This is my first kernel only competition. It sucks. A lot.
\n\nThe reason for that is that most of the time Kaggle can't save my changes or committing hangs, whatever. Notebook or pure code. No difference.
\n\nCommitting does not work with Chromium at all, but with Firefox (sometimes). On the other hand loading the leaderboard hangs in Firefox and works in Chromium.
\n\nIf the kernel runs its much slower than on my computer (which is a quite average model).\nI can't store features properly, they have to be recalculated every time I fork (aka rollback) the kernel.
\n\nI'm a developer. I like to use functions for reducing duplicate code, which is not possible with notebooks (you have to copy them into every notebook). I like to have a proper IDE and version control.
\n\nI spend more time fighting with the Kaggle website than working on the model.\n(I even gave up on making text bold in this editor. Completely unusable.)
\n\nIn my view it makes absolutely no sense to host kernel only competitions with such a platform.
\n\nDon't get me wrong. I like Kaggle competitions a lot. But being forced to use such a flaky environment is just frustrating.
", "messages": [ { "id": "458779", "postDate": "01/20/2019 13:31:56", "content": "This is my first kernel only competition. It sucks. A lot.
\n\nThe reason for that is that most of the time Kaggle can't save my changes or committing hangs, whatever. Notebook or pure code. No difference.
\n\nCommitting does not work with Chromium at all, but with Firefox (sometimes). On the other hand loading the leaderboard hangs in Firefox and works in Chromium.
\n\nIf the kernel runs its much slower than on my computer (which is a quite average model).\nI can't store features properly, they have to be recalculated every time I fork (aka rollback) the kernel.
\n\nI'm a developer. I like to use functions for reducing duplicate code, which is not possible with notebooks (you have to copy them into every notebook). I like to have a proper IDE and version control.
\n\nI spend more time fighting with the Kaggle website than working on the model.\n(I even gave up on making text bold in this editor. Completely unusable.)
\n\nIn my view it makes absolutely no sense to host kernel only competitions with such a platform.
\n\nDon't get me wrong. I like Kaggle competitions a lot. But being forced to use such a flaky environment is just frustrating.
", "rawMarkdown": "This is my first kernel only competition. It sucks. A lot.\n\nThe reason for that is that most of the time Kaggle can't save my changes or committing hangs, whatever. Notebook or pure code. No difference.\n\nCommitting does not work with Chromium at all, but with Firefox (sometimes). On the other hand loading the leaderboard hangs in Firefox and works in Chromium.\n\nIf the kernel runs its much slower than on my computer (which is a quite average model).\nI can't store features properly, they have to be recalculated every time I fork (aka rollback) the kernel.\n\nI'm a developer. I like to use functions for reducing duplicate code, which is not possible with notebooks (you have to copy them into every notebook). I like to have a proper IDE and version control.\n\nI spend more time fighting with the Kaggle website than working on the model.\n(I even gave up on making text bold in this editor. Completely unusable.)\n\nIn my view it makes absolutely no sense to host kernel only competitions with such a platform.\n\nDon't get me wrong. I like Kaggle competitions a lot. But being forced to use such a flaky environment is just frustrating.", "votes": null }, { "id": "458782", "postDate": "01/20/2019 13:41:42", "content": "I think today working experience is quite strange. Almost every time today, my editor cannot \"save draft\" and I have to refresh a lot before I can commit.
\n\nMy experience in the other day is better, though.
", "rawMarkdown": "I think today working experience is quite strange. Almost every time today, my editor cannot \"save draft\" and I have to refresh a lot before I can commit.\n\nMy experience in the other day is better, though.", "votes": null }, { "id": "458791", "postDate": "01/20/2019 13:56:44", "content": "I am facing the issue of kernels loading slow as well as drafts not getting saved. Since the changes made to the UI, it has gone downhill for me.
", "rawMarkdown": "I am facing the issue of kernels loading slow as well as drafts not getting saved. Since the changes made to the UI, it has gone downhill for me.", "votes": null }, { "id": "458798", "postDate": "01/20/2019 14:01:35", "content": "The Kaggle website is slow for me most of the time.
