Hi,
\n\nDisclaimer: I am new to Data Science and Kaggle. hence my queries can sound naive. Kindly pardon me for the same.
\n\nI am reaching out to the masters to point to some reference, some reading, which can help a novice like me understand how to deal with problems like this, where the source data is a bunch of images.
\n\nWhat are the ways in which we create datasets from images? What sort of feature engineering do we do in these cases?
\n\nThanks and regards\nSayantan Raha
", "messages": [ { "id": 113944, "postDate": "2016-04-06T11:32:27.887Z", "content": "+1 for Keras
\n\nAlso, a resource I recently saw that may come in handy:
\n\n"A guide to convolution arithmetic for deep learning"
\n\nhttp://arxiv.org/abs/1603.07285
", "rawMarkdown": "+1 for Keras\r\n\r\nAlso, a resource I recently saw that may come in handy:\r\n\r\n\"A guide to convolution arithmetic for deep learning\"\r\n\r\nhttp://arxiv.org/abs/1603.07285", "votes": 12 }, { "id": 114523, "postDate": "2016-04-11T15:59:00.917Z", "content": "Hi Sayantan,
\n\nPlease check Stanford's course on Convolution Networks. Possibly the best place to learn convolution networks and it's applications in Machine Vision. However, this course assume basics in Machine Learning.
\n\n", "rawMarkdown": "Hi Sayantan,\r\n\r\nPlease check Stanford's course on Convolution Networks. Possibly the best place to learn convolution networks and it's applications in Machine Vision. However, this course assume basics in Machine Learning. \r\n\r\nhttp://cs231n.stanford.edu/", "votes": 5 }, { "id": 113926, "postDate": "2016-04-06T09:15:14.237Z", "content": "Hi,
\n\nDisclaimer: I am new to Data Science and Kaggle. hence my queries can sound naive. Kindly pardon me for the same.
\n\nI am reaching out to the masters to point to some reference, some reading, which can help a novice like me understand how to deal with problems like this, where the source data is a bunch of images.
\n\nWhat are the ways in which we create datasets from images? What sort of feature engineering do we do in these cases?
\n\nThanks and regards\nSayantan Raha
", "rawMarkdown": "Hi,\r\n\r\nDisclaimer: I am new to Data Science and Kaggle. hence my queries can sound naive. Kindly pardon me for the same.\r\n\r\nI am reaching out to the masters to point to some reference, some reading, which can help a novice like me understand how to deal with problems like this, where the source data is a bunch of images. \r\n\r\nWhat are the ways in which we create datasets from images? What sort of feature engineering do we do in these cases?\r\n\r\nThanks and regards\r\nSayantan Raha", "votes": 6 }, { "id": 113943, "postDate": "2016-04-06T11:16:01.087Z", "content": "Deep learning is the nuclear weapon for such computer vision problem :-) Unlike other competitions on Kaggle, there is no way to apply typical feature engineering in this competition.
\n\nOn the condition that you are familiar with Python, Keras (http://keras.io/) will be a wise choice.
", "rawMarkdown": "Deep learning is the nuclear weapon for such computer vision problem :-) Unlike other competitions on Kaggle, there is no way to apply typical feature engineering in this competition.\r\n\r\nOn the condition that you are familiar with Python, Keras (http://keras.io/) will be a wise choice.", "votes": 5 }, { "id": 113960, "postDate": "2016-04-06T13:16:59.703Z", "content": "[quote=pasqui;113953]
\n\n[quote=ChaseWind;113943]
\n\nDeep learning is the nuclear weapon for such computer vision problem :-) Unlike other competitions on Kaggle, there is no way to apply typical feature engineering in this competition.
\n\nOn the condition that you are familiar with Python, Keras (http://keras.io/) will be a wise choice.
\n\n[/quote]
\n\nHi, \n[I am also a novice; no libs war]\ncan I ask you why you suggested Keras and not Caffe?
\n\nbest,
\n\n[/quote]
\n\n1, Installation of Caffe is much more complicated than Keras.
\n\n2, Keras is a high level package backended by Theano and TensorFlow. The syntax of Keras is simpler.
\n\n3, Keras supports Caffe models.
\n\n4, Personally, I have ever encountered many nasty defects with Caffe, such as compatibility problems.
", "rawMarkdown": "[quote=pasqui;113953]\r\n\r\n[quote=ChaseWind;113943]\r\n\r\nDeep learning is the nuclear weapon for such computer vision problem :-) Unlike other competitions on Kaggle, there is no way to apply typical feature engineering in this competition.\r\n\r\nOn the condition that you are familiar with Python, Keras (http://keras.io/) will be a wise choice.\r\n\r\n[/quote]\r\n\r\nHi, \r\n[I am also a novice; no libs war]\r\ncan I ask you why you suggested Keras and not Caffe?\r\n\r\nbest,\r\n\r\n[/quote]\r\n\r\n1, Installation of Caffe is much more complicated than Keras.\r\n\r\n2, Keras is a high level package backended by Theano and TensorFlow. The syntax of Keras is simpler.\r\n\r\n3, Keras supports Caffe models.\r\n\r\n4, Personally, I have ever encountered many nasty defects with Caffe, such as compatibility problems.", "votes": 4 }, { "id": 114597, "postDate": "2016-04-12T10:34:05.943Z", "content": "@narsil\nThe deep learning libraries available for R are far less developed and far less supported than for Python, which has numerous great tools (Keras, Lasagne, TensorFlow, etc..).
\n\nMXNet is now available, and seems to have a lot of potential, but I've had lots of trouble with it (random crashes...), and the documentation is very sparse for the the moment.
\n\nSo yeah, if you want to use some deep neural networks, Python is probably the way to go.
", "rawMarkdown": "@narsil\r\nThe deep learning libraries available for R are far less developed and far less supported than for Python, which has numerous great tools (Keras, Lasagne, TensorFlow, etc..). \r\n\r\nMXNet is now available, and seems to have a lot of potential, but I've had lots of trouble with it (random crashes...), and the documentation is very sparse for the the moment.\r\n\r\nSo yeah, if you want to use some deep neural networks, Python is probably the way to go.", "votes": 1 }, { "id": 114163, "postDate": "2016-04-07T21:25:13.687Z", "content": "g2.2xlarge AWS spot instances are no longer an option: the price instability is way too high since some months ago, and they are usually more expensive (sometimes 10x more) than standard instances.
", "rawMarkdown": "g2.2xlarge AWS spot instances are no longer an option: the price instability is way too high since some months ago, and they are usually more expensive (sometimes 10x more) than standard instances. ", "votes": 1 }, { "id": 113994, "postDate": "2016-04-06T18:20:19.993Z", "content": "[quote=RohinArora;113991]
\n\n[quote=inversion;113977]
\n\n[quote=the1owl;113975]
\n\nNow we just need one of these
\n\nhttp://www.nvidia.com/object/deep-learning-system.html
\n\n[/quote]
\n\n$129,000. Whoa.
\n\nSeems better to just get a few Titan X GPUs. :-)
\n\n[/quote]
\n\nHi. Another novice here. Is it possible to take part in this competition without a Nvidia GPU? My laptop has Intel i5 3rd gen and AMD graphic card. My guess is no. I wanted your opinion.
\n\n[/quote]
\n\nYou can use a cloud GPU instance. AWS offers such instances (it is the G2 family).
\n\nYou can find more documentation here: https://aws.amazon.com/ec2/instance-types/.
\n\nLet me know if you need more help!
