I am using a 4 cores i7 with 64 GB and a GTX 1080 Ti. Memory is not really an issue here, my processes are way smaller than the available RAM.
", "messages": [ { "id": 339701, "postDate": "2018-06-07T12:24:53.293Z", "content": "The kaggle kernel is ok~~
", "rawMarkdown": "The kaggle kernel is ok~~", "votes": 3, "replies": [ { "id": 340434, "postDate": "2018-06-09T06:30:21.370Z", "content": "Well, it takes me about 30 hours to compute a solution, it would not run on a Kaggle Kernel due to the time limitL
", "rawMarkdown": "Well, it takes me about 30 hours to compute a solution, it would not run on a Kaggle Kernel due to the time limitL", "votes": 1 }, { "id": 340444, "postDate": "2018-06-09T07:20:42.930Z", "content": "It takes me 7 hours to obtain the score 0.62 using my 2x4 cores Xeon and probably every additional 0.02 will take me twice longer time, not talking about code optimisation and new ideas needed. Best wishes.
", "rawMarkdown": "It takes me 7 hours to obtain the score 0.62 using my 2x4 cores Xeon and probably every additional 0.02 will take me twice longer time, not talking about code optimisation and new ideas needed. Best wishes.", "votes": 1 }, { "id": 349567, "postDate": "2018-06-28T09:28:46.830Z", "content": "Really?Kaggle kernel can handle this?I consider to join this competition,but I only have 8GB ram PC.
", "rawMarkdown": "Really?Kaggle kernel can handle this?I consider to join this competition,but I only have 8GB ram PC." }, { "id": 349936, "postDate": "2018-06-28T22:11:54.943Z", "content": "The kernels can run decent submissions, but it seems many of the better scoring solutions are having very long processing times, so parallel processing will probably be more of a limitation than memory.
", "rawMarkdown": "The kernels can run decent submissions, but it seems many of the better scoring solutions are having very long processing times, so parallel processing will probably be more of a limitation than memory. ", "votes": 2 }, { "id": 350008, "postDate": "2018-06-29T02:52:08.353Z", "content": "You can start this competion, refer to https://www.kaggle.com/sionek/bayesian-optimization @Grzegorz Sionkowski
", "rawMarkdown": "You can start this competion, refer to https://www.kaggle.com/sionek/bayesian-optimization @Grzegorz Sionkowski", "votes": 1 } ] }, { "id": 339696, "postDate": "2018-06-07T12:08:15.843Z", "content": "I am using a 4 cores i7 with 64 GB and a GTX 1080 Ti. Memory is not really an issue here, my processes are way smaller than the available RAM.
", "rawMarkdown": "I am using a 4 cores i7 with 64 GB and a GTX 1080 Ti. Memory is not really an issue here, my processes are way smaller than the available RAM.", "votes": 3 }, { "id": 341557, "postDate": "2018-06-11T19:40:23.227Z", "content": "Newbie here. Setup an I9-7900 with 64GB memory and two 1080 ti cards, with two stripped 500GB NVNe pcie drives. I got the i9 for the 44 lanes to push data to the 1080 ti cards. The pcie drives give me ~6 GB of throughput. I may be a little light on memory.
\n\nI have no dreams of doing good in this being my first kaggle, just want to learn more about data science.
", "rawMarkdown": "Newbie here. Setup an I9-7900 with 64GB memory and two 1080 ti cards, with two stripped 500GB NVNe pcie drives. I got the i9 for the 44 lanes to push data to the 1080 ti cards. The pcie drives give me ~6 GB of throughput. I may be a little light on memory. \n\nI have no dreams of doing good in this being my first kaggle, just want to learn more about data science.", "votes": 1, "replies": [ { "id": 341568, "postDate": "2018-06-11T20:08:10.813Z", "content": "That's a monster! Welcome from a fellow beginner to the competition, there is plenty to learn.
\n\nWhere did this machine come from? Is it a personal project or a workstation used for something else?