", "rawMarkdown": "The Kaggle website is slow for me most of the time.", "votes": null }, { "id": "458799", "postDate": "01/20/2019 14:02:36", "content": "I try to change one line. If it can be saved as draft I continue, otherwise I reload the kernel.
", "rawMarkdown": "I try to change one line. If it can be saved as draft I continue, otherwise I reload the kernel.", "votes": null }, { "id": "458805", "postDate": "01/20/2019 14:11:10", "content": "It is a problem of the last 36 hours, I hope they are going to fix it soon because at the time there is no use for Kernels
", "rawMarkdown": "It is a problem of the last 36 hours, I hope they are going to fix it soon because at the time there is no use for Kernels", "votes": null }, { "id": "458806", "postDate": "01/20/2019 14:13:02", "content": "I totally agree! I have also problem when try add a second dataset to a new kernel...
", "rawMarkdown": "I totally agree! I have also problem when try add a second dataset to a new kernel...", "votes": null }, { "id": "458811", "postDate": "01/20/2019 14:22:03", "content": "The issue is raised here. I am also facing the same issue. I expect kaggle admins to look into the issue\nhttps://www.kaggle.com/product-feedback/78031#458790
", "rawMarkdown": "The issue is raised here. I am also facing the same issue. I expect kaggle admins to look into the issue\nhttps://www.kaggle.com/product-feedback/78031#458790", "votes": null }, { "id": "458818", "postDate": "01/20/2019 14:39:58", "content": "Nothing stops you from developing your stuff locally and then only copy the script or notebook to Kaggle when you are done. Alternatively, you can also use Google Colab. I don't see the issue. Kernels can be a bit shaky lately, but I am happy that Kaggle/Google offers a GPU infrastructure basically for free. At least in this format everyone has access to the same infrastructure. That does not mean though that I would not be happy if they would make it more stable.
", "rawMarkdown": "Nothing stops you from developing your stuff locally and then only copy the script or notebook to Kaggle when you are done. Alternatively, you can also use Google Colab. I don't see the issue. Kernels can be a bit shaky lately, but I am happy that Kaggle/Google offers a GPU infrastructure basically for free. At least in this format everyone has access to the same infrastructure. That does not mean though that I would not be happy if they would make it more stable.", "votes": null }, { "id": "458845", "postDate": "01/20/2019 15:55:39", "content": "Yes. And I try to submit that code for about 4h now...
", "rawMarkdown": "Yes. And I try to submit that code for about 4h now...", "votes": null }, { "id": "458853", "postDate": "01/20/2019 16:34:46", "content": "I have organized my code into a library consisting of multiple different files that I develop locally and keep under version control. When I want to do a kernel commit, I use a python package called stickytape to compile my codebase into a single script that can be copy and pasted into a script kernel. This way, I can autogenerate different variations of the script programatically with different hyperparameter choices and models.
\n\nCopy and pasting into the webUI is not a good use of time, but the kaggle API gives me errors whenever I try to push code to a kernel. The README implies that kernels are not fully supported in the API yet.
", "rawMarkdown": "I have organized my code into a library consisting of multiple different files that I develop locally and keep under version control. When I want to do a kernel commit, I use a python package called stickytape to compile my codebase into a single script that can be copy and pasted into a script kernel. This way, I can autogenerate different variations of the script programatically with different hyperparameter choices and models.\n\nCopy and pasting into the webUI is not a good use of time, but the kaggle API gives me errors whenever I try to push code to a kernel. The README implies that kernels are not fully supported in the API yet.", "votes": null }, { "id": "458866", "postDate": "01/20/2019 17:13:35", "content": "Thats a great idea. Thanks. I'll check that out.
", "rawMarkdown": "Thats a great idea. Thanks. I'll check that out.", "votes": null }, { "id": "458867", "postDate": "01/20/2019 17:14:24", "content": "Thanks for the link.
", "rawMarkdown": "Thanks for the link.", "votes": null }, { "id": "458918", "postDate": "01/20/2019 20:04:57", "content": "I 'fixed' the API by just forking my kernel and it seemed to work. Now I have a workflow for hyperparameter optimization by running each hyperparameter configuration as a kernel version. The API can collect the outputs from each version for local analysis to identify the best performing configuration.
