Can anyone win this kind of competition using only Kaggle kernel (8GB ram and gpu), I'm still new to competition like this so I wonder if I will blow the ram and GPU.
", "messages": [ { "id": 1371386, "postDate": "2021-07-01T00:40:32.600Z", "content": "Thanks @michael127001 for the mention. I have not taken a look at this competition in particular, so can't comment on how much of an advantage more compute here would be compared to other recent Kaggle competitions. On one hand, this doesn't seem to be an image competition - those are usually by far the most compute intensive. On the other hand, it's still a massive dataset with signal data, so my instinct would be yes, it will require substantial computing resources.
\nIn general, if you are serious about intently competing on Kaggle I would strongly recommend that you invest in your own compute. The \"free\" online resources have come a long way in recent years, but IMHO there is nothing that compares to having a computer at your fingertips.
\nYou should also be clear with yourself about what you want to get out of Kaggle. Many top Kagglers these days are professionals or semi-professionals, so winning Kaggle competitions has become exponentially more difficult than a few years ago. However, if your goal is to learn and get better at various aspects of machine learning (which is what I initially came here for), then just doing your best and improving steadily is well worthwhile endeavor.
\nThat's my 2c.
", "rawMarkdown": "Thanks @michael127001 for the mention. I have not taken a look at this competition in particular, so can't comment on how much of an advantage more compute here would be compared to other recent Kaggle competitions. On one hand, this doesn't seem to be an image competition - those are usually by far the most compute intensive. On the other hand, it's still a massive dataset with signal data, so my instinct would be yes, it will require substantial computing resources. \n\nIn general, if you are serious about intently competing on Kaggle I would strongly recommend that you invest in your own compute. The \"free\" online resources have come a long way in recent years, but IMHO there is nothing that compares to having a computer at your fingertips. \n\nYou should also be clear with yourself about what you want to get out of Kaggle. Many top Kagglers these days are professionals or semi-professionals, so winning Kaggle competitions has become exponentially more difficult than a few years ago. However, if your goal is to learn and get better at various aspects of machine learning (which is what I initially came here for), then just doing your best and improving steadily is well worthwhile endeavor. \n\nThat's my 2c.", "votes": 18, "replies": [ { "id": 1371715, "postDate": "2021-07-01T07:40:04.290Z", "content": "Completely agree with @tunguz,
\nBut I really think having a good compute resource will be a big benefit in this comp. There are almost 500k targets in the training set, and for each data point, there are 3 sources that make the whole training set more than 1 million signals.
Can anyone win this kind of competition using only Kaggle kernel (8GB ram and gpu), I'm still new to competition like this so I wonder if I will blow the ram and GPU.
", "rawMarkdown": "Can anyone win this kind of competition using only Kaggle kernel (8GB ram and gpu), I'm still new to competition like this so I wonder if I will blow the ram and GPU.", "votes": 15 }, { "id": 1401246, "postDate": "2021-07-27T06:30:54.743Z", "content": "I have tried a few products (Kaggle, Colab, Paperspace, Genesis Cloud) and so far I like Genesis Cloud best.
", "rawMarkdown": "I have tried a few products (Kaggle, Colab, Paperspace, Genesis Cloud) and so far I like Genesis Cloud best.", "votes": 1 }, { "id": 1371416, "postDate": "2021-07-01T01:31:59.987Z", "content": "As @tunguz and @michael127001 have told that, we have to invest our own compute.Anybody can suggest best cloud computing technologies to buy to train our model, which one is cost effective.
", "rawMarkdown": "As @tunguz and @michael127001 have told that, we have to invest our own compute.Anybody can suggest best cloud computing technologies to buy to train our model, which one is cost effective.", "votes": 2, "replies": [ { "id": 1406865, "postDate": "2021-08-01T09:10:48.423Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 1401057, "postDate": "2021-07-26T21:37:32.623Z", "content": "\n\nCan anyone win this kind of competition using only Kaggle kernel (8GB ram and gpu)
\n
Lets take it other way round, having loads of computational resource doesn't give the guarantee of winning, I feel it is right strategy at multiple levels which counts. Having good resources helps but not always and at the end of the day it's not only about winning it is also about learning new ways to do things.