", "rawMarkdown": "That's a monster! Welcome from a fellow beginner to the competition, there is plenty to learn. \n\nWhere did this machine come from? Is it a personal project or a workstation used for something else?", "votes": 1 }, { "id": 341592, "postDate": "2018-06-11T20:51:30.283Z", "content": "I built it. I have a full up DC/OS cluster in my house. Added this little puppy to test out implementations of MxNet, PyTorch, TensorFlow and other goddies my customer wants to add to their cluster. Also, this site is a great place to learn and add another skill to my toolbox, so I figured a good GPU box would help me learn.
", "rawMarkdown": "I built it. I have a full up DC/OS cluster in my house. Added this little puppy to test out implementations of MxNet, PyTorch, TensorFlow and other goddies my customer wants to add to their cluster. Also, this site is a great place to learn and add another skill to my toolbox, so I figured a good GPU box would help me learn." }, { "id": 341713, "postDate": "2018-06-12T05:43:29.467Z", "content": "> I have no dreams of doing good in this being my first kaggle
\n\nThis competition is very open as it is quite different from usual machine learning competitions. You have all your chances even if you are not a seasoned ML competitor. For instance, it is not yet clear how to best apply supervised machine learning. And you can do quite well without machine learning (above 0.6).
\n\nEdited (thanks to Nicole Finnie): replaced 'machine learning' by 'supervised machine learning' as clustering is a form of machine learning.
", "rawMarkdown": "> I have no dreams of doing good in this being my first kaggle\n\nThis competition is very open as it is quite different from usual machine learning competitions. You have all your chances even if you are not a seasoned ML competitor. For instance, it is not yet clear how to best apply supervised machine learning. And you can do quite well without machine learning (above 0.6).\n\nEdited (thanks to Nicole Finnie): replaced 'machine learning' by 'supervised machine learning' as clustering is a form of machine learning.", "votes": 1 } ] }, { "id": 339720, "postDate": "2018-06-07T14:01:37.550Z", "content": "A beginner's question regarding GPU: apart from using it for neural network or gradient boosting frameworks, do you often try to vectorize regular functions which are time-consuming and run them on GPU? I suppose the answer depends on the details?
\n\nAnswer to your question: for now just a notebook with i5, 12GB and GeForce 840m.
", "rawMarkdown": "A beginner's question regarding GPU: apart from using it for neural network or gradient boosting frameworks, do you often try to vectorize regular functions which are time-consuming and run them on GPU? I suppose the answer depends on the details?\n\nAnswer to your question: for now just a notebook with i5, 12GB and GeForce 840m.", "votes": 1, "replies": [ { "id": 339732, "postDate": "2018-06-07T14:25:38.643Z", "content": "I was thinking about it as many operations we do here are indeed vectorized ones. I wish I could just use a gpu accelerator for numpy.
\n\ngoogling the topic leads to some interesting hits. I'll look into them:
\n\nhttps://github.com/lebedov/scikit-cuda
\n\nhttps://docs.anaconda.com/accelerate/
\n\nhttp://minpy.readthedocs.io/en/latest/tutorial/numpy_under_minpy.html
", "rawMarkdown": "I was thinking about it as many operations we do here are indeed vectorized ones. I wish I could just use a gpu accelerator for numpy.\n\ngoogling the topic leads to some interesting hits. I'll look into them:\n\nhttps://github.com/lebedov/scikit-cuda\n\nhttps://docs.anaconda.com/accelerate/\n\nhttp://minpy.readthedocs.io/en/latest/tutorial/numpy_under_minpy.html\n\n\n", "votes": 3 }, { "id": 340342, "postDate": "2018-06-09T00:50:02.867Z", "content": "http://gpu.userbenchmark.com/Compare/Nvidia-GTX-1080-Ti-vs-Nvidia-GeForce-840M/3918vsm8643
\n\nGTX 1080Ti VS GeForce 840M
\n\nGTX 1080Ti is much faster, much better, much more efficient. But at the time I got my laptop (2015), I didn't look so precise, there was much better option than 840M, but it was rather good as well. I see that the age of 840M is already 4 years (so when I ordered it, GeForce 840M was rather young, but GTX 1080Ti is only ~1 year old)
\n\nInsanely faster multi rendering. +2,033%\n Insanely faster NBody calculation. +1,271%
\n\nAmazing!
