I am a student interested in competing in Kaggle competitions and at least winning a bronze medal. It is not that I ain't ambitious but to even reach out for bronze medals in such competitions would require me to invest a lot of money to train models. Are there resources available or is it something that competitors have to just invest in it?
", "messages": [ { "id": "269713", "postDate": "01/17/2018 06:41:31", "content": "I am a student interested in competing in Kaggle competitions and at least winning a bronze medal. It is not that I ain't ambitious but to even reach out for bronze medals in such competitions would require me to invest a lot of money to train models. Are there resources available or is it something that competitors have to just invest in it?
", "rawMarkdown": "I am a student interested in competing in Kaggle competitions and at least winning a bronze medal. It is not that I ain't ambitious but to even reach out for bronze medals in such competitions would require me to invest a lot of money to train models. Are there resources available or is it something that competitors have to just invest in it?", "votes": null }, { "id": "269903", "postDate": "01/17/2018 14:03:22", "content": "I am new too and am interested in participating. I don't understand what you mean by investing money to train models? Is it specific to this competition? Or are you talking in general
", "rawMarkdown": "I am new too and am interested in participating. I don't understand what you mean by investing money to train models? Is it specific to this competition? Or are you talking in general", "votes": null }, { "id": "269907", "postDate": "01/17/2018 14:06:39", "content": "I think unless you have access to a decent GPU in this competition is going to be hard to get a bronze medal. I think there will be 1000+ teams.
", "rawMarkdown": "I think unless you have access to a decent GPU in this competition is going to be hard to get a bronze medal. I think there will be 1000+ teams.", "votes": null }, { "id": "269960", "postDate": "01/17/2018 15:25:16", "content": "Hey, I'm a student too. I think it won't be that big of an issue, just be sure to ration your resources wisely. You can get about $300 Google Cloud platform credits and $225 AWS credits( $150 free +75 starter) after that floydhub has about 20 CPU hours free/month iirc
\n\nThat being said, this involves switching between a lot of platforms and can potentially be quite tiring. Like what Andres said it's much better to have a decent GPU. What I'm currently doing is using my University's resources to train the models. Perhaps you could explore something similar
", "rawMarkdown": "Hey, I'm a student too. I think it won't be that big of an issue, just be sure to ration your resources wisely. You can get about $300 Google Cloud platform credits and $225 AWS credits( $150 free +75 starter) after that floydhub has about 20 CPU hours free/month iirc\n\nThat being said, this involves switching between a lot of platforms and can potentially be quite tiring. Like what Andres said it's much better to have a decent GPU. What I'm currently doing is using my University's resources to train the models. Perhaps you could explore something similar", "votes": null }, { "id": "269961", "postDate": "01/17/2018 15:26:56", "content": "I think he means that you'll need to spend money to pay for GPU instances to train the convolutional neural network models which are likely to play a key role in this task. For other kaggle competitions there are instances where people don't use Neural networks and instead use conventional Machine Learning techniques but nowadays it's quite rare.
\n\nIf cost is an issue, you can consider exploring some of the options that I've listed below to get free credits.
", "rawMarkdown": "I think he means that you'll need to spend money to pay for GPU instances to train the convolutional neural network models which are likely to play a key role in this task. For other kaggle competitions there are instances where people don't use Neural networks and instead use conventional Machine Learning techniques but nowadays it's quite rare. \n\nIf cost is an issue, you can consider exploring some of the options that I've listed below to get free credits.", "votes": null }, { "id": "270098", "postDate": "01/17/2018 18:48:54", "content": "Thank you for the suggestion, Andres. I am looking to buy Dell Workstation with NVIDIA GPUs. Would you suggest any specific laptops (not gaming laptops) or one that you prefer using for competitions?
", "rawMarkdown": "Thank you for the suggestion, Andres. I am looking to buy Dell Workstation with NVIDIA GPUs. Would you suggest any specific laptops (not gaming laptops) or one that you prefer using for competitions?", "votes": null }, { "id": "270100", "postDate": "01/17/2018 18:51:01", "content": "I doubt using University's resources will be ethical. I would rather go with AWS, GCP, and Azure (if they provide free credits). Even then, with the limited credits, one can only compete in a competition or two.
", "rawMarkdown": "I doubt using University's resources will be ethical. I would rather go with AWS, GCP, and Azure (if they provide free credits). Even then, with the limited credits, one can only compete in a competition or two.", "votes": null }, { "id": "270101", "postDate": "01/17/2018 18:51:59", "content": "Single most important thing is the GPU. Get an Nvidia card (AMD has lousy support for DL) with 8 Gbytes of RAM or more so you can have decent batch size. A 1070 Ti, 1080, or 1080 Ti are pretty solid choices (although expensive). Re: CPU rule of thumb is to get number of cores = 2 x number of GPUs + 1.