", "rawMarkdown": "> Can anyone win this kind of competition using only Kaggle kernel (8GB ram and gpu)\n\nLets take it other way round, having loads of computational resource doesn't give the guarantee of winning, I feel it is right strategy at multiple levels which counts. Having good resources helps but not always and at the end of the day it's not only about winning it is also about learning new ways to do things.", "votes": 1 }, { "id": 1561289, "postDate": "2021-10-27T13:12:31.950Z", "content": "Hey All,
\nThank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey https://forms.gle/QP9L16niPexozyhu5.
\nThank you all,
\nRegards,
\nChris
From Kaggle Grandmaster @tunguz in a recent interview:
\n\n\nFinally, if you are really serious about Kaggling, you will need to invest in your own high-end hardware. Kaggle notebooks and other free online compute resources (such as Colab) can be helpful to get you off the ground, but aside from a handful of competitions these days, most Kaggle competitions require access to a lot of computational resources.
\n
These competitions have a nasty barrier of entry to obtain fast and large computing components. Big data is always a source of contention and much of the drive for research is fitting the largest models on the smallest machines while maximizing accessibility and speed. Google has the tech to cross validate massive networks with distributed training. At least we're able to step our foot in the door with the resources Kaggle gives us 🙁
", "rawMarkdown": "From Kaggle Grandmaster @tunguz in a [recent interview](https://www.businessofbusiness.com/articles/Kaggle-grandmaster-top-10-bojan-tunguz/):\n\n> Finally, if you are really serious about Kaggling, you will need to invest in your own high-end hardware. Kaggle notebooks and other free online compute resources (such as Colab) can be helpful to get you off the ground, but aside from a handful of competitions these days, most Kaggle competitions require access to a lot of computational resources.\n\nThese competitions have a nasty barrier of entry to obtain fast and large computing components. Big data is always a source of contention and much of the drive for research is fitting the largest models on the smallest machines while maximizing accessibility and speed. Google has the tech to cross validate massive networks with distributed training. At least we're able to step our foot in the door with the resources Kaggle gives us 🙁", "votes": 4, "isDeleted": true, "replies": [ { "id": 1371309, "postDate": "2021-06-30T21:28:15.360Z", "content": "Thank you for your reply, Mike. I was wondering.
", "rawMarkdown": "Thank you for your reply, Mike. I was wondering.", "votes": 1 } ] } ], "comments": [ { "id": 1371386, "author_name": "Bojan Tunguz", "author_url": "", "post_date": "2021-07-01T00:40:32.600000", "content": "Thanks @michael127001 for the mention. I have not taken a look at this competition in particular, so can't comment on how much of an advantage more compute here would be compared to other recent Kaggle competitions. On one hand, this doesn't seem to be an image competition - those are usually by far the most compute intensive. On the other hand, it's still a massive dataset with signal data, so my instinct would be yes, it will require substantial computing resources.
\nIn general, if you are serious about intently competing on Kaggle I would strongly recommend that you invest in your own compute. The \"free\" online resources have come a long way in recent years, but IMHO there is nothing that compares to having a computer at your fingertips.
\nYou should also be clear with yourself about what you want to get out of Kaggle. Many top Kagglers these days are professionals or semi-professionals, so winning Kaggle competitions has become exponentially more difficult than a few years ago. However, if your goal is to learn and get better at various aspects of machine learning (which is what I initially came here for), then just doing your best and improving steadily is well worthwhile endeavor.
\nThat's my 2c.
", "votes": 18, "replies": [ { "id": 1371715, "author_name": "DeepUnderstanding", "author_url": "", "post_date": "2021-07-01T07:40:04.290000", "content": "Completely agree with @tunguz,
\nBut I really think having a good compute resource will be a big benefit in this comp. There are almost 500k targets in the training set, and for each data point, there are 3 sources that make the whole training set more than 1 million signals.
I have tried a few products (Kaggle, Colab, Paperspace, Genesis Cloud) and so far I like Genesis Cloud best.
", "votes": 1, "replies": [] }, { "id": 1371416, "author_name": "laxman kusuma", "author_url": "", "post_date": "2021-07-01T01:31:59.987000", "content": "As @tunguz and @michael127001 have told that, we have to invest our own compute.Anybody can suggest best cloud computing technologies to buy to train our model, which one is cost effective.