", "rawMarkdown": "http://gpu.userbenchmark.com/Compare/Nvidia-GTX-1080-Ti-vs-Nvidia-GeForce-840M/3918vsm8643\n\nGTX 1080Ti VS GeForce 840M\n\nGTX 1080Ti is much faster, much better, much more efficient. But at the time I got my laptop (2015), I didn't look so precise, there was much better option than 840M, but it was rather good as well. I see that the age of 840M is already 4 years (so when I ordered it, GeForce 840M was rather young, but GTX 1080Ti is only ~1 year old)\n\nInsanely faster multi rendering. +2,033%\n Insanely faster NBody calculation. +1,271% \n\n\nAmazing!" }, { "id": 340343, "postDate": "2018-06-09T00:50:28.873Z", "content": "How comfortable you are with GTX 1080Ti?
", "rawMarkdown": "How comfortable you are with GTX 1080Ti? " }, { "id": 340344, "postDate": "2018-06-09T00:52:41.647Z", "content": "How likely you will recommend to have some this kind of nice fast graphics card?\nBecause I did not use anything for comparison, but you might be done and can share experience.\nMerci beaucoup!
", "rawMarkdown": "How likely you will recommend to have some this kind of nice fast graphics card?\nBecause I did not use anything for comparison, but you might be done and can share experience.\nMerci beaucoup!" }, { "id": 340433, "postDate": "2018-06-09T06:29:19.910Z", "content": "1080 TI (Pascal architecture) or more recent Volta architecture (Titan XP) are definitely making a difference when training complex CNN like resnet models. If you enter image recognition competition then having some of these really makes a difference. If you look at Carvana segmentation, the top teams all had several of these GPUs.
\n\nHere, I am doing some deep learning models, but nothing serious. My current score doe snot use any of them, hence the 1080 Ti is not used so far.
", "rawMarkdown": "1080 TI (Pascal architecture) or more recent Volta architecture (Titan XP) are definitely making a difference when training complex CNN like resnet models. If you enter image recognition competition then having some of these really makes a difference. If you look at Carvana segmentation, the top teams all had several of these GPUs.\n\nHere, I am doing some deep learning models, but nothing serious. My current score doe snot use any of them, hence the 1080 Ti is not used so far.", "votes": 1 }, { "id": 340546, "postDate": "2018-06-09T14:54:50.947Z", "content": "And the good news is that you can buy the high end cards for something close to list price again. It seems like the mining craziness is slowing down.
", "rawMarkdown": "And the good news is that you can buy the high end cards for something close to list price again. It seems like the mining craziness is slowing down." }, { "id": 340621, "postDate": "2018-06-09T19:25:24.260Z", "content": "Thanks for the info! Very useful.
", "rawMarkdown": "Thanks for the info! Very useful.", "votes": 1 }, { "id": 340863, "postDate": "2018-06-10T13:56:55.550Z", "content": "You can actually use pytorch to do vectorized operations on the gpu, similar to working with numpy.
", "rawMarkdown": "You can actually use pytorch to do vectorized operations on the gpu, similar to working with numpy.", "votes": 2 }, { "id": 340867, "postDate": "2018-06-10T14:02:26.390Z", "content": "\n\n\nYou can actually use pytorch to do vectorized operations on the gpu, similar to working with numpy.
\n
Sure, but I don't have DBSCAN on top of that. If it was a gpu accelerated numpy then code using numpy would run unmodified. Anyway, that's what I am dreaming of.
", "rawMarkdown": "> You can actually use pytorch to do vectorized operations on the gpu, similar to working with numpy.\n\nSure, but I don't have DBSCAN on top of that. If it was a gpu accelerated numpy then code using numpy would run unmodified. Anyway, that's what I am dreaming of.", "votes": 1 }, { "id": 340954, "postDate": "2018-06-10T17:37:39.923Z", "content": "Oh, I thought you were trying to speed up your own code.
