Hi I am new to Kaggle Competitions. I would like to know the requirements to run your DL algorithms for competitions. Do you have to use distributed computing? or can you accomplish the task with you PC. I have 4GB GPU with me. Please let me know. Thanks!
", "messages": [ { "id": 2400257, "postDate": "2023-08-20T23:45:47.613Z", "content": "For this competition you can train your models on anything you have available. For prediction and scoring you must use a kaggle notebook.
\n4GB on graphics card might force you to models with small image size or small batch size (longer training time).
\nMy suggestion is that this competition is a tough place to start your kaggle experience. The data set is large, the labels very weak and you need to do segmentation at some point in the process and than do a classification model. Only 1 person seems to have developed a model that does better than just predicting the means. Your 4GB only adds to the issues.
\nThe playground is a much better point to learn the issues related to using kaggle and likely will fit your computer so you can do much of the work on local machine. I think the current playground ends today, would expect to see a new one shortly.
", "rawMarkdown": "For this competition you can train your models on anything you have available. For prediction and scoring you must use a kaggle notebook.\n\n4GB on graphics card might force you to models with small image size or small batch size (longer training time).\n\nMy suggestion is that this competition is a tough place to start your kaggle experience. The data set is large, the labels very weak and you need to do segmentation at some point in the process and than do a classification model. Only 1 person seems to have developed a model that does better than just predicting the means. Your 4GB only adds to the issues.\n\nThe [playground](https://www.kaggle.com/competitions/playground-series-s3e20) is a much better point to learn the issues related to using kaggle and likely will fit your computer so you can do much of the work on local machine. I think the current playground ends today, would expect to see a new one shortly.\n", "votes": 1, "replies": [ { "id": 2400278, "postDate": "2023-08-21T00:17:24.990Z", "content": "Thanks Jimmmy. If I were to set up a PC to accomplish my tasks, what is the spec I am looking at?
", "rawMarkdown": "Thanks Jimmmy. If I were to set up a PC to accomplish my tasks, what is the spec I am looking at?", "replies": [ { "id": 2400322, "postDate": "2023-08-21T02:01:27.397Z", "content": "I have dual GPU with 11GB each - with 512x512 image size my batch size for a tensorflow model is 8. That generally results in one training run per day with the full set of train images.
\nKaggle has resources better than my local machines - but the quota on hours does not fit my life style (retired with nothing else to do :).
\nIf your just getting started in machine learning and have limited hours of your life to spend - just use kaggle. Write your code to use GPU and TPU and you can learn within the hours quota.
", "rawMarkdown": "I have dual GPU with 11GB each - with 512x512 image size my batch size for a tensorflow model is 8. That generally results in one training run per day with the full set of train images. \n\nKaggle has resources better than my local machines - but the quota on hours does not fit my life style (retired with nothing else to do :). \n\nIf your just getting started in machine learning and have limited hours of your life to spend - just use kaggle. Write your code to use GPU and TPU and you can learn within the hours quota." } ] } ] }, { "id": 2400227, "postDate": "2023-08-20T22:24:38.073Z", "content": "Hello @asokans11, Kaggle provides all the resources you need to get started.
", "rawMarkdown": "Hello @asokans11, Kaggle provides all the resources you need to get started." }, { "id": 2400220, "postDate": "2023-08-20T22:20:18.277Z", "content": "Hi I am new to Kaggle Competitions. I would like to know the requirements to run your DL algorithms for competitions. Do you have to use distributed computing? or can you accomplish the task with you PC. I have 4GB GPU with me. Please let me know. Thanks!
", "rawMarkdown": "Hi I am new to Kaggle Competitions. I would like to know the requirements to run your DL algorithms for competitions. Do you have to use distributed computing? or can you accomplish the task with you PC. I have 4GB GPU with me. Please let me know. Thanks!\n" }, { "id": 2400245, "postDate": "2023-08-20T22:59:34.737Z", "content": "In addition to the free GPU provided by Kaggle, you can also try platforms such as Colab. After you build the pipeline, you can try to rent more stable and powerful GPUs on the GPU rental platform to train the model. Generally speaking, Kaggle will have a free quota of GPU usage, but I guess maybe you will not have enough.