", "rawMarkdown": "[quote=RohinArora;113991]\r\n\r\n[quote=inversion;113977]\r\n\r\n[quote=the1owl;113975]\r\n\r\nNow we just need one of these\r\n\r\n[http://www.nvidia.com/object/deep-learning-system.html][1]\r\n\r\n\r\n[/quote]\r\n\r\n$129,000. Whoa.\r\n\r\nSeems better to just get a few Titan X GPUs. :-)\r\n\r\n\r\n[/quote]\r\n\r\n\r\nHi. Another novice here. Is it possible to take part in this competition without a Nvidia GPU? My laptop has Intel i5 3rd gen and AMD graphic card. My guess is no. I wanted your opinion. \r\n\r\n[/quote]\r\n\r\n\r\nYou can use a cloud GPU instance. [**AWS**][2] offers such instances (it is the **G2** family). \r\n\r\nYou can find more documentation here: https://aws.amazon.com/ec2/instance-types/. \r\n\r\nLet me know if you need more help!\r\n\r\n\r\n [1]: http://www.nvidia.com/object/deep-learning-system.html\r\n [2]: https://aws.amazon.com/ec2/instance-types/", "votes": 1 }, { "id": 114141, "postDate": "2016-04-07T18:35:27.297Z", "content": "$0.3/hour sounds pretty cheap but I found that I very quickly ran up a bill of over $100 just get install everything and run a few sample networks. In the end I decided that I should just put together a desktop Linux box with a decent Nvidia card in it. I built one around a 980 Ti and it runs 2.5x faster than the AWS instance.
\n\nBut it wasn't cheap. Probably close to $2,000 for everything.
\n\nIf you do go the AWS route, note that you are charged for the instance just being up even if it's idle.
", "rawMarkdown": "$0.3/hour sounds pretty cheap but I found that I very quickly ran up a bill of over $100 just get install everything and run a few sample networks. In the end I decided that I should just put together a desktop Linux box with a decent Nvidia card in it. I built one around a 980 Ti and it runs 2.5x faster than the AWS instance. \r\n\r\nBut it wasn't cheap. Probably close to $2,000 for everything.\r\n\r\nIf you do go the AWS route, note that you are charged for the instance just being up even if it's idle.", "votes": 2 }, { "id": 114000, "postDate": "2016-04-06T18:35:10.060Z", "content": "[quote=RohinArora;113991]
\n\n[quote=inversion;113977]
\n\n[quote=the1owl;113975]
\n\nNow we just need one of these
\n\nhttp://www.nvidia.com/object/deep-learning-system.html
\n\n[/quote]
\n\n$129,000. Whoa.
\n\nSeems better to just get a few Titan X GPUs. :-)
\n\n[/quote]
\n\nHi. Another novice here. Is it possible to take part in this competition without a Nvidia GPU? My laptop has Intel i5 3rd gen and AMD graphic card. My guess is no. I wanted your opinion.
\n\n[/quote]
\n\nIt should be possible to take part, but training times will be too long to do much in the way of tuning. To get started, you will need to limit the scope of what you attempt - such as reducing the size of the images and trying out architectures on just 10% of the data.
\n\nI don't have any GPU-based hardware, but I still intend to play along, and pick up a little experience.
\n\nIf I think I will do well enough, I might try AWS this time for the experience of that, too. But I'm going to get started on my 3-year-old Macbook Pro, and expect the first serious training session to take several days once I've settled on some working code.
\n\nUpdate: Trying out ZFTurbo's demo code using Keras without a GPU takes about 3 hours on my laptop. Reducing the image size makes it run even faster. Plenty to look at and learn inside that script, and opportunity to try changing things, tuning etc.
", "rawMarkdown": "[quote=RohinArora;113991]\r\n\r\n[quote=inversion;113977]\r\n\r\n[quote=the1owl;113975]\r\n\r\nNow we just need one of these\r\n\r\n[http://www.nvidia.com/object/deep-learning-system.html][1]\r\n\r\n\r\n [1]: http://www.nvidia.com/object/deep-learning-system.html\r\n\r\n[/quote]\r\n\r\n$129,000. Whoa.\r\n\r\nSeems better to just get a few Titan X GPUs. :-)\r\n\r\n\r\n[/quote]\r\n\r\n\r\nHi. Another novice here. Is it possible to take part in this competition without a Nvidia GPU? My laptop has Intel i5 3rd gen and AMD graphic card. My guess is no. I wanted your opinion. \r\n\r\n[/quote]\r\n\r\nIt should be possible to take part, but training times will be too long to do much in the way of tuning. To get started, you will need to limit the scope of what you attempt - such as reducing the size of the images and trying out architectures on just 10% of the data. \r\n\r\nI don't have any GPU-based hardware, but I still intend to play along, and pick up a little experience.\r\n\r\nIf I think I will do well enough, I *might* try AWS this time for the experience of that, too. But I'm going to get started on my 3-year-old Macbook Pro, and expect the first serious training session to take several days once I've settled on some working code.\r\n\r\n----\r\n\r\nUpdate: Trying out *ZFTurbo*'s demo code using Keras without a GPU takes about 3 hours on my laptop. Reducing the image size makes it run even faster. Plenty to look at and learn inside that script, and opportunity to try changing things, tuning etc.\r\n\r\n\r\n", "votes": 2 }, { "id": 113977, "postDate": "2016-04-06T15:16:31.133Z", "content": "[quote=the1owl;113975]
\n\nNow we just need one of these
\n\nhttp://www.nvidia.com/object/deep-learning-system.html
\n\n[/quote]
\n\n$129,000. Whoa.
\n\nSeems better to just get a few Titan X GPUs. :-)
", "rawMarkdown": "[quote=the1owl;113975]\r\n\r\nNow we just need one of these\r\n\r\n[http://www.nvidia.com/object/deep-learning-system.html][1]\r\n\r\n\r\n [1]: http://www.nvidia.com/object/deep-learning-system.html\r\n\r\n[/quote]\r\n\r\n$129,000. Whoa.\r\n\r\nSeems better to just get a few Titan X GPUs. :-)\r\n", "votes": 2 }, { "id": 113958, "postDate": "2016-04-06T13:13:44.323Z", "content": "[quote=inversion;113944]
\n\n+1 for Keras
\n\nAlso, a resource I recently saw that may come in handy:
\n\n"A guide to convolution arithmetic for deep learning"
\n\nhttp://arxiv.org/abs/1603.07285
\n\n[/quote]
\n\nThanks for the article
", "rawMarkdown": "[quote=inversion;113944]\r\n\r\n+1 for Keras\r\n\r\nAlso, a resource I recently saw that may come in handy:\r\n\r\n\"A guide to convolution arithmetic for deep learning\"\r\n\r\nhttp://arxiv.org/abs/1603.07285\r\n\r\n[/quote]\r\n\r\nThanks for the article", "votes": 2 }, { "id": 113953, "postDate": "2016-04-06T13:02:07.243Z", "content": "[quote=ChaseWind;113943]
\n\nDeep learning is the nuclear weapon for such computer vision problem :-) Unlike other competitions on Kaggle, there is no way to apply typical feature engineering in this competition.
\n\nOn the condition that you are familiar with Python, Keras (http://keras.io/) will be a wise choice.
\n\n[/quote]
\n\nHi, \n[I am also a novice; no libs war]\ncan I ask you why you suggested Keras and not Caffe?
\n\nbest,
", "rawMarkdown": "[quote=ChaseWind;113943]\r\n\r\nDeep learning is the nuclear weapon for such computer vision problem :-) Unlike other competitions on Kaggle, there is no way to apply typical feature engineering in this competition.\r\n\r\nOn the condition that you are familiar with Python, Keras (http://keras.io/) will be a wise choice.\r\n\r\n[/quote]\r\n\r\nHi, \r\n[I am also a novice; no libs war]\r\ncan I ask you why you suggested Keras and not Caffe?\r\n\r\nbest,", "votes": 2 }, { "id": 124793, "postDate": "2016-06-22T12:25:28.700Z", "content": "[quote=Yassine Alouini;114384]
\n\nAs mentioned by other participants, ZFTurbo code is a great place to start for this challenge.
\n\nI am playing with the code right now, doing some tweaking and hopefully improving it.
\n\nIf I find something interesting, I will post it. Stay tuned!
\n\nYou can find the original code here: https://github.com/ZFTurbo/KAGGLE_DISTRACTED_DRIVER
\n\n[/quote]
\n\nWe need R copy for the Code , How we can translate it to R?