", "rawMarkdown": "Oh, I thought you were trying to speed up your own code." } ] }, { "id": 352659, "postDate": "2018-07-04T20:14:13.553Z", "content": "AMD 1 core 2400 MHz 2 Gb RAM - 1 hour to make solution.
", "rawMarkdown": "AMD 1 core 2400 MHz 2 Gb RAM - 1 hour to make solution.", "votes": 2, "replies": [ { "id": 352663, "postDate": "2018-07-04T20:44:23.043Z", "content": "1 hour per event?
", "rawMarkdown": "1 hour per event?" }, { "id": 352670, "postDate": "2018-07-04T21:10:18.983Z", "content": "\n\n\nmacfarll wrote
\n \n\n\n1 hour per event?
\n
...per all 125
", "rawMarkdown": "\n> **macfarll wrote**\n> \n> > 1 hour per event?\n\n...per all 125", "votes": 6 }, { "id": 352675, "postDate": "2018-07-04T21:35:00.883Z", "content": "That's insane! Very impressive, your solution must work very differently from the dbscan unrolling approaches
", "rawMarkdown": "That's insane! Very impressive, your solution must work very differently from the dbscan unrolling approaches " }, { "id": 353074, "postDate": "2018-07-05T20:15:03.223Z", "content": "@Victor, you should join the 2nd phase competition. Your result is impressive!
", "rawMarkdown": "@Victor, you should join the 2nd phase competition. Your result is impressive! " } ] }, { "id": 342716, "postDate": "2018-06-14T00:09:33.637Z", "content": "Currently using my $200 in Azure credit on a 4vCPU/8GB virtual machine. I cloned a backup VM so I can still tweak my model while creating a submission.
", "rawMarkdown": "Currently using my $200 in Azure credit on a 4vCPU/8GB virtual machine. I cloned a backup VM so I can still tweak my model while creating a submission." }, { "id": 341736, "postDate": "2018-06-12T07:08:06.243Z", "content": "For education just regular macbook pro 2015 with core i7 + 16G ram + SSD.
", "rawMarkdown": "For education just regular macbook pro 2015 with core i7 + 16G ram + SSD." }, { "id": 340225, "postDate": "2018-06-08T16:53:15.280Z", "content": "I am using cloud VM to parallel. i am just using my old 2012 macbook air to run locally.
", "rawMarkdown": "I am using cloud VM to parallel. i am just using my old 2012 macbook air to run locally.", "replies": [ { "id": 342743, "postDate": "2018-06-14T02:33:26.917Z", "content": "Not sure if everyone has the similar pattern. Our code runs about 7 minutes per dataset, without GPU on regular mac laptop. So just to make one submission, we used to run 14 hours. So I wrote a script to use Google Cloud preemptible VMs(much cheaper cost), spin up 125 VMs, then use simple script to combine all data into one file. Boom! 8 minutes, all done. Cost os about $1.5 USD
", "rawMarkdown": "Not sure if everyone has the similar pattern. Our code runs about 7 minutes per dataset, without GPU on regular mac laptop. So just to make one submission, we used to run 14 hours. So I wrote a script to use Google Cloud preemptible VMs(much cheaper cost), spin up 125 VMs, then use simple script to combine all data into one file. Boom! 8 minutes, all done. Cost os about $1.5 USD", "votes": 2 }, { "id": 342744, "postDate": "2018-06-14T02:34:32.157Z", "content": "https://github.com/liuxiao/CFlow, in case someone is interested
", "rawMarkdown": "https://github.com/liuxiao/CFlow, in case someone is interested", "votes": 1 } ] }, { "id": 339786, "postDate": "2018-06-07T16:39:43.880Z", "content": "I7 4770k \nGTX 1070\n16 GB ram (I just upped from 8)\n120gb ssd (I need more...)
\n\nI reused old ram and went with a cheap ssd, which are choices that made since at the time but are starting to feel more and more like a limitation.