", "rawMarkdown": "Single most important thing is the GPU. Get an Nvidia card (AMD has lousy support for DL) with 8 Gbytes of RAM or more so you can have decent batch size. A 1070 Ti, 1080, or 1080 Ti are pretty solid choices (although expensive). Re: CPU rule of thumb is to get number of cores = 2 x number of GPUs + 1.", "votes": null }, { "id": "270218", "postDate": "01/17/2018 23:17:14", "content": "I am new to ML and don't have a computer with GPU, May I know how can I get AWS/Google credits?
", "rawMarkdown": "I am new to ML and don't have a computer with GPU, May I know how can I get AWS/Google credits?", "votes": null }, { "id": "270232", "postDate": "01/17/2018 23:49:52", "content": "Google Cloud: https://cloud.google.com/free/
\n\nAWS:https://aws.amazon.com/awscredits/
", "rawMarkdown": "Google Cloud: https://cloud.google.com/free/\n\nAWS:https://aws.amazon.com/awscredits/", "votes": null }, { "id": "270239", "postDate": "01/18/2018 00:11:24", "content": "On the contrary, I think it’s perfectly ethical. University researchers typically don’t use their GPUs 24/7. Resources (GPUs) will be lying idle if unused. Kaggle competitions aim to provide a platform for learning which is what the university wishes to support too. So if you able to use the downtime effectively you are converting a wasted resource into an aid for learning. You won't be depriving other people of resources. You can even rope in your classmates so it's more cost-effective
\n\nIf it’s an issue of having an unfair advantage, I do believe many teams here have better rigs than most university researchers.
\n\nHowever, I do acknowledge that I ‘m privileged to come from a well-funded school with excess resources. Not all universities may be in this situation. AWS and GCP should suffice for a competition or two— beyond that, it would be a good idea to save up for a GPU. Personally, I’m working part-time in order to buy a GPU and have explored the option of sharing one with three other people.
\n\nShould these methods not suffice, feel free to let me know if there's any way I can help
", "rawMarkdown": "On the contrary, I think it’s perfectly ethical. University researchers typically don’t use their GPUs 24/7. Resources (GPUs) will be lying idle if unused. Kaggle competitions aim to provide a platform for learning which is what the university wishes to support too. So if you able to use the downtime effectively you are converting a wasted resource into an aid for learning. You won't be depriving other people of resources. You can even rope in your classmates so it's more cost-effective\n\nIf it’s an issue of having an unfair advantage, I do believe many teams here have better rigs than most university researchers.\n\nHowever, I do acknowledge that I ‘m privileged to come from a well-funded school with excess resources. Not all universities may be in this situation. AWS and GCP should suffice for a competition or two— beyond that, it would be a good idea to save up for a GPU. Personally, I’m working part-time in order to buy a GPU and have explored the option of sharing one with three other people.\n\nShould these methods not suffice, feel free to let me know if there's any way I can help", "votes": null }, { "id": "274100", "postDate": "01/25/2018 21:53:43", "content": "After a pytorch program ran over 10 hrs on my ancient laptop, I am now looking for free computing (GPU) resources. OnePanel is promising and helpful and they are resolving one issue for me, another one I just found is Floydhub. Anyone has any experience with the? I tried Colaboratory also but can't figure out how to put 100+ images there to process (and potential some processing image like mask need to go somewhere).
", "rawMarkdown": "After a pytorch program ran over 10 hrs on my ancient laptop, I am now looking for free computing (GPU) resources. OnePanel is promising and helpful and they are resolving one issue for me, another one I just found is Floydhub. Anyone has any experience with the? I tried Colaboratory also but can't figure out how to put 100+ images there to process (and potential some processing image like mask need to go somewhere).", "votes": null }, { "id": "274129", "postDate": "01/25/2018 23:16:58", "content": "Student here as well. You could simply avoid image based comps since those require more processing power. But you could look into buying a used GPU, even the cheap ones will get you far.
\n\nAlso you can definitely medal just using kaggle kernels in some competitions.