", "rawMarkdown": "[quote=Yassine Alouini;114384]\r\n\r\nAs mentioned by other participants, **ZFTurbo** code is a great place to start for this challenge. \r\n\r\nI am playing with the code right now, doing some tweaking and hopefully improving it. \r\n\r\nIf I find something interesting, I will post it. Stay tuned!\r\n\r\nYou can find the original code here: https://github.com/ZFTurbo/KAGGLE_DISTRACTED_DRIVER\r\n\r\n[/quote]\r\n\r\nWe need R copy for the Code , How we can translate it to R?\r\n", "votes": -2 }, { "id": 124801, "postDate": "2016-06-22T13:16:46.267Z", "content": "I am thinking of getting that little NUC :) Do you have something like Razer Core for dedicated GPU with NUC?
\n\n[quote=the1owl;115077]
\n\nNow I want one of these...
\n\nhttp://www.intel.com/content/www/us/en/processors/xeon/xeon-phi-detail.html
\n\nBut I guess I have to settle for this one, my current little NUC did ok these past few competitions.
\n\nhttp://www.intel.com/content/www/us/en/nuc/change-the-game-with-nuc.html
\n\n[/quote]
", "rawMarkdown": "I am thinking of getting that little NUC :) Do you have something like Razer Core for dedicated GPU with NUC? \r\n\r\n[quote=the1owl;115077]\r\n\r\nNow I want one of these...\r\n\r\nhttp://www.intel.com/content/www/us/en/processors/xeon/xeon-phi-detail.html\r\n\r\nBut I guess I have to settle for this one, my current little NUC did ok these past few competitions.\r\n\r\nhttp://www.intel.com/content/www/us/en/nuc/change-the-game-with-nuc.html\r\n\r\n[/quote]\r\n" }, { "id": 124794, "postDate": "2016-06-22T12:26:28.130Z", "content": "[quote=NaomiFridman;114588]
\n\nHi,
\n\nDisclaimer: I am total newbie in Data Science and Kaggle. I am a mathematician, I did some courses in Machine learning, but experienced only on small clean data on matlab and Python.
\n\nMy first challenge is to set up working environment. \nDo you think setting it up on windows is a good Idea ? I found this article : \nhttps://datanoord.com/2016/02/01/setup-a-deep-learning-environment-on-windows-theano-keras-with-gpu-enabled/
\n\nI installed Ubuntu on Oracle virtual machine, so I can try setting up working environment there as well.
\n\nThanks and regards,
\n\nNaomi Fridman
\n\n[/quote]\n Page cannot be found !!!
", "rawMarkdown": "[quote=NaomiFridman;114588]\r\n\r\nHi,\r\n\r\nDisclaimer: I am total newbie in Data Science and Kaggle. I am a mathematician, I did some courses in Machine learning, but experienced only on small clean data on matlab and Python. \r\n\r\nMy first challenge is to set up working environment. \r\nDo you think setting it up on windows is a good Idea ? I found this article : \r\nhttps://datanoord.com/2016/02/01/setup-a-deep-learning-environment-on-windows-theano-keras-with-gpu-enabled/\r\n\r\nI installed Ubuntu on Oracle virtual machine, so I can try setting up working environment there as well.\r\n\r\nThanks and regards,\r\n\r\nNaomi Fridman\r\n\r\n\r\n[/quote]\r\n Page cannot be found !!!" }, { "id": 123212, "postDate": "2016-06-10T11:23:07.433Z", "content": "Hi Every One
\n\nThe tutorial of Facial Keypoint Detection : https://www.kaggle.com/c/facial-keypoints-detection/details/getting-started-with-r
\n\nStill need more other tutorials , I have question based on R is my favorite language to start coding:
\n\n1-Is there any other sources do step by step tutorial using R for Image processing and extraction (recognition)?
\n\n2- Can you use Matlab Instead for creating the algorithm ?
\n\n3-Is there a learning sources to combine R with Matlab in coding our project (StateFarm)?
\n\nThank you in Advanced
", "rawMarkdown": "Hi Every One \r\n\r\nThe tutorial of Facial Keypoint Detection : https://www.kaggle.com/c/facial-keypoints-detection/details/getting-started-with-r\r\n\r\nStill need more other tutorials , I have question based on R is my favorite language to start coding:\r\n\r\n1-Is there any other sources do step by step tutorial using R for Image processing and extraction (recognition)?\r\n\r\n2- Can you use Matlab Instead for creating the algorithm ?\r\n\r\n3-Is there a learning sources to combine R with Matlab in coding our project (StateFarm)?\r\n\r\nThank you in Advanced " }, { "id": 114843, "postDate": "2016-04-14T07:35:06.917Z", "rawMarkdown": "" }, { "id": 114637, "postDate": "2016-04-12T16:56:52.410Z", "content": "@NaomiFridman Linux is much better supported for deep learning than Windows. You probably won't be happy running Ubuntu in a VM, except for initial explorations. It's too slow and doesn't support a GPU. If you need to keep Windows you could consider a dual-boot setup.
\n\nI'm using a dedicated Linux machine (Ubuntu 14.04) on which I run
\n\n[quote=NaomiFridman;114588]
\n\nHi,
\n\nDisclaimer: I am total newbie in Data Science and Kaggle. I am a mathematician, I did some courses in Machine learning, but experienced only on small clean data on matlab and Python.
\n\nMy first challenge is to set up working environment. \nDo you think setting it up on windows is a good Idea ? I found this article : \nhttps://datanoord.com/2016/02/01/setup-a-deep-learning-environment-on-windows-theano-keras-with-gpu-enabled/
\n\nI installed Ubuntu on Oracle virtual machine, so I can try setting up working environment there as well.
\n\nThanks and regards,
\n\nNaomi Fridman
\n\n[/quote]
\n\nHi, \nusually, for Computer Vision tasks I would like to use all the computational power available; thus, if we are talking about desktops/laptops, I would not go for a VM,
\n\nbest,
", "rawMarkdown": "[quote=NaomiFridman;114588]\r\n\r\nHi,\r\n\r\nDisclaimer: I am total newbie in Data Science and Kaggle. I am a mathematician, I did some courses in Machine learning, but experienced only on small clean data on matlab and Python. \r\n\r\nMy first challenge is to set up working environment. \r\nDo you think setting it up on windows is a good Idea ? I found this article : \r\nhttps://datanoord.com/2016/02/01/setup-a-deep-learning-environment-on-windows-theano-keras-with-gpu-enabled/\r\n\r\nI installed Ubuntu on Oracle virtual machine, so I can try setting up working environment there as well.\r\n\r\nThanks and regards,\r\n\r\nNaomi Fridman\r\n\r\n\r\n[/quote]\r\n\r\nHi, \r\nusually, for Computer Vision tasks I would like to use all the computational power available; thus, if we are talking about desktops/laptops, I would not go for a VM, \r\n\r\nbest," }, { "id": 114588, "postDate": "2016-04-12T09:04:04.913Z", "content": "Hi,
\n\nDisclaimer: I am total newbie in Data Science and Kaggle. I am a mathematician, I did some courses in Machine learning, but experienced only on small clean data on matlab and Python.
\n\nMy first challenge is to set up working environment. \nDo you think setting it up on windows is a good Idea ? I found this article : \nhttps://datanoord.com/2016/02/01/setup-a-deep-learning-environment-on-windows-theano-keras-with-gpu-enabled/
\n\nI installed Ubuntu on Oracle virtual machine, so I can try setting up working environment there as well.
\n\nThanks and regards,
\n\nNaomi Fridman
", "rawMarkdown": "Hi,\r\n\r\nDisclaimer: I am total newbie in Data Science and Kaggle. I am a mathematician, I did some courses in Machine learning, but experienced only on small clean data on matlab and Python. \r\n\r\nMy first challenge is to set up working environment. \r\nDo you think setting it up on windows is a good Idea ? I found this article : \r\nhttps://datanoord.com/2016/02/01/setup-a-deep-learning-environment-on-windows-theano-keras-with-gpu-enabled/\r\n\r\nI installed Ubuntu on Oracle virtual machine, so I can try setting up working environment there as well.\r\n\r\nThanks and regards,\r\n\r\nNaomi Fridman\r\n" }, { "id": 114573, "postDate": "2016-04-12T02:29:03.953Z", "content": "For someone who wants to do it in R, is there an obvious way to go (like Keras for Python)?\nOr you suggest that I forget R, and learn python to do it?