", "rawMarkdown": "I7 4770k \nGTX 1070\n16 GB ram (I just upped from 8)\n120gb ssd (I need more...)\n\nI reused old ram and went with a cheap ssd, which are choices that made since at the time but are starting to feel more and more like a limitation. ", "replies": [ { "id": 340346, "postDate": "2018-06-09T00:53:20.200Z", "content": "How much you paid for a laptop with such options?
", "rawMarkdown": "How much you paid for a laptop with such options?" }, { "id": 343096, "postDate": "2018-06-14T17:24:17.203Z", "content": "I just realized this was directed to me...
\n\nMy computer is a desktop, hard to put a price on it since I built/upgraded incrementally, but current parts are a little over 1000. Luckily I got my gpu before the crypto currency fad.
\n\nAlso worth noting that my current solution doesn't use my gpu at all.
", "rawMarkdown": "I just realized this was directed to me...\n\nMy computer is a desktop, hard to put a price on it since I built/upgraded incrementally, but current parts are a little over 1000. Luckily I got my gpu before the crypto currency fad.\n\nAlso worth noting that my current solution doesn't use my gpu at all." } ] }, { "id": 339765, "postDate": "2018-06-07T15:33:27.177Z", "content": "I also have a 4 cores i7, with (just) 6GB. \nI started this contest with an old i5, then I decided it was a good time to invest in a new hardware :)
", "rawMarkdown": "I also have a 4 cores i7, with (just) 6GB. \nI started this contest with an old i5, then I decided it was a good time to invest in a new hardware :)\n" }, { "id": 340338, "postDate": "2018-06-09T00:43:48.550Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 340341, "postDate": "2018-06-09T00:49:15.803Z", "rawMarkdown": "", "isDeleted": true } ] } ], "comments": [ { "id": 339701, "author_name": "yyll008", "author_url": "", "post_date": "2018-06-07T12:24:53.293000", "content": "The kaggle kernel is ok~~
", "votes": 3, "replies": [ { "id": 340434, "author_name": "CPMP", "author_url": "", "post_date": "2018-06-09T06:30:21.370000", "content": "Well, it takes me about 30 hours to compute a solution, it would not run on a Kaggle Kernel due to the time limitL
", "votes": 1, "replies": [] }, { "id": 340444, "author_name": "Grzegorz Sionkowski", "author_url": "", "post_date": "2018-06-09T07:20:42.930000", "content": "It takes me 7 hours to obtain the score 0.62 using my 2x4 cores Xeon and probably every additional 0.02 will take me twice longer time, not talking about code optimisation and new ideas needed. Best wishes.
", "votes": 1, "replies": [] }, { "id": 349567, "author_name": "Johnny Liu", "author_url": "", "post_date": "2018-06-28T09:28:46.830000", "content": "Really?Kaggle kernel can handle this?I consider to join this competition,but I only have 8GB ram PC.
", "votes": 0, "replies": [] }, { "id": 349936, "author_name": "macfarll", "author_url": "", "post_date": "2018-06-28T22:11:54.943000", "content": "The kernels can run decent submissions, but it seems many of the better scoring solutions are having very long processing times, so parallel processing will probably be more of a limitation than memory.
", "votes": 2, "replies": [] }, { "id": 350008, "author_name": "yyll008", "author_url": "", "post_date": "2018-06-29T02:52:08.353000", "content": "You can start this competion, refer to https://www.kaggle.com/sionek/bayesian-optimization @Grzegorz Sionkowski
", "votes": 1, "replies": [] } ] }, { "id": 341557, "author_name": "Dan Brown", "author_url": "", "post_date": "2018-06-11T19:40:23.227000", "content": "Newbie here. Setup an I9-7900 with 64GB memory and two 1080 ti cards, with two stripped 500GB NVNe pcie drives. I got the i9 for the 44 lanes to push data to the 1080 ti cards. The pcie drives give me ~6 GB of throughput. I may be a little light on memory.
\n\nI have no dreams of doing good in this being my first kaggle, just want to learn more about data science.
