mind. Just like in the BNP challenge the $15000 prize wasn't on my mind, that doesn't mean I don't want $15000.\r\n\r\nNow I'm on my way to Bimbo.", "126196": "EDIT: Moved this to the solution sharing thread", "126199": "Radu Does an interview require top 10? They say in the competition description that there are 10+ positions available! Surely they would be looking to interview more than 10 individuals. Although I suspect they have other hiring avenues as well.\r\n\r\n@Jared I also think you deserve an interview here! Very early on in the comp you introduced the idea of using a grid to use xgboost. I wonder how many people's results lean heavily on your public sharing off that idea! Of course, some people would have come up with it on their own.", "126435": "I'll be the first honest person to say that I want an interview, and that's why I took this competition as seriously as I did, and it's also what kept me going after hitting the wall and getting stuck many times. Of course, I also wanted the badge of a top-10 finish and the Kaggle ranking points, but in my opinion a machine learning related position at Facebook is probably one of the best out there for anyone who wants a career doing this. \r\n\r\nI think overall this was a very well setup competition problem due to the many different approaches available, and that resulted in a lot of theorizing, experimentation, and new discoveries.", "126613": "Only 6 out of first 30 submitted their resumes so far, and it would be a big exaggeration to assume that many are looking for a job at FB. My primary motivation was points. I will submit my resume just for a case, but I don't expect a job offering from FB due to visa consideration (I posted this question when the competition started, they did not respond).", "128254": "Has anyone got a call from FB?", "128270": "Radu Stoicescu, please be patient, Facebook will start reviewing the candidates soon.", "129533": "[quote=Wendy Kan;128270]\r\n\r\n@Radu Stoicescu, please be patient, Facebook will start reviewing the candidates soon. \r\n\r\n[/quote]\r\n\r\nHas Facebook started reviewing candidates yet? Has anyone gotten contacted?", "130776": "[quote=Jared Turkewitz;129533]\r\n\r\n[quote=Wendy Kan;128270]\r\n\r\n@Radu Stoicescu, please be patient, Facebook will start reviewing the candidates soon. \r\n\r\n[/quote]\r\n\r\nHas Facebook started reviewing candidates yet? Has anyone gotten contacted?\r\n\r\n[/quote]\r\nWhile I did not apply, I was contacted today. (#28 in on LB)", "130795": "[quote=Frederik;130776]\r\n\r\nWhile I did not apply, I was contacted today. (#28 in on LB)\r\n\r\n[/quote]\r\n\r\n\r\nGreat to hear that.\r\n\r\nWhile I haven't applied either (I didn't have a great result nor do I have enough experience), after an unpleasant experience with [airbnb][1] I was curious to find out how serious companies really are about hiring from a kaggle challenge. I wish you all good luck, and please share your experience with us.\r\n\r\n[1]: https://www.kaggle.com/c/airbnb-recruiting-new-user-bookings/forums/t/19268/did-anyone-get-contacted-by-airbnb-at-this-point" }, "source": "meta" }