", "rawMarkdown": "For someone who wants to do it in R, is there an obvious way to go (like Keras for Python)?\r\nOr you suggest that I forget R, and learn python to do it?\r\n\r\n" }, { "id": 114444, "postDate": "2016-04-10T22:56:47.767Z", "content": "Hey Yassine\nHope all is well\nIt is a great place to start, however which file would you tackle first.\nThere are 3 posted thus far. Thanks
", "rawMarkdown": "Hey Yassine\r\nHope all is well\r\nIt is a great place to start, however which file would you tackle first.\r\nThere are 3 posted thus far. Thanks" }, { "id": 114384, "postDate": "2016-04-10T08:21:31.530Z", "content": "As mentioned by other participants, ZFTurbo code is a great place to start for this challenge.
\n\nI am playing with the code right now, doing some tweaking and hopefully improving it.
\n\nIf I find something interesting, I will post it. Stay tuned!
\n\nYou can find the original code here: https://github.com/ZFTurbo/KAGGLE_DISTRACTED_DRIVER
", "rawMarkdown": "As mentioned by other participants, **ZFTurbo** code is a great place to start for this challenge. \r\n\r\nI am playing with the code right now, doing some tweaking and hopefully improving it. \r\n\r\nIf I find something interesting, I will post it. Stay tuned!\r\n\r\nYou can find the original code here: https://github.com/ZFTurbo/KAGGLE_DISTRACTED_DRIVER" }, { "id": 114257, "postDate": "2016-04-08T16:28:05.010Z", "content": "Thanks Inversion, ChaseWind for the references. I will try to learn Python, Keras over the next 3 month period. Atleast I will get started. \nThanks for the references, will go over them. BTW I have successfully installed all the components CUDA, VC, Anaconda Theano and Keras. Time to start studying !!!
\n\nThanks and regards\nSayantan Raha
", "rawMarkdown": "Thanks Inversion, ChaseWind for the references. I will try to learn Python, Keras over the next 3 month period. Atleast I will get started. \r\nThanks for the references, will go over them. BTW I have successfully installed all the components CUDA, VC, Anaconda Theano and Keras. Time to start studying !!!\r\n\r\nThanks and regards\r\nSayantan Raha" }, { "id": 114120, "postDate": "2016-04-07T16:44:04.497Z", "content": "[quote=BrettAmdur;114092]
\n\nCan anyone take a ballpark stab at what the AWS fees would be for someone to just have fun and learn with this competition, as opposed to someone who needs to train enough data to be one of the leaders?
\n\n[/quote]
\n\nIt is possible to have a spot instance of "g2.2xlarge" at the price $0.3/h. Please keep in mind that the price is not stable at all. The execution time of the script could be less than 1 hour.
", "rawMarkdown": "[quote=BrettAmdur;114092]\r\n\r\nCan anyone take a ballpark stab at what the AWS fees would be for someone to just have fun and learn with this competition, as opposed to someone who needs to train enough data to be one of the leaders? \r\n\r\n[/quote]\r\n\r\nIt is possible to have a spot instance of \"g2.2xlarge\" at the price $0.3/h. Please keep in mind that the price is not stable at all. The execution time of the script could be less than 1 hour." }, { "id": 114092, "postDate": "2016-04-07T14:46:34.057Z", "content": "Can anyone take a ballpark stab at what the AWS fees would be for someone to just have fun and learn with this competition, as opposed to someone who needs to train enough data to be one of the leaders?
", "rawMarkdown": "Can anyone take a ballpark stab at what the AWS fees would be for someone to just have fun and learn with this competition, as opposed to someone who needs to train enough data to be one of the leaders? " }, { "id": 114044, "postDate": "2016-04-07T07:46:32.397Z", "content": "[quote=Neil Slater;114043]
\n\n[quote=pasqui;114001]
\n\n\n\n\nBut I'm going to get started on my 3-year-old Macbook Pro
\n
It should be CUDA-capable
\n\n[/quote]
\n\nNot in any meaningful way. I checked. Integrated graphics board: Intel HD Graphics 4000 1024 MB. See http://apple.stackexchange.com/questions/93354/is-my-mac-book-pro-suitable-for-cuda
\n\nSome of the top-end models had graphics cards where it might be possible.
\n\n[/quote]
\n\nI am sorry. 3 years means that only the 15" in its top conf has a a Nvidia card.
", "rawMarkdown": "[quote=Neil Slater;114043]\r\n\r\n[quote=pasqui;114001]\r\n\r\n> But I'm going to get started on my 3-year-old Macbook Pro\r\n\r\nIt should be CUDA-capable\r\n\r\n[/quote]\r\n\r\nNot in any meaningful way. I checked. Integrated graphics board: Intel HD Graphics 4000 1024 MB. See http://apple.stackexchange.com/questions/93354/is-my-mac-book-pro-suitable-for-cuda\r\n\r\nSome of the top-end models had graphics cards where it might be possible.\r\n\r\n\r\n[/quote]\r\n\r\nI am sorry. 3 years means that only the 15\" in its top conf has a a Nvidia card." }, { "id": 114043, "postDate": "2016-04-07T07:30:19.007Z", "content": "[quote=pasqui;114001]
\n\n\n\n\nBut I'm going to get started on my 3-year-old Macbook Pro
\n
It should be CUDA-capable
\n\n[/quote]
\n\nNot in any meaningful way. I checked. Integrated graphics board: Intel HD Graphics 4000 1024 MB. See http://apple.stackexchange.com/questions/93354/is-my-mac-book-pro-suitable-for-cuda
\n\nSome of the top-end models had graphics cards where it might be possible.
", "rawMarkdown": "[quote=pasqui;114001]\r\n\r\n> But I'm going to get started on my 3-year-old Macbook Pro\r\n\r\nIt should be CUDA-capable\r\n\r\n[/quote]\r\n\r\nNot in any meaningful way. I checked. Integrated graphics board: Intel HD Graphics 4000 1024 MB. See http://apple.stackexchange.com/questions/93354/is-my-mac-book-pro-suitable-for-cuda\r\n\r\nSome of the top-end models had graphics cards where it might be possible.\r\n" }, { "id": 114001, "postDate": "2016-04-06T18:38:35.270Z", "content": "\n\n\nBut I'm going to get started on my 3-year-old Macbook Pro
\n
It should be CUDA-capable
", "rawMarkdown": "> But I'm going to get started on my 3-year-old Macbook Pro\r\n\r\nIt should be CUDA-capable" }, { "id": 115077, "postDate": "2016-04-16T03:50:40.657Z", "rawMarkdown": "", "isDeleted": true }, { "id": 113999, "postDate": "2016-04-06T18:29:28.683Z", "rawMarkdown": "", "isDeleted": true }, { "id": 113991, "postDate": "2016-04-06T18:13:46.027Z", "rawMarkdown": "", "isDeleted": true }, { "id": 113975, "postDate": "2016-04-06T14:46:47.887Z", "rawMarkdown": "", "votes": 1, "isDeleted": true } ], "comments": [ { "id": 113944, "author_name": "inversion", "author_url": "", "post_date": "2016-04-06T11:32:27.887000", "content": "+1 for Keras
\n\nAlso, a resource I recently saw that may come in handy:
\n\n"A guide to convolution arithmetic for deep learning"
\n\nhttp://arxiv.org/abs/1603.07285
", "votes": 12, "replies": [] }, { "id": 114523, "author_name": "Venkata Dikshit", "author_url": "", "post_date": "2016-04-11T15:59:00.917000", "content": "Hi Sayantan,
\n\nPlease check Stanford's course on Convolution Networks. Possibly the best place to learn convolution networks and it's applications in Machine Vision. However, this course assume basics in Machine Learning.