", "rawMarkdown": "I 'fixed' the API by just forking my kernel and it seemed to work. Now I have a workflow for hyperparameter optimization by running each hyperparameter configuration as a kernel version. The API can collect the outputs from each version for local analysis to identify the best performing configuration.", "votes": null }, { "id": "459005", "postDate": "01/21/2019 01:57:53", "content": "Don't get why some people are downvoting you. You've raised some completely valid points; the experience with the Kaggle interface has been extremely frustrating recently.
", "rawMarkdown": "Don't get why some people are downvoting you. You've raised some completely valid points; the experience with the Kaggle interface has been extremely frustrating recently.", "votes": null }, { "id": "459020", "postDate": "01/21/2019 03:10:55", "content": "Yes, it is sometimes painful but freely playing with many sessions of 12 gb memory vga is a pure pleasure. And I'm impressed by their fast response time for new kernel generation also.
", "rawMarkdown": "Yes, it is sometimes painful but freely playing with many sessions of 12 gb memory vga is a pure pleasure. And I'm impressed by their fast response time for new kernel generation also.", "votes": null }, { "id": "459021", "postDate": "01/21/2019 03:12:29", "content": "@Anshul Rai: excellent point! those who downvoted him are the same with kernel issues recently..lol
", "rawMarkdown": "Anshul Rai: excellent point! those who downvoted him are the same with kernel issues recently..lol", "votes": null }, { "id": "459127", "postDate": "01/21/2019 08:39:59", "content": "Download Anaconda, use the Jupyter Notebooks there. Upload the code later on. If you are collaborating just download the latest working version from kaggle. Regarding data uploading updating please consult https://codelabs.developers.google.com/codelabs/upload-update-data-kaggle-api/index.html#1
", "rawMarkdown": "Download Anaconda, use the Jupyter Notebooks there. Upload the code later on. If you are collaborating just download the latest working version from kaggle. Regarding data uploading updating please consult https://codelabs.developers.google.com/codelabs/upload-update-data-kaggle-api/index.html#1", "votes": null }, { "id": "459192", "postDate": "01/21/2019 11:20:23", "content": "I think Kaggle kernels work best with Chrome, so I am doing that. The only thing that happens to me once in a while is that changes do not get saved (see a round, red circle that is busy instead of quickly showing \"draft saved). I just keep an eye on that, so that I don't lose much if that happens. Have not had any failed runs lately.
", "rawMarkdown": "I think Kaggle kernels work best with Chrome, so I am doing that. The only thing that happens to me once in a while is that changes do not get saved (see a round, red circle that is busy instead of quickly showing \"draft saved). I just keep an eye on that, so that I don't lose much if that happens. Have not had any failed runs lately.", "votes": null }, { "id": "459285", "postDate": "01/21/2019 14:00:31", "content": "@arigion Today, my kernel experience returns back to normal (i mean 'good') ... Maybe you can give our kernel a 2nd chance :)
", "rawMarkdown": "arigion Today, my kernel experience returns back to normal (i mean 'good') ... Maybe you can give our kernel a 2nd chance :)", "votes": null }, { "id": "459310", "postDate": "01/21/2019 15:07:35", "content": "I don't understand the downvotes either. But frankly speaking I don't mind. I do not write posts for artificial medals but to talk about things that matter to me.
", "rawMarkdown": "I don't understand the downvotes either. But frankly speaking I don't mind. I do not write posts for artificial medals but to talk about things that matter to me.", "votes": null }, { "id": "459313", "postDate": "01/21/2019 15:16:24", "content": "Actually I use the Atom editor with the Hydrogen plugin (https://atom.io/packages/hydrogen). You can use it like a normal Python IDE or just execute single lines/blocks like in a Jupyter notebook with Hydrogen, but with git integration and refactoring. I don't like Jupyter Notebooks for actual models. They are great for exploring or testing stuff, but I don't really like them for coding. So my workflow is a Jupyter notebook for testing and writing reusable code into a library which I import into my models/notebooks.\nThat's one of the reasons why the kernel only competition annoys me, because I can't use my normal workflow and many tasks are so inconvenient.