", "votes": 2, "replies": [ { "id": 1406865, "author_name": "", "author_url": "", "post_date": "2021-08-01T09:10:48.423000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 1401057, "author_name": "cyberia", "author_url": "", "post_date": "2021-07-26T21:37:32.623000", "content": "\n\nCan anyone win this kind of competition using only Kaggle kernel (8GB ram and gpu)
\n
Lets take it other way round, having loads of computational resource doesn't give the guarantee of winning, I feel it is right strategy at multiple levels which counts. Having good resources helps but not always and at the end of the day it's not only about winning it is also about learning new ways to do things.
", "votes": 1, "replies": [] }, { "id": 1561289, "author_name": "ChristopherZerafa", "author_url": "", "post_date": "2021-10-27T13:12:31.950000", "content": "Hey All,
\nThank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey https://forms.gle/QP9L16niPexozyhu5.
\nThank you all,
\nRegards,
\nChris
From Kaggle Grandmaster @tunguz in a recent interview:
\n\n\nFinally, if you are really serious about Kaggling, you will need to invest in your own high-end hardware. Kaggle notebooks and other free online compute resources (such as Colab) can be helpful to get you off the ground, but aside from a handful of competitions these days, most Kaggle competitions require access to a lot of computational resources.
\n
These competitions have a nasty barrier of entry to obtain fast and large computing components. Big data is always a source of contention and much of the drive for research is fitting the largest models on the smallest machines while maximizing accessibility and speed. Google has the tech to cross validate massive networks with distributed training. At least we're able to step our foot in the door with the resources Kaggle gives us 🙁
", "votes": 4, "replies": [ { "id": 1371309, "author_name": "Mphaka j", "author_url": "", "post_date": "2021-06-30T21:28:15.360000", "content": "Thank you for your reply, Mike. I was wondering.
", "votes": 1, "replies": [] } ] } ], "raw_markdown_by_id": { "1371386": "Thanks @michael127001 for the mention. I have not taken a look at this competition in particular, so can't comment on how much of an advantage more compute here would be compared to other recent Kaggle competitions. On one hand, this doesn't seem to be an image competition - those are usually by far the most compute intensive. On the other hand, it's still a massive dataset with signal data, so my instinct would be yes, it will require substantial computing resources. \n\nIn general, if you are serious about intently competing on Kaggle I would strongly recommend that you invest in your own compute. The \"free\" online resources have come a long way in recent years, but IMHO there is nothing that compares to having a computer at your fingertips. \n\nYou should also be clear with yourself about what you want to get out of Kaggle. Many top Kagglers these days are professionals or semi-professionals, so winning Kaggle competitions has become exponentially more difficult than a few years ago. However, if your goal is to learn and get better at various aspects of machine learning (which is what I initially came here for), then just doing your best and improving steadily is well worthwhile endeavor. \n\nThat's my 2c.", "1371218": "Can anyone win this kind of competition using only Kaggle kernel (8GB ram and gpu), I'm still new to competition like this so I wonder if I will blow the ram and GPU.", "1401246": "I have tried a few products (Kaggle, Colab, Paperspace, Genesis Cloud) and so far I like Genesis Cloud best.", "1371416": "As @tunguz and @michael127001 have told that, we have to invest our own compute.Anybody can suggest best cloud computing technologies to buy to train our model, which one is cost effective.", "1401057": "> Can anyone win this kind of competition using only Kaggle kernel (8GB ram and gpu)\n\nLets take it other way round, having loads of computational resource doesn't give the guarantee of winning, I feel it is right strategy at multiple levels which counts. Having good resources helps but not always and at the end of the day it's not only about winning it is also about learning new ways to do things.", "1561289": "Hey All,\n\nThank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey https://forms.gle/QP9L16niPexozyhu5.\n\nThank you all,\n\nRegards,\nChris", "1371307": "", "1371293": "From Kaggle Grandmaster @tunguz in a [recent interview](https://www.businessofbusiness.com/articles/Kaggle-grandmaster-top-10-bojan-tunguz/):\n\n> Finally, if you are really serious about Kaggling, you will need to invest in your own high-end hardware. Kaggle notebooks and other free online compute resources (such as Colab) can be helpful to get you off the ground, but aside from a handful of competitions these days, most Kaggle competitions require access to a lot of computational resources.\n\nThese competitions have a nasty barrier of entry to obtain fast and large computing components. Big data is always a source of contention and much of the drive for research is fitting the largest models on the smallest machines while maximizing accessibility and speed. Google has the tech to cross validate massive networks with distributed training. At least we're able to step our foot in the door with the resources Kaggle gives us 🙁" } }