\n\n", "votes": 5, "replies": [] }, { "id": 113943, "author_name": "ChaseWind", "author_url": "", "post_date": "2016-04-06T11:16:01.087000", "content": "Deep learning is the nuclear weapon for such computer vision problem :-) Unlike other competitions on Kaggle, there is no way to apply typical feature engineering in this competition.
\n\nOn the condition that you are familiar with Python, Keras (http://keras.io/) will be a wise choice.
", "votes": 5, "replies": [] }, { "id": 113960, "author_name": "ChaseWind", "author_url": "", "post_date": "2016-04-06T13:16:59.703000", "content": "[quote=pasqui;113953]
\n\n[quote=ChaseWind;113943]
\n\nDeep learning is the nuclear weapon for such computer vision problem :-) Unlike other competitions on Kaggle, there is no way to apply typical feature engineering in this competition.
\n\nOn the condition that you are familiar with Python, Keras (http://keras.io/) will be a wise choice.
\n\n[/quote]
\n\nHi, \n[I am also a novice; no libs war]\ncan I ask you why you suggested Keras and not Caffe?
\n\nbest,
\n\n[/quote]
\n\n1, Installation of Caffe is much more complicated than Keras.
\n\n2, Keras is a high level package backended by Theano and TensorFlow. The syntax of Keras is simpler.
\n\n3, Keras supports Caffe models.
\n\n4, Personally, I have ever encountered many nasty defects with Caffe, such as compatibility problems.
", "votes": 4, "replies": [] }, { "id": 114597, "author_name": "Gerome Pistre", "author_url": "", "post_date": "2016-04-12T10:34:05.943000", "content": "@narsil\nThe deep learning libraries available for R are far less developed and far less supported than for Python, which has numerous great tools (Keras, Lasagne, TensorFlow, etc..).
\n\nMXNet is now available, and seems to have a lot of potential, but I've had lots of trouble with it (random crashes...), and the documentation is very sparse for the the moment.
\n\nSo yeah, if you want to use some deep neural networks, Python is probably the way to go.
", "votes": 1, "replies": [] }, { "id": 114163, "author_name": "juandoso", "author_url": "", "post_date": "2016-04-07T21:25:13.687000", "content": "g2.2xlarge AWS spot instances are no longer an option: the price instability is way too high since some months ago, and they are usually more expensive (sometimes 10x more) than standard instances.
", "votes": 1, "replies": [] }, { "id": 113994, "author_name": "Yassine Alouini", "author_url": "", "post_date": "2016-04-06T18:20:19.993000", "content": "[quote=RohinArora;113991]
\n\n[quote=inversion;113977]
\n\n[quote=the1owl;113975]
\n\nNow we just need one of these
\n\nhttp://www.nvidia.com/object/deep-learning-system.html
\n\n[/quote]
\n\n$129,000. Whoa.
\n\nSeems better to just get a few Titan X GPUs. :-)
\n\n[/quote]
\n\nHi. Another novice here. Is it possible to take part in this competition without a Nvidia GPU? My laptop has Intel i5 3rd gen and AMD graphic card. My guess is no. I wanted your opinion.
\n\n[/quote]
\n\nYou can use a cloud GPU instance. AWS offers such instances (it is the G2 family).
\n\nYou can find more documentation here: https://aws.amazon.com/ec2/instance-types/.
\n\nLet me know if you need more help!
", "votes": 1, "replies": [] }, { "id": 114141, "author_name": "gauss256", "author_url": "", "post_date": "2016-04-07T18:35:27.297000", "content": "$0.3/hour sounds pretty cheap but I found that I very quickly ran up a bill of over $100 just get install everything and run a few sample networks. In the end I decided that I should just put together a desktop Linux box with a decent Nvidia card in it. I built one around a 980 Ti and it runs 2.5x faster than the AWS instance.
\n\nBut it wasn't cheap. Probably close to $2,000 for everything.
\n\nIf you do go the AWS route, note that you are charged for the instance just being up even if it's idle.
", "votes": 2, "replies": [] }, { "id": 114000, "author_name": "Neil Slater", "author_url": "", "post_date": "2016-04-06T18:35:10.060000", "content": "[quote=RohinArora;113991]
\n\n[quote=inversion;113977]
\n\n[quote=the1owl;113975]
\n\nNow we just need one of these
\n\nhttp://www.nvidia.com/object/deep-learning-system.html
\n\n[/quote]
\n\n$129,000. Whoa.
\n\nSeems better to just get a few Titan X GPUs. :-)
\n\n[/quote]
\n\nHi. Another novice here. Is it possible to take part in this competition without a Nvidia GPU? My laptop has Intel i5 3rd gen and AMD graphic card. My guess is no. I wanted your opinion.
\n\n[/quote]
\n\nIt should be possible to take part, but training times will be too long to do much in the way of tuning. To get started, you will need to limit the scope of what you attempt - such as reducing the size of the images and trying out architectures on just 10% of the data.
\n\nI don't have any GPU-based hardware, but I still intend to play along, and pick up a little experience.
\n\nIf I think I will do well enough, I might try AWS this time for the experience of that, too. But I'm going to get started on my 3-year-old Macbook Pro, and expect the first serious training session to take several days once I've settled on some working code.
\n\nUpdate: Trying out ZFTurbo's demo code using Keras without a GPU takes about 3 hours on my laptop. Reducing the image size makes it run even faster. Plenty to look at and learn inside that script, and opportunity to try changing things, tuning etc.
", "votes": 2, "replies": [] }, { "id": 113977, "author_name": "inversion", "author_url": "", "post_date": "2016-04-06T15:16:31.133000", "content": "[quote=the1owl;113975]
\n\nNow we just need one of these
\n\nhttp://www.nvidia.com/object/deep-learning-system.html
\n\n[/quote]
\n\n$129,000. Whoa.
\n\nSeems better to just get a few Titan X GPUs. :-)
", "votes": 2, "replies": [] }, { "id": 113958, "author_name": "OncuKayalar", "author_url": "", "post_date": "2016-04-06T13:13:44.323000", "content": "[quote=inversion;113944]
\n\n+1 for Keras
\n\nAlso, a resource I recently saw that may come in handy:
\n\n"A guide to convolution arithmetic for deep learning"
\n\nhttp://arxiv.org/abs/1603.07285
\n\n[/quote]
\n\nThanks for the article
", "votes": 2, "replies": [] }, { "id": 113953, "author_name": "pasqui", "author_url": "", "post_date": "2016-04-06T13:02:07.243000", "content": "[quote=ChaseWind;113943]
\n\nDeep learning is the nuclear weapon for such computer vision problem :-) Unlike other competitions on Kaggle, there is no way to apply typical feature engineering in this competition.
\n\nOn the condition that you are familiar with Python, Keras (http://keras.io/) will be a wise choice.
\n\n[/quote]
\n\nHi, \n[I am also a novice; no libs war]\ncan I ask you why you suggested Keras and not Caffe?
\n\nbest,
", "votes": 2, "replies": [] }, { "id": 124793, "author_name": "MohammadAbdullah", "author_url": "", "post_date": "2016-06-22T12:25:28.700000", "content": "[quote=Yassine Alouini;114384]
\n\nAs mentioned by other participants, ZFTurbo code is a great place to start for this challenge.
\n\nI am playing with the code right now, doing some tweaking and hopefully improving it.
\n\nIf I find something interesting, I will post it. Stay tuned!
\n\nYou can find the original code here: https://github.com/ZFTurbo/KAGGLE_DISTRACTED_DRIVER
\n\n[/quote]
\n\nWe need R copy for the Code , How we can translate it to R?
", "votes": -2, "replies": [] }, { "id": 124801, "author_name": "", "author_url": "", "post_date": "2016-06-22T13:16:46.267000", "content": "I am thinking of getting that little NUC :) Do you have something like Razer Core for dedicated GPU with NUC?
\n\n[quote=the1owl;115077]
\n\nNow I want one of these...