", "rawMarkdown": "Actually I use the Atom editor with the Hydrogen plugin (https://atom.io/packages/hydrogen). You can use it like a normal Python IDE or just execute single lines/blocks like in a Jupyter notebook with Hydrogen, but with git integration and refactoring. I don't like Jupyter Notebooks for actual models. They are great for exploring or testing stuff, but I don't really like them for coding. So my workflow is a Jupyter notebook for testing and writing reusable code into a library which I import into my models/notebooks.\nThat's one of the reasons why the kernel only competition annoys me, because I can't use my normal workflow and many tasks are so inconvenient.", "votes": null }, { "id": "459377", "postDate": "01/21/2019 16:22:35", "content": "my kernel is stuck for more than an hour on a simple one lgbm model, that on my machine run in less than 7 minutes..;\ni'm a niubie, ok, but I doubt that most of real expert analysts can waste their time on this... \"only (inefficient) kernel competition\"...please...
", "rawMarkdown": "my kernel is stuck for more than an hour on a simple one lgbm model, that on my machine run in less than 7 minutes..;\ni'm a niubie, ok, but I doubt that most of real expert analysts can waste their time on this... \"only (inefficient) kernel competition\"...please...", "votes": null }, { "id": "459460", "postDate": "01/21/2019 19:22:15", "content": "Yes, you're right, At least in this way everyone has access to the same infrastructure.
", "rawMarkdown": "Yes, you're right, At least in this way everyone has access to the same infrastructure.", "votes": null }, { "id": "459567", "postDate": "01/22/2019 02:15:35", "content": "I guess it's just for avoiding cheats or something
", "rawMarkdown": "I guess it's just for avoiding cheats or something", "votes": null }, { "id": "459770", "postDate": "01/22/2019 10:40:54", "content": "i always work online, and it works flawless, except on iPad or old PC with small memory.
", "rawMarkdown": "i always work online, and it works flawless, except on iPad or old PC with small memory.", "votes": null }, { "id": "459772", "postDate": "01/22/2019 10:43:05", "content": "i personally think when you have to create your submission online, the competition solution is more levelplaying for everybody and doesnot give a benefit to people having an extreme fast computer available. I don't think competitions have to work on data of 1 terabyte fe, to generate a solution that could be tested on 1Gb of data ?
", "rawMarkdown": "i personally think when you have to create your submission online, the competition solution is more levelplaying for everybody and doesnot give a benefit to people having an extreme fast computer available. I don't think competitions have to work on data of 1 terabyte fe, to generate a solution that could be tested on 1Gb of data ?", "votes": null }, { "id": "459848", "postDate": "01/22/2019 13:39:07", "content": "Whatever doesn't kill you simply makes you stranger.
", "rawMarkdown": "Whatever doesn't kill you simply makes you stranger.", "votes": null }, { "id": "460090", "postDate": "01/23/2019 01:34:08", "content": "How about click the download button to keep your script?
", "rawMarkdown": "How about click the download button to keep your script?", "votes": null }, { "id": "460709", "postDate": "01/24/2019 09:23:43", "content": "I sort of agree. But what's stopping you from using functions, IDE and version control? :) I use py files, not notebooks.
", "rawMarkdown": "I sort of agree. But what's stopping you from using functions, IDE and version control? :) I use py files, not notebooks.", "votes": null }, { "id": "460885", "postDate": "01/24/2019 16:05:17", "content": "This competition actually necessitates using a kernel based competition due to the nature of the data. Their dataset is derived from actual questions posted to Quora. The insincere questions are removed from their site since they violate their site guidelines. As a result one could get good results simply by cross referencing questions in the dataset and whether the question is still up on Quora's website. In addition, it helps reduce the complexity of solutions so it isn't some crazy blend of hundreds of models.
", "rawMarkdown": "This competition actually necessitates using a kernel based competition due to the nature of the data. Their dataset is derived from actual questions posted to Quora. The insincere questions are removed from their site since they violate their site guidelines. As a result one could get good results simply by cross referencing questions in the dataset and whether the question is still up on Quora's website. In addition, it helps reduce the complexity of solutions so it isn't some crazy blend of hundreds of models.", "votes": null }, { "id": "461761", "postDate": "01/26/2019 23:59:28", "content": "I liked the idea of the kernel competition with the time limits etc and level setting what resources folks can use. however the kernel environment needs improvement. Too many hangs, errors, and just unstable experience made it frustrating. Hopefully these kernel only comps are used as a way to improve the kernel environment.