", "votes": 1, "replies": [ { "id": 341568, "author_name": "macfarll", "author_url": "", "post_date": "2018-06-11T20:08:10.813000", "content": "That's a monster! Welcome from a fellow beginner to the competition, there is plenty to learn.
\n\nWhere did this machine come from? Is it a personal project or a workstation used for something else?
", "votes": 1, "replies": [] }, { "id": 341592, "author_name": "Dan Brown", "author_url": "", "post_date": "2018-06-11T20:51:30.283000", "content": "I built it. I have a full up DC/OS cluster in my house. Added this little puppy to test out implementations of MxNet, PyTorch, TensorFlow and other goddies my customer wants to add to their cluster. Also, this site is a great place to learn and add another skill to my toolbox, so I figured a good GPU box would help me learn.
", "votes": 0, "replies": [] }, { "id": 341713, "author_name": "CPMP", "author_url": "", "post_date": "2018-06-12T05:43:29.467000", "content": "> I have no dreams of doing good in this being my first kaggle
\n\nThis competition is very open as it is quite different from usual machine learning competitions. You have all your chances even if you are not a seasoned ML competitor. For instance, it is not yet clear how to best apply supervised machine learning. And you can do quite well without machine learning (above 0.6).
\n\nEdited (thanks to Nicole Finnie): replaced 'machine learning' by 'supervised machine learning' as clustering is a form of machine learning.
", "votes": 1, "replies": [] } ] }, { "id": 339720, "author_name": "Jakub Guzowski", "author_url": "", "post_date": "2018-06-07T14:01:37.550000", "content": "A beginner's question regarding GPU: apart from using it for neural network or gradient boosting frameworks, do you often try to vectorize regular functions which are time-consuming and run them on GPU? I suppose the answer depends on the details?
\n\nAnswer to your question: for now just a notebook with i5, 12GB and GeForce 840m.
", "votes": 1, "replies": [ { "id": 339732, "author_name": "CPMP", "author_url": "", "post_date": "2018-06-07T14:25:38.643000", "content": "I was thinking about it as many operations we do here are indeed vectorized ones. I wish I could just use a gpu accelerator for numpy.
\n\ngoogling the topic leads to some interesting hits. I'll look into them:
\n\nhttps://github.com/lebedov/scikit-cuda
\n\nhttps://docs.anaconda.com/accelerate/
\n\nhttp://minpy.readthedocs.io/en/latest/tutorial/numpy_under_minpy.html
", "votes": 3, "replies": [] }, { "id": 340342, "author_name": "Mukharbek Organokov", "author_url": "", "post_date": "2018-06-09T00:50:02.867000", "content": "http://gpu.userbenchmark.com/Compare/Nvidia-GTX-1080-Ti-vs-Nvidia-GeForce-840M/3918vsm8643
\n\nGTX 1080Ti VS GeForce 840M
\n\nGTX 1080Ti is much faster, much better, much more efficient. But at the time I got my laptop (2015), I didn't look so precise, there was much better option than 840M, but it was rather good as well. I see that the age of 840M is already 4 years (so when I ordered it, GeForce 840M was rather young, but GTX 1080Ti is only ~1 year old)
\n\nInsanely faster multi rendering. +2,033%\n Insanely faster NBody calculation. +1,271%
\n\nAmazing!
", "votes": 0, "replies": [] }, { "id": 340343, "author_name": "Mukharbek Organokov", "author_url": "", "post_date": "2018-06-09T00:50:28.873000", "content": "How comfortable you are with GTX 1080Ti?
", "votes": 0, "replies": [] }, { "id": 340344, "author_name": "Mukharbek Organokov", "author_url": "", "post_date": "2018-06-09T00:52:41.647000", "content": "How likely you will recommend to have some this kind of nice fast graphics card?\nBecause I did not use anything for comparison, but you might be done and can share experience.\nMerci beaucoup!
", "votes": 0, "replies": [] }, { "id": 340433, "author_name": "CPMP", "author_url": "", "post_date": "2018-06-09T06:29:19.910000", "content": "1080 TI (Pascal architecture) or more recent Volta architecture (Titan XP) are definitely making a difference when training complex CNN like resnet models. If you enter image recognition competition then having some of these really makes a difference. If you look at Carvana segmentation, the top teams all had several of these GPUs.