\n\nhttp://www.intel.com/content/www/us/en/processors/xeon/xeon-phi-detail.html
\n\nBut I guess I have to settle for this one, my current little NUC did ok these past few competitions.
\n\nhttp://www.intel.com/content/www/us/en/nuc/change-the-game-with-nuc.html
\n\n[/quote]
", "votes": 0, "replies": [] }, { "id": 124794, "author_name": "MohammadAbdullah", "author_url": "", "post_date": "2016-06-22T12:26:28.130000", "content": "[quote=NaomiFridman;114588]
\n\nHi,
\n\nDisclaimer: I am total newbie in Data Science and Kaggle. I am a mathematician, I did some courses in Machine learning, but experienced only on small clean data on matlab and Python.
\n\nMy first challenge is to set up working environment. \nDo you think setting it up on windows is a good Idea ? I found this article : \nhttps://datanoord.com/2016/02/01/setup-a-deep-learning-environment-on-windows-theano-keras-with-gpu-enabled/
\n\nI installed Ubuntu on Oracle virtual machine, so I can try setting up working environment there as well.
\n\nThanks and regards,
\n\nNaomi Fridman
\n\n[/quote]\n Page cannot be found !!!
", "votes": 0, "replies": [] }, { "id": 123212, "author_name": "MohammadAbdullah", "author_url": "", "post_date": "2016-06-10T11:23:07.433000", "content": "Hi Every One
\n\nThe tutorial of Facial Keypoint Detection : https://www.kaggle.com/c/facial-keypoints-detection/details/getting-started-with-r
\n\nStill need more other tutorials , I have question based on R is my favorite language to start coding:
\n\n1-Is there any other sources do step by step tutorial using R for Image processing and extraction (recognition)?
\n\n2- Can you use Matlab Instead for creating the algorithm ?
\n\n3-Is there a learning sources to combine R with Matlab in coding our project (StateFarm)?
\n\nThank you in Advanced
", "votes": 0, "replies": [] }, { "id": 114843, "author_name": "Sparrow", "author_url": "", "post_date": "2016-04-14T07:35:06.917000", "content": "", "votes": 0, "replies": [] }, { "id": 114637, "author_name": "gauss256", "author_url": "", "post_date": "2016-04-12T16:56:52.410000", "content": "@NaomiFridman Linux is much better supported for deep learning than Windows. You probably won't be happy running Ubuntu in a VM, except for initial explorations. It's too slow and doesn't support a GPU. If you need to keep Windows you could consider a dual-boot setup.
\n\nI'm using a dedicated Linux machine (Ubuntu 14.04) on which I run
\n\n[quote=NaomiFridman;114588]
\n\nHi,
\n\nDisclaimer: I am total newbie in Data Science and Kaggle. I am a mathematician, I did some courses in Machine learning, but experienced only on small clean data on matlab and Python.
\n\nMy first challenge is to set up working environment. \nDo you think setting it up on windows is a good Idea ? I found this article : \nhttps://datanoord.com/2016/02/01/setup-a-deep-learning-environment-on-windows-theano-keras-with-gpu-enabled/
\n\nI installed Ubuntu on Oracle virtual machine, so I can try setting up working environment there as well.
\n\nThanks and regards,
\n\nNaomi Fridman
\n\n[/quote]
\n\nHi, \nusually, for Computer Vision tasks I would like to use all the computational power available; thus, if we are talking about desktops/laptops, I would not go for a VM,
\n\nbest,
", "votes": 0, "replies": [] }, { "id": 114588, "author_name": "NaomiFridman", "author_url": "", "post_date": "2016-04-12T09:04:04.913000", "content": "Hi,
\n\nDisclaimer: I am total newbie in Data Science and Kaggle. I am a mathematician, I did some courses in Machine learning, but experienced only on small clean data on matlab and Python.
\n\nMy first challenge is to set up working environment. \nDo you think setting it up on windows is a good Idea ? I found this article : \nhttps://datanoord.com/2016/02/01/setup-a-deep-learning-environment-on-windows-theano-keras-with-gpu-enabled/
\n\nI installed Ubuntu on Oracle virtual machine, so I can try setting up working environment there as well.
\n\nThanks and regards,
\n\nNaomi Fridman
", "votes": 0, "replies": [] }, { "id": 114573, "author_name": "narsil (jobs-in-data.com)", "author_url": "", "post_date": "2016-04-12T02:29:03.953000", "content": "For someone who wants to do it in R, is there an obvious way to go (like Keras for Python)?\nOr you suggest that I forget R, and learn python to do it?
", "votes": 0, "replies": [] }, { "id": 114444, "author_name": "Mustyy", "author_url": "", "post_date": "2016-04-10T22:56:47.767000", "content": "Hey Yassine\nHope all is well\nIt is a great place to start, however which file would you tackle first.\nThere are 3 posted thus far. Thanks
", "votes": 0, "replies": [] }, { "id": 114384, "author_name": "Yassine Alouini", "author_url": "", "post_date": "2016-04-10T08:21:31.530000", "content": "As mentioned by other participants, ZFTurbo code is a great place to start for this challenge.
\n\nI am playing with the code right now, doing some tweaking and hopefully improving it.
\n\nIf I find something interesting, I will post it. Stay tuned!
\n\nYou can find the original code here: https://github.com/ZFTurbo/KAGGLE_DISTRACTED_DRIVER
", "votes": 0, "replies": [] }, { "id": 114257, "author_name": "TheFOX", "author_url": "", "post_date": "2016-04-08T16:28:05.010000", "content": "Thanks Inversion, ChaseWind for the references. I will try to learn Python, Keras over the next 3 month period. Atleast I will get started. \nThanks for the references, will go over them. BTW I have successfully installed all the components CUDA, VC, Anaconda Theano and Keras. Time to start studying !!!
\n\nThanks and regards\nSayantan Raha
", "votes": 0, "replies": [] }, { "id": 114120, "author_name": "ChaseWind", "author_url": "", "post_date": "2016-04-07T16:44:04.497000", "content": "[quote=BrettAmdur;114092]
\n\nCan anyone take a ballpark stab at what the AWS fees would be for someone to just have fun and learn with this competition, as opposed to someone who needs to train enough data to be one of the leaders?
\n\n[/quote]
\n\nIt is possible to have a spot instance of "g2.2xlarge" at the price $0.3/h. Please keep in mind that the price is not stable at all. The execution time of the script could be less than 1 hour.
", "votes": 0, "replies": [] }, { "id": 114092, "author_name": "staggerlee", "author_url": "", "post_date": "2016-04-07T14:46:34.057000", "content": "Can anyone take a ballpark stab at what the AWS fees would be for someone to just have fun and learn with this competition, as opposed to someone who needs to train enough data to be one of the leaders?
", "votes": 0, "replies": [] }, { "id": 114044, "author_name": "pasqui", "author_url": "", "post_date": "2016-04-07T07:46:32.397000", "content": "[quote=Neil Slater;114043]
\n\n[quote=pasqui;114001]
\n\n\n\n\nBut I'm going to get started on my 3-year-old Macbook Pro
\n
It should be CUDA-capable
\n\n[/quote]
\n\nNot in any meaningful way. I checked. Integrated graphics board: Intel HD Graphics 4000 1024 MB. See http://apple.stackexchange.com/questions/93354/is-my-mac-book-pro-suitable-for-cuda
\n\nSome of the top-end models had graphics cards where it might be possible.
\n\n[/quote]
\n\nI am sorry. 3 years means that only the 15" in its top conf has a a Nvidia card.
", "votes": 0, "replies": [] }, { "id": 114043, "author_name": "Neil Slater", "author_url": "", "post_date": "2016-04-07T07:30:19.007000", "content": "[quote=pasqui;114001]
\n\n\n\n\nBut I'm going to get started on my 3-year-old Macbook Pro
\n
It should be CUDA-capable
\n\n[/quote]
\n\nNot in any meaningful way. I checked. Integrated graphics board: Intel HD Graphics 4000 1024 MB. See http://apple.stackexchange.com/questions/93354/is-my-mac-book-pro-suitable-for-cuda
\n\nSome of the top-end models had graphics cards where it might be possible.