", "rawMarkdown": "I liked the idea of the kernel competition with the time limits etc and level setting what resources folks can use. however the kernel environment needs improvement. Too many hangs, errors, and just unstable experience made it frustrating. Hopefully these kernel only comps are used as a way to improve the kernel environment.", "votes": null }, { "id": "462466", "postDate": "01/28/2019 10:40:17", "content": "We should commit the kernel and after a few hours, we need to go back to see if the code has been executed correctly.\nA complete waste of time!
", "rawMarkdown": "We should commit the kernel and after a few hours, we need to go back to see if the code has been executed correctly.\nA complete waste of time!", "votes": null }, { "id": "463405", "postDate": "01/29/2019 23:51:37", "content": "I wanted to jump in here and let everyone know that we very much appreciate this direct feedback. Our product and engineering teams see feedback like this and either work to address them in the short term, or prioritize them on a longer-term product roadmap. Community members who have been with Kaggle over the years know that we do listen and work hard to respond to the pain points our users experience. Stay tuned for some updates on specific issues. Thanks!
", "rawMarkdown": "I wanted to jump in here and let everyone know that we very much appreciate this direct feedback. Our product and engineering teams see feedback like this and either work to address them in the short term, or prioritize them on a longer-term product roadmap. Community members who have been with Kaggle over the years know that we do listen and work hard to respond to the pain points our users experience. Stay tuned for some updates on specific issues. Thanks!", "votes": null }, { "id": "463927", "postDate": "01/30/2019 22:54:56", "content": "There are some hiccups particularly when you are trying to commit code. Sometimes commit hangs and often it is very slow.
", "rawMarkdown": "There are some hiccups particularly when you are trying to commit code. Sometimes commit hangs and often it is very slow.", "votes": null }, { "id": "464269", "postDate": "01/31/2019 14:14:00", "content": "Is there any Plugin for Visual Studio Code or PyCharm to edit the Jupyter Notebook locally, but run it remotely on Kaggle?
", "rawMarkdown": "Is there any **Plugin** for **Visual Studio Code** or **PyCharm** to edit the Jupyter Notebook locally, but run it remotely on Kaggle?", "votes": null }, { "id": "464840", "postDate": "02/01/2019 16:01:49", "content": "You can always work offline using Jupyter Notebook and upload them later.\nBut I do had the same problems when I did try to commit a kernal with hyperparameter optimization, had to comment the optimization and hardcode these parameters later.
", "rawMarkdown": "You can always work offline using Jupyter Notebook and upload them later.\nBut I do had the same problems when I did try to commit a kernal with hyperparameter optimization, had to comment the optimization and hardcode these parameters later.", "votes": null }, { "id": "465143", "postDate": "02/02/2019 12:35:16", "content": "While I do agree that this post has some valid points, kernel only competitions give people like me a chance to participate on a contest without needing to have access to state of the art hardware or (quite expensive) cloud compute.
", "rawMarkdown": "While I do agree that this post has some valid points, kernel only competitions give people like me a chance to participate on a contest without needing to have access to state of the art hardware or (quite expensive) cloud compute.", "votes": null }, { "id": "465192", "postDate": "02/02/2019 15:28:32", "content": "I am unable to select my kernel for final score. Does anyone get this?
", "rawMarkdown": "I am unable to select my kernel for final score. Does anyone get this?", "votes": null }, { "id": "484190", "postDate": "03/05/2019 17:04:54", "content": "@happyprancer \nCan you explain how you use stickytape ?\nI tried it, but it produces a lot of garbage into a single file...
Could you install the GPU-edtion lightgbm please~~~\nI have to install it each time I open the notebook.\nThanks
", "rawMarkdown": "Could you install the GPU-edtion lightgbm please~~~\nI have to install it each time I open the notebook.\nThanks", "votes": null }, { "id": "1200749", "postDate": "02/15/2021 00:41:25", "content": "I just came back after quitting for this exact reason.
", "rawMarkdown": "I just came back after quitting for this exact reason.", "votes": null } ], "comments": [ { "id": 1200749, "author_name": "willrice", "author_url": "", "post_date": "02/15/2021 00:41:25", "content": "I just came back after quitting for this exact reason.