", "rawMarkdown": "Student here as well. You could simply avoid image based comps since those require more processing power. But you could look into buying a used GPU, even the cheap ones will get you far.\n\nAlso you can definitely medal just using kaggle kernels in some competitions.", "votes": null } ], "comments": [ { "id": 269903, "author_name": "tonmoyj", "author_url": "", "post_date": "01/17/2018 14:03:22", "content": "I am new too and am interested in participating. I don't understand what you mean by investing money to train models? Is it specific to this competition? Or are you talking in general
", "votes": null, "replies": [ { "id": 269961, "author_name": "jzerox2", "author_url": "", "post_date": "01/17/2018 15:26:56", "content": "I think he means that you'll need to spend money to pay for GPU instances to train the convolutional neural network models which are likely to play a key role in this task. For other kaggle competitions there are instances where people don't use Neural networks and instead use conventional Machine Learning techniques but nowadays it's quite rare.
\n\nIf cost is an issue, you can consider exploring some of the options that I've listed below to get free credits.
", "votes": null, "replies": [] } ] }, { "id": 269907, "author_name": "antorsae", "author_url": "", "post_date": "01/17/2018 14:06:39", "content": "I think unless you have access to a decent GPU in this competition is going to be hard to get a bronze medal. I think there will be 1000+ teams.
", "votes": null, "replies": [ { "id": 270098, "author_name": "gosuddin", "author_url": "", "post_date": "01/17/2018 18:48:54", "content": "Thank you for the suggestion, Andres. I am looking to buy Dell Workstation with NVIDIA GPUs. Would you suggest any specific laptops (not gaming laptops) or one that you prefer using for competitions?
", "votes": null, "replies": [] }, { "id": 270101, "author_name": "antorsae", "author_url": "", "post_date": "01/17/2018 18:51:59", "content": "Single most important thing is the GPU. Get an Nvidia card (AMD has lousy support for DL) with 8 Gbytes of RAM or more so you can have decent batch size. A 1070 Ti, 1080, or 1080 Ti are pretty solid choices (although expensive). Re: CPU rule of thumb is to get number of cores = 2 x number of GPUs + 1.
", "votes": null, "replies": [] } ] }, { "id": 269960, "author_name": "jzerox2", "author_url": "", "post_date": "01/17/2018 15:25:16", "content": "Hey, I'm a student too. I think it won't be that big of an issue, just be sure to ration your resources wisely. You can get about $300 Google Cloud platform credits and $225 AWS credits( $150 free +75 starter) after that floydhub has about 20 CPU hours free/month iirc
\n\nThat being said, this involves switching between a lot of platforms and can potentially be quite tiring. Like what Andres said it's much better to have a decent GPU. What I'm currently doing is using my University's resources to train the models. Perhaps you could explore something similar
", "votes": null, "replies": [ { "id": 270100, "author_name": "gosuddin", "author_url": "", "post_date": "01/17/2018 18:51:01", "content": "I doubt using University's resources will be ethical. I would rather go with AWS, GCP, and Azure (if they provide free credits). Even then, with the limited credits, one can only compete in a competition or two.
", "votes": null, "replies": [] }, { "id": 270218, "author_name": "xsw7910", "author_url": "", "post_date": "01/17/2018 23:17:14", "content": "I am new to ML and don't have a computer with GPU, May I know how can I get AWS/Google credits?
", "votes": null, "replies": [] }, { "id": 270232, "author_name": "jzerox2", "author_url": "", "post_date": "01/17/2018 23:49:52", "content": "Google Cloud: https://cloud.google.com/free/
\n\nAWS:https://aws.amazon.com/awscredits/
", "votes": null, "replies": [] }, { "id": 270239, "author_name": "jzerox2", "author_url": "", "post_date": "01/18/2018 00:11:24", "content": "On the contrary, I think it’s perfectly ethical. University researchers typically don’t use their GPUs 24/7. Resources (GPUs) will be lying idle if unused. Kaggle competitions aim to provide a platform for learning which is what the university wishes to support too. So if you able to use the downtime effectively you are converting a wasted resource into an aid for learning. You won't be depriving other people of resources. You can even rope in your classmates so it's more cost-effective
\n\nIf it’s an issue of having an unfair advantage, I do believe many teams here have better rigs than most university researchers.
\n\nHowever, I do acknowledge that I ‘m privileged to come from a well-funded school with excess resources. Not all universities may be in this situation. AWS and GCP should suffice for a competition or two— beyond that, it would be a good idea to save up for a GPU. Personally, I’m working part-time in order to buy a GPU and have explored the option of sharing one with three other people.