", "votes": 0, "replies": [] }, { "id": 114001, "author_name": "pasqui", "author_url": "", "post_date": "2016-04-06T18:38:35.270000", "content": "\n\n\nBut I'm going to get started on my 3-year-old Macbook Pro
\n
It should be CUDA-capable
", "votes": 0, "replies": [] }, { "id": 115077, "author_name": "", "author_url": "", "post_date": "2016-04-16T03:50:40.657000", "content": "", "votes": 0, "replies": [] }, { "id": 113999, "author_name": "", "author_url": "", "post_date": "2016-04-06T18:29:28.683000", "content": "", "votes": 0, "replies": [] }, { "id": 113991, "author_name": "", "author_url": "", "post_date": "2016-04-06T18:13:46.027000", "content": "", "votes": 0, "replies": [] }, { "id": 113975, "author_name": "", "author_url": "", "post_date": "2016-04-06T14:46:47.887000", "content": "", "votes": 1, "replies": [] } ], "raw_markdown_by_id": { "113944": "+1 for Keras\r\n\r\nAlso, a resource I recently saw that may come in handy:\r\n\r\n\"A guide to convolution arithmetic for deep learning\"\r\n\r\nhttp://arxiv.org/abs/1603.07285", "114523": "Hi Sayantan,\r\n\r\nPlease check Stanford's course on Convolution Networks. Possibly the best place to learn convolution networks and it's applications in Machine Vision. However, this course assume basics in Machine Learning. \r\n\r\nhttp://cs231n.stanford.edu/", "113926": "Hi,\r\n\r\nDisclaimer: I am new to Data Science and Kaggle. hence my queries can sound naive. Kindly pardon me for the same.\r\n\r\nI am reaching out to the masters to point to some reference, some reading, which can help a novice like me understand how to deal with problems like this, where the source data is a bunch of images. \r\n\r\nWhat are the ways in which we create datasets from images? What sort of feature engineering do we do in these cases?\r\n\r\nThanks and regards\r\nSayantan Raha", "113943": "Deep learning is the nuclear weapon for such computer vision problem :-) Unlike other competitions on Kaggle, there is no way to apply typical feature engineering in this competition.\r\n\r\nOn the condition that you are familiar with Python, Keras (http://keras.io/) will be a wise choice.", "113960": "[quote=pasqui;113953]\r\n\r\n[quote=ChaseWind;113943]\r\n\r\nDeep learning is the nuclear weapon for such computer vision problem :-) Unlike other competitions on Kaggle, there is no way to apply typical feature engineering in this competition.\r\n\r\nOn the condition that you are familiar with Python, Keras (http://keras.io/) will be a wise choice.\r\n\r\n[/quote]\r\n\r\nHi, \r\n[I am also a novice; no libs war]\r\ncan I ask you why you suggested Keras and not Caffe?\r\n\r\nbest,\r\n\r\n[/quote]\r\n\r\n1, Installation of Caffe is much more complicated than Keras.\r\n\r\n2, Keras is a high level package backended by Theano and TensorFlow. The syntax of Keras is simpler.\r\n\r\n3, Keras supports Caffe models.\r\n\r\n4, Personally, I have ever encountered many nasty defects with Caffe, such as compatibility problems.", "114597": "@narsil\r\nThe deep learning libraries available for R are far less developed and far less supported than for Python, which has numerous great tools (Keras, Lasagne, TensorFlow, etc..). \r\n\r\nMXNet is now available, and seems to have a lot of potential, but I've had lots of trouble with it (random crashes...), and the documentation is very sparse for the the moment.\r\n\r\nSo yeah, if you want to use some deep neural networks, Python is probably the way to go.", "114163": "g2.2xlarge AWS spot instances are no longer an option: the price instability is way too high since some months ago, and they are usually more expensive (sometimes 10x more) than standard instances. ", "113994": "[quote=RohinArora;113991]\r\n\r\n[quote=inversion;113977]\r\n\r\n[quote=the1owl;113975]\r\n\r\nNow we just need one of these\r\n\r\n[http://www.nvidia.com/object/deep-learning-system.html][1]\r\n\r\n\r\n[/quote]\r\n\r\n$129,000. Whoa.\r\n\r\nSeems better to just get a few Titan X GPUs. :-)\r\n\r\n\r\n[/quote]\r\n\r\n\r\nHi. Another novice here. Is it possible to take part in this competition without a Nvidia GPU? My laptop has Intel i5 3rd gen and AMD graphic card. My guess is no. I wanted your opinion. \r\n\r\n[/quote]\r\n\r\n\r\nYou can use a cloud GPU instance. [**AWS**][2] offers such instances (it is the **G2** family). \r\n\r\nYou can find more documentation here: https://aws.amazon.com/ec2/instance-types/. \r\n\r\nLet me know if you need more help!\r\n\r\n\r\n [1]: http://www.nvidia.com/object/deep-learning-system.html\r\n [2]: https://aws.amazon.com/ec2/instance-types/", "114141": "$0.3/hour sounds pretty cheap but I found that I very quickly ran up a bill of over $100 just get install everything and run a few sample networks. In the end I decided that I should just put together a desktop Linux box with a decent Nvidia card in it. I built one around a 980 Ti and it runs 2.5x faster than the AWS instance. \r\n\r\nBut it wasn't cheap. Probably close to $2,000 for everything.\r\n\r\nIf you do go the AWS route, note that you are charged for the instance just being up even if it's idle.", "114000": "[quote=RohinArora;113991]\r\n\r\n[quote=inversion;113977]\r\n\r\n[quote=the1owl;113975]\r\n\r\nNow we just need one of these\r\n\r\n[http://www.nvidia.com/object/deep-learning-system.html][1]\r\n\r\n\r\n [1]: http://www.nvidia.com/object/deep-learning-system.html\r\n\r\n[/quote]\r\n\r\n$129,000. Whoa.\r\n\r\nSeems better to just get a few Titan X GPUs. :-)\r\n\r\n\r\n[/quote]\r\n\r\n\r\nHi. Another novice here. Is it possible to take part in this competition without a Nvidia GPU? My laptop has Intel i5 3rd gen and AMD graphic card. My guess is no. I wanted your opinion. \r\n\r\n[/quote]\r\n\r\nIt should be possible to take part, but training times will be too long to do much in the way of tuning. To get started, you will need to limit the scope of what you attempt - such as reducing the size of the images and trying out architectures on just 10% of the data. \r\n\r\nI don't have any GPU-based hardware, but I still intend to play along, and pick up a little experience.\r\n\r\nIf I think I will do well enough, I *might* try AWS this time for the experience of that, too. But I'm going to get started on my 3-year-old Macbook Pro, and expect the first serious training session to take several days once I've settled on some working code.\r\n\r\n----\r\n\r\nUpdate: Trying out *ZFTurbo*'s demo code using Keras without a GPU takes about 3 hours on my laptop. Reducing the image size makes it run even faster. Plenty to look at and learn inside that script, and opportunity to try changing things, tuning etc.\r\n\r\n\r\n", "113977": "[quote=the1owl;113975]\r\n\r\nNow we just need one of these\r\n\r\n[http://www.nvidia.com/object/deep-learning-system.html][1]\r\n\r\n\r\n [1]: http://www.nvidia.com/object/deep-learning-system.html\r\n\r\n[/quote]\r\n\r\n$129,000. Whoa.\r\n\r\nSeems better to just get a few Titan X GPUs. :-)\r\n", "113958": "[quote=inversion;113944]\r\n\r\n+1 for Keras\r\n\r\nAlso, a resource I recently saw that may come in handy:\r\n\r\n\"A guide to convolution arithmetic for deep learning\"\r\n\r\nhttp://arxiv.org/abs/1603.07285\r\n\r\n[/quote]\r\n\r\nThanks for the article", "113953": "[quote=ChaseWind;113943]\r\n\r\nDeep learning is the nuclear weapon for such computer vision problem :-) Unlike other competitions on Kaggle, there is no way to apply typical feature engineering in this competition.