", "rawMarkdown": "In addition to the free GPU provided by Kaggle, you can also try platforms such as Colab. After you build the pipeline, you can try to rent more stable and powerful GPUs on the GPU rental platform to train the model. Generally speaking, Kaggle will have a free quota of GPU usage, but I guess maybe you will not have enough.\n\n", "isDeleted": true } ], "comments": [ { "id": 2400257, "author_name": "PC Jimmmy", "author_url": "", "post_date": "2023-08-20T23:45:47.613000", "content": "For this competition you can train your models on anything you have available. For prediction and scoring you must use a kaggle notebook.
\n4GB on graphics card might force you to models with small image size or small batch size (longer training time).
\nMy suggestion is that this competition is a tough place to start your kaggle experience. The data set is large, the labels very weak and you need to do segmentation at some point in the process and than do a classification model. Only 1 person seems to have developed a model that does better than just predicting the means. Your 4GB only adds to the issues.
\nThe playground is a much better point to learn the issues related to using kaggle and likely will fit your computer so you can do much of the work on local machine. I think the current playground ends today, would expect to see a new one shortly.
", "votes": 1, "replies": [ { "id": 2400278, "author_name": "SillyPoint", "author_url": "", "post_date": "2023-08-21T00:17:24.990000", "content": "Thanks Jimmmy. If I were to set up a PC to accomplish my tasks, what is the spec I am looking at?
", "votes": 0, "replies": [ { "id": 2400322, "author_name": "PC Jimmmy", "author_url": "", "post_date": "2023-08-21T02:01:27.397000", "content": "I have dual GPU with 11GB each - with 512x512 image size my batch size for a tensorflow model is 8. That generally results in one training run per day with the full set of train images.
\nKaggle has resources better than my local machines - but the quota on hours does not fit my life style (retired with nothing else to do :).
\nIf your just getting started in machine learning and have limited hours of your life to spend - just use kaggle. Write your code to use GPU and TPU and you can learn within the hours quota.
", "votes": 0, "replies": [] } ] } ] }, { "id": 2400227, "author_name": "C4rl05/V", "author_url": "", "post_date": "2023-08-20T22:24:38.073000", "content": "Hello @asokans11, Kaggle provides all the resources you need to get started.
", "votes": 0, "replies": [] }, { "id": 2400245, "author_name": "", "author_url": "", "post_date": "2023-08-20T22:59:34.737000", "content": "In addition to the free GPU provided by Kaggle, you can also try platforms such as Colab. After you build the pipeline, you can try to rent more stable and powerful GPUs on the GPU rental platform to train the model. Generally speaking, Kaggle will have a free quota of GPU usage, but I guess maybe you will not have enough.
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "2400257": "For this competition you can train your models on anything you have available. For prediction and scoring you must use a kaggle notebook.\n\n4GB on graphics card might force you to models with small image size or small batch size (longer training time).\n\nMy suggestion is that this competition is a tough place to start your kaggle experience. The data set is large, the labels very weak and you need to do segmentation at some point in the process and than do a classification model. Only 1 person seems to have developed a model that does better than just predicting the means. Your 4GB only adds to the issues.\n\nThe [playground](https://www.kaggle.com/competitions/playground-series-s3e20) is a much better point to learn the issues related to using kaggle and likely will fit your computer so you can do much of the work on local machine. I think the current playground ends today, would expect to see a new one shortly.\n", "2400227": "Hello @asokans11, Kaggle provides all the resources you need to get started.", "2400220": "Hi I am new to Kaggle Competitions. I would like to know the requirements to run your DL algorithms for competitions. Do you have to use distributed computing? or can you accomplish the task with you PC. I have 4GB GPU with me. Please let me know. Thanks!\n", "2400245": "In addition to the free GPU provided by Kaggle, you can also try platforms such as Colab. After you build the pipeline, you can try to rent more stable and powerful GPUs on the GPU rental platform to train the model. Generally speaking, Kaggle will have a free quota of GPU usage, but I guess maybe you will not have enough.\n\n" } }