\n\nHere, I am doing some deep learning models, but nothing serious. My current score doe snot use any of them, hence the 1080 Ti is not used so far.
", "votes": 1, "replies": [] }, { "id": 340546, "author_name": "John Sweeney", "author_url": "", "post_date": "2018-06-09T14:54:50.947000", "content": "And the good news is that you can buy the high end cards for something close to list price again. It seems like the mining craziness is slowing down.
", "votes": 0, "replies": [] }, { "id": 340621, "author_name": "Mukharbek Organokov", "author_url": "", "post_date": "2018-06-09T19:25:24.260000", "content": "Thanks for the info! Very useful.
", "votes": 1, "replies": [] }, { "id": 340863, "author_name": "Robert", "author_url": "", "post_date": "2018-06-10T13:56:55.550000", "content": "You can actually use pytorch to do vectorized operations on the gpu, similar to working with numpy.
", "votes": 2, "replies": [] }, { "id": 340867, "author_name": "CPMP", "author_url": "", "post_date": "2018-06-10T14:02:26.390000", "content": "\n\n\nYou can actually use pytorch to do vectorized operations on the gpu, similar to working with numpy.
\n
Sure, but I don't have DBSCAN on top of that. If it was a gpu accelerated numpy then code using numpy would run unmodified. Anyway, that's what I am dreaming of.
", "votes": 1, "replies": [] }, { "id": 340954, "author_name": "Robert", "author_url": "", "post_date": "2018-06-10T17:37:39.923000", "content": "Oh, I thought you were trying to speed up your own code.
", "votes": 0, "replies": [] } ] }, { "id": 352659, "author_name": "Victor", "author_url": "", "post_date": "2018-07-04T20:14:13.553000", "content": "AMD 1 core 2400 MHz 2 Gb RAM - 1 hour to make solution.
", "votes": 2, "replies": [ { "id": 352663, "author_name": "macfarll", "author_url": "", "post_date": "2018-07-04T20:44:23.043000", "content": "1 hour per event?
", "votes": 0, "replies": [] }, { "id": 352670, "author_name": "Victor", "author_url": "", "post_date": "2018-07-04T21:10:18.983000", "content": "\n\n\nmacfarll wrote
\n \n\n\n1 hour per event?
\n
...per all 125
", "votes": 6, "replies": [] }, { "id": 352675, "author_name": "macfarll", "author_url": "", "post_date": "2018-07-04T21:35:00.883000", "content": "That's insane! Very impressive, your solution must work very differently from the dbscan unrolling approaches
", "votes": 0, "replies": [] }, { "id": 353074, "author_name": "Nicole Finnie", "author_url": "", "post_date": "2018-07-05T20:15:03.223000", "content": "@Victor, you should join the 2nd phase competition. Your result is impressive!
", "votes": 0, "replies": [] } ] }, { "id": 342716, "author_name": "Matt", "author_url": "", "post_date": "2018-06-14T00:09:33.637000", "content": "Currently using my $200 in Azure credit on a 4vCPU/8GB virtual machine. I cloned a backup VM so I can still tweak my model while creating a submission.
", "votes": 0, "replies": [] }, { "id": 341736, "author_name": "Borys Pratsiuk", "author_url": "", "post_date": "2018-06-12T07:08:06.243000", "content": "For education just regular macbook pro 2015 with core i7 + 16G ram + SSD.
", "votes": 0, "replies": [] }, { "id": 340225, "author_name": "Lewis Liu", "author_url": "", "post_date": "2018-06-08T16:53:15.280000", "content": "I am using cloud VM to parallel. i am just using my old 2012 macbook air to run locally.