\r\n\r\nOn the condition that you are familiar with Python, Keras (http://keras.io/) will be a wise choice.\r\n\r\n[/quote]\r\n\r\nHi, \r\n[I am also a novice; no libs war]\r\ncan I ask you why you suggested Keras and not Caffe?\r\n\r\nbest,", "124793": "[quote=Yassine Alouini;114384]\r\n\r\nAs mentioned by other participants, **ZFTurbo** code is a great place to start for this challenge. \r\n\r\nI am playing with the code right now, doing some tweaking and hopefully improving it. \r\n\r\nIf I find something interesting, I will post it. Stay tuned!\r\n\r\nYou can find the original code here: https://github.com/ZFTurbo/KAGGLE_DISTRACTED_DRIVER\r\n\r\n[/quote]\r\n\r\nWe need R copy for the Code , How we can translate it to R?\r\n", "124801": "I am thinking of getting that little NUC :) Do you have something like Razer Core for dedicated GPU with NUC? \r\n\r\n[quote=the1owl;115077]\r\n\r\nNow I want one of these...\r\n\r\nhttp://www.intel.com/content/www/us/en/processors/xeon/xeon-phi-detail.html\r\n\r\nBut I guess I have to settle for this one, my current little NUC did ok these past few competitions.\r\n\r\nhttp://www.intel.com/content/www/us/en/nuc/change-the-game-with-nuc.html\r\n\r\n[/quote]\r\n", "124794": "[quote=NaomiFridman;114588]\r\n\r\nHi,\r\n\r\nDisclaimer: I am total newbie in Data Science and Kaggle. I am a mathematician, I did some courses in Machine learning, but experienced only on small clean data on matlab and Python. \r\n\r\nMy first challenge is to set up working environment. \r\nDo you think setting it up on windows is a good Idea ? I found this article : \r\nhttps://datanoord.com/2016/02/01/setup-a-deep-learning-environment-on-windows-theano-keras-with-gpu-enabled/\r\n\r\nI installed Ubuntu on Oracle virtual machine, so I can try setting up working environment there as well.\r\n\r\nThanks and regards,\r\n\r\nNaomi Fridman\r\n\r\n\r\n[/quote]\r\n Page cannot be found !!!", "123212": "Hi Every One \r\n\r\nThe tutorial of Facial Keypoint Detection : https://www.kaggle.com/c/facial-keypoints-detection/details/getting-started-with-r\r\n\r\nStill need more other tutorials , I have question based on R is my favorite language to start coding:\r\n\r\n1-Is there any other sources do step by step tutorial using R for Image processing and extraction (recognition)?\r\n\r\n2- Can you use Matlab Instead for creating the algorithm ?\r\n\r\n3-Is there a learning sources to combine R with Matlab in coding our project (StateFarm)?\r\n\r\nThank you in Advanced ", "114843": "", "114637": "@NaomiFridman Linux is much better supported for deep learning than Windows. You probably won't be happy running Ubuntu in a VM, except for initial explorations. It's too slow and doesn't support a GPU. If you need to keep Windows you could consider a dual-boot setup.\r\n\r\nI'm using a dedicated Linux machine (Ubuntu 14.04) on which I run\r\n\r\n - Python (Anaconda distribution)\r\n - TensorFlow\r\n - Keras\r\n - Nvidia 980 Ti GPU", "114590": "[quote=NaomiFridman;114588]\r\n\r\nHi,\r\n\r\nDisclaimer: I am total newbie in Data Science and Kaggle. I am a mathematician, I did some courses in Machine learning, but experienced only on small clean data on matlab and Python. \r\n\r\nMy first challenge is to set up working environment. \r\nDo you think setting it up on windows is a good Idea ? I found this article : \r\nhttps://datanoord.com/2016/02/01/setup-a-deep-learning-environment-on-windows-theano-keras-with-gpu-enabled/\r\n\r\nI installed Ubuntu on Oracle virtual machine, so I can try setting up working environment there as well.\r\n\r\nThanks and regards,\r\n\r\nNaomi Fridman\r\n\r\n\r\n[/quote]\r\n\r\nHi, \r\nusually, for Computer Vision tasks I would like to use all the computational power available; thus, if we are talking about desktops/laptops, I would not go for a VM, \r\n\r\nbest,", "114588": "Hi,\r\n\r\nDisclaimer: I am total newbie in Data Science and Kaggle. I am a mathematician, I did some courses in Machine learning, but experienced only on small clean data on matlab and Python. \r\n\r\nMy first challenge is to set up working environment. \r\nDo you think setting it up on windows is a good Idea ? I found this article : \r\nhttps://datanoord.com/2016/02/01/setup-a-deep-learning-environment-on-windows-theano-keras-with-gpu-enabled/\r\n\r\nI installed Ubuntu on Oracle virtual machine, so I can try setting up working environment there as well.\r\n\r\nThanks and regards,\r\n\r\nNaomi Fridman\r\n", "114573": "For someone who wants to do it in R, is there an obvious way to go (like Keras for Python)?\r\nOr you suggest that I forget R, and learn python to do it?\r\n\r\n", "114444": "Hey Yassine\r\nHope all is well\r\nIt is a great place to start, however which file would you tackle first.\r\nThere are 3 posted thus far. Thanks", "114384": "As mentioned by other participants, **ZFTurbo** code is a great place to start for this challenge. \r\n\r\nI am playing with the code right now, doing some tweaking and hopefully improving it. \r\n\r\nIf I find something interesting, I will post it. Stay tuned!\r\n\r\nYou can find the original code here: https://github.com/ZFTurbo/KAGGLE_DISTRACTED_DRIVER", "114257": "Thanks Inversion, ChaseWind for the references. I will try to learn Python, Keras over the next 3 month period. Atleast I will get started. \r\nThanks for the references, will go over them. BTW I have successfully installed all the components CUDA, VC, Anaconda Theano and Keras. Time to start studying !!!\r\n\r\nThanks and regards\r\nSayantan Raha", "114120": "[quote=BrettAmdur;114092]\r\n\r\nCan anyone take a ballpark stab at what the AWS fees would be for someone to just have fun and learn with this competition, as opposed to someone who needs to train enough data to be one of the leaders? \r\n\r\n[/quote]\r\n\r\nIt is possible to have a spot instance of \"g2.2xlarge\" at the price $0.3/h. Please keep in mind that the price is not stable at all. The execution time of the script could be less than 1 hour.", "114092": "Can anyone take a ballpark stab at what the AWS fees would be for someone to just have fun and learn with this competition, as opposed to someone who needs to train enough data to be one of the leaders? ", "114044": "[quote=Neil Slater;114043]\r\n\r\n[quote=pasqui;114001]\r\n\r\n> But I'm going to get started on my 3-year-old Macbook Pro\r\n\r\nIt should be CUDA-capable\r\n\r\n[/quote]\r\n\r\nNot in any meaningful way. I checked. Integrated graphics board: Intel HD Graphics 4000 1024 MB. See http://apple.stackexchange.com/questions/93354/is-my-mac-book-pro-suitable-for-cuda\r\n\r\nSome of the top-end models had graphics cards where it might be possible.\r\n\r\n\r\n[/quote]\r\n\r\nI am sorry. 3 years means that only the 15\" in its top conf has a a Nvidia card.", "114043": "[quote=pasqui;114001]\r\n\r\n> But I'm going to get started on my 3-year-old Macbook Pro\r\n\r\nIt should be CUDA-capable\r\n\r\n[/quote]\r\n\r\nNot in any meaningful way. I checked. Integrated graphics board: Intel HD Graphics 4000 1024 MB. See http://apple.stackexchange.com/questions/93354/is-my-mac-book-pro-suitable-for-cuda\r\n\r\nSome of the top-end models had graphics cards where it might be possible.\r\n", "114001": "> But I'm going to get started on my 3-year-old Macbook Pro\r\n\r\nIt should be CUDA-capable", "115077": "", "113999": "", "113991": "", "113975": "" } }