\n\nShould these methods not suffice, feel free to let me know if there's any way I can help
", "votes": null, "replies": [] } ] }, { "id": 274100, "author_name": "ppirrip", "author_url": "", "post_date": "01/25/2018 21:53:43", "content": "After a pytorch program ran over 10 hrs on my ancient laptop, I am now looking for free computing (GPU) resources. OnePanel is promising and helpful and they are resolving one issue for me, another one I just found is Floydhub. Anyone has any experience with the? I tried Colaboratory also but can't figure out how to put 100+ images there to process (and potential some processing image like mask need to go somewhere).
", "votes": null, "replies": [] }, { "id": 274129, "author_name": "michaelsnell", "author_url": "", "post_date": "01/25/2018 23:16:58", "content": "Student here as well. You could simply avoid image based comps since those require more processing power. But you could look into buying a used GPU, even the cheap ones will get you far.
\n\nAlso you can definitely medal just using kaggle kernels in some competitions.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "269713": "I am a student interested in competing in Kaggle competitions and at least winning a bronze medal. It is not that I ain't ambitious but to even reach out for bronze medals in such competitions would require me to invest a lot of money to train models. Are there resources available or is it something that competitors have to just invest in it?", "269903": "I am new too and am interested in participating. I don't understand what you mean by investing money to train models? Is it specific to this competition? Or are you talking in general", "269907": "I think unless you have access to a decent GPU in this competition is going to be hard to get a bronze medal. I think there will be 1000+ teams.", "269960": "Hey, I'm a student too. I think it won't be that big of an issue, just be sure to ration your resources wisely. You can get about $300 Google Cloud platform credits and $225 AWS credits( $150 free +75 starter) after that floydhub has about 20 CPU hours free/month iirc\n\nThat being said, this involves switching between a lot of platforms and can potentially be quite tiring. Like what Andres said it's much better to have a decent GPU. What I'm currently doing is using my University's resources to train the models. Perhaps you could explore something similar", "269961": "I think he means that you'll need to spend money to pay for GPU instances to train the convolutional neural network models which are likely to play a key role in this task. For other kaggle competitions there are instances where people don't use Neural networks and instead use conventional Machine Learning techniques but nowadays it's quite rare. \n\nIf cost is an issue, you can consider exploring some of the options that I've listed below to get free credits.", "270098": "Thank you for the suggestion, Andres. I am looking to buy Dell Workstation with NVIDIA GPUs. Would you suggest any specific laptops (not gaming laptops) or one that you prefer using for competitions?", "270100": "I doubt using University's resources will be ethical. I would rather go with AWS, GCP, and Azure (if they provide free credits). Even then, with the limited credits, one can only compete in a competition or two.", "270101": "Single most important thing is the GPU. Get an Nvidia card (AMD has lousy support for DL) with 8 Gbytes of RAM or more so you can have decent batch size. A 1070 Ti, 1080, or 1080 Ti are pretty solid choices (although expensive). Re: CPU rule of thumb is to get number of cores = 2 x number of GPUs + 1.", "270218": "I am new to ML and don't have a computer with GPU, May I know how can I get AWS/Google credits?", "270232": "Google Cloud: https://cloud.google.com/free/\n\nAWS:https://aws.amazon.com/awscredits/", "270239": "On the contrary, I think it’s perfectly ethical. University researchers typically don’t use their GPUs 24/7. Resources (GPUs) will be lying idle if unused. Kaggle competitions aim to provide a platform for learning which is what the university wishes to support too. So if you able to use the downtime effectively you are converting a wasted resource into an aid for learning. You won't be depriving other people of resources. You can even rope in your classmates so it's more cost-effective\n\nIf it’s an issue of having an unfair advantage, I do believe many teams here have better rigs than most university researchers.\n\nHowever, I do acknowledge that I ‘m privileged to come from a well-funded school with excess resources. Not all universities may be in this situation. AWS and GCP should suffice for a competition or two— beyond that, it would be a good idea to save up for a GPU. Personally, I’m working part-time in order to buy a GPU and have explored the option of sharing one with three other people.\n\nShould these methods not suffice, feel free to let me know if there's any way I can help", "274100": "After a pytorch program ran over 10 hrs on my ancient laptop, I am now looking for free computing (GPU) resources. OnePanel is promising and helpful and they are resolving one issue for me, another one I just found is Floydhub. Anyone has any experience with the? I tried Colaboratory also but can't figure out how to put 100+ images there to process (and potential some processing image like mask need to go somewhere).", "274129": "Student here as well. You could simply avoid image based comps since those require more processing power. But you could look into buying a used GPU, even the cheap ones will get you far.\n\nAlso you can definitely medal just using kaggle kernels in some competitions." }, "source": "meta" }