", "votes": 0, "replies": [ { "id": 342743, "author_name": "Lewis Liu", "author_url": "", "post_date": "2018-06-14T02:33:26.917000", "content": "Not sure if everyone has the similar pattern. Our code runs about 7 minutes per dataset, without GPU on regular mac laptop. So just to make one submission, we used to run 14 hours. So I wrote a script to use Google Cloud preemptible VMs(much cheaper cost), spin up 125 VMs, then use simple script to combine all data into one file. Boom! 8 minutes, all done. Cost os about $1.5 USD
", "votes": 2, "replies": [] }, { "id": 342744, "author_name": "Lewis Liu", "author_url": "", "post_date": "2018-06-14T02:34:32.157000", "content": "https://github.com/liuxiao/CFlow, in case someone is interested
", "votes": 1, "replies": [] } ] }, { "id": 339786, "author_name": "macfarll", "author_url": "", "post_date": "2018-06-07T16:39:43.880000", "content": "I7 4770k \nGTX 1070\n16 GB ram (I just upped from 8)\n120gb ssd (I need more...)
\n\nI reused old ram and went with a cheap ssd, which are choices that made since at the time but are starting to feel more and more like a limitation.
", "votes": 0, "replies": [ { "id": 340346, "author_name": "Mukharbek Organokov", "author_url": "", "post_date": "2018-06-09T00:53:20.200000", "content": "How much you paid for a laptop with such options?
", "votes": 0, "replies": [] }, { "id": 343096, "author_name": "macfarll", "author_url": "", "post_date": "2018-06-14T17:24:17.203000", "content": "I just realized this was directed to me...
\n\nMy computer is a desktop, hard to put a price on it since I built/upgraded incrementally, but current parts are a little over 1000. Luckily I got my gpu before the crypto currency fad.
\n\nAlso worth noting that my current solution doesn't use my gpu at all.
", "votes": 0, "replies": [] } ] }, { "id": 339765, "author_name": "Zidmie", "author_url": "", "post_date": "2018-06-07T15:33:27.177000", "content": "I also have a 4 cores i7, with (just) 6GB. \nI started this contest with an old i5, then I decided it was a good time to invest in a new hardware :)
", "votes": 0, "replies": [] }, { "id": 340338, "author_name": "", "author_url": "", "post_date": "2018-06-09T00:43:48.550000", "content": "", "votes": 0, "replies": [ { "id": 340341, "author_name": "", "author_url": "", "post_date": "2018-06-09T00:49:15.803000", "content": "", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "339701": "The kaggle kernel is ok~~", "339696": "I am using a 4 cores i7 with 64 GB and a GTX 1080 Ti. Memory is not really an issue here, my processes are way smaller than the available RAM.", "341557": "Newbie here. Setup an I9-7900 with 64GB memory and two 1080 ti cards, with two stripped 500GB NVNe pcie drives. I got the i9 for the 44 lanes to push data to the 1080 ti cards. The pcie drives give me ~6 GB of throughput. I may be a little light on memory. \n\nI have no dreams of doing good in this being my first kaggle, just want to learn more about data science.", "339720": "A beginner's question regarding GPU: apart from using it for neural network or gradient boosting frameworks, do you often try to vectorize regular functions which are time-consuming and run them on GPU? I suppose the answer depends on the details?\n\nAnswer to your question: for now just a notebook with i5, 12GB and GeForce 840m.", "352659": "AMD 1 core 2400 MHz 2 Gb RAM - 1 hour to make solution.", "342716": "Currently using my $200 in Azure credit on a 4vCPU/8GB virtual machine. I cloned a backup VM so I can still tweak my model while creating a submission.", "341736": "For education just regular macbook pro 2015 with core i7 + 16G ram + SSD.", "340225": "I am using cloud VM to parallel. i am just using my old 2012 macbook air to run locally.", "339786": "I7 4770k \nGTX 1070\n16 GB ram (I just upped from 8)\n120gb ssd (I need more...)\n\nI reused old ram and went with a cheap ssd, which are choices that made since at the time but are starting to feel more and more like a limitation. ", "339765": "I also have a 4 cores i7, with (just) 6GB. \nI started this contest with an old i5, then I decided it was a good time to invest in a new hardware :)\n", "340338": "" } }