", "votes": null, "replies": [] }, { "id": 458782, "author_name": "ratthachat", "author_url": "", "post_date": "01/20/2019 13:41:42", "content": "I think today working experience is quite strange. Almost every time today, my editor cannot \"save draft\" and I have to refresh a lot before I can commit.
\n\nMy experience in the other day is better, though.
", "votes": null, "replies": [ { "id": 458799, "author_name": "arigion", "author_url": "", "post_date": "01/20/2019 14:02:36", "content": "I try to change one line. If it can be saved as draft I continue, otherwise I reload the kernel.
", "votes": null, "replies": [] }, { "id": 458805, "author_name": "lucabasa", "author_url": "", "post_date": "01/20/2019 14:11:10", "content": "It is a problem of the last 36 hours, I hope they are going to fix it soon because at the time there is no use for Kernels
", "votes": null, "replies": [] }, { "id": 459285, "author_name": "ratthachat", "author_url": "", "post_date": "01/21/2019 14:00:31", "content": "@arigion Today, my kernel experience returns back to normal (i mean 'good') ... Maybe you can give our kernel a 2nd chance :)
", "votes": null, "replies": [] } ] }, { "id": 458791, "author_name": "satian", "author_url": "", "post_date": "01/20/2019 13:56:44", "content": "I am facing the issue of kernels loading slow as well as drafts not getting saved. Since the changes made to the UI, it has gone downhill for me.
", "votes": null, "replies": [ { "id": 458798, "author_name": "arigion", "author_url": "", "post_date": "01/20/2019 14:01:35", "content": "The Kaggle website is slow for me most of the time.
", "votes": null, "replies": [] } ] }, { "id": 458806, "author_name": "madara", "author_url": "", "post_date": "01/20/2019 14:13:02", "content": "I totally agree! I have also problem when try add a second dataset to a new kernel...
", "votes": null, "replies": [] }, { "id": 458811, "author_name": "suchith0312", "author_url": "", "post_date": "01/20/2019 14:22:03", "content": "The issue is raised here. I am also facing the same issue. I expect kaggle admins to look into the issue\nhttps://www.kaggle.com/product-feedback/78031#458790
", "votes": null, "replies": [ { "id": 458867, "author_name": "arigion", "author_url": "", "post_date": "01/20/2019 17:14:24", "content": "Thanks for the link.
", "votes": null, "replies": [] } ] }, { "id": 458818, "author_name": "philippsinger", "author_url": "", "post_date": "01/20/2019 14:39:58", "content": "Nothing stops you from developing your stuff locally and then only copy the script or notebook to Kaggle when you are done. Alternatively, you can also use Google Colab. I don't see the issue. Kernels can be a bit shaky lately, but I am happy that Kaggle/Google offers a GPU infrastructure basically for free. At least in this format everyone has access to the same infrastructure. That does not mean though that I would not be happy if they would make it more stable.
", "votes": null, "replies": [ { "id": 458845, "author_name": "arigion", "author_url": "", "post_date": "01/20/2019 15:55:39", "content": "Yes. And I try to submit that code for about 4h now...
", "votes": null, "replies": [] }, { "id": 459460, "author_name": "jmourad100", "author_url": "", "post_date": "01/21/2019 19:22:15", "content": "Yes, you're right, At least in this way everyone has access to the same infrastructure.
", "votes": null, "replies": [] } ] }, { "id": 458853, "author_name": "happyprancer", "author_url": "", "post_date": "01/20/2019 16:34:46", "content": "I have organized my code into a library consisting of multiple different files that I develop locally and keep under version control. When I want to do a kernel commit, I use a python package called stickytape to compile my codebase into a single script that can be copy and pasted into a script kernel. This way, I can autogenerate different variations of the script programatically with different hyperparameter choices and models.
\n\nCopy and pasting into the webUI is not a good use of time, but the kaggle API gives me errors whenever I try to push code to a kernel. The README implies that kernels are not fully supported in the API yet.
", "votes": null, "replies": [ { "id": 458866, "author_name": "arigion", "author_url": "", "post_date": "01/20/2019 17:13:35", "content": "Thats a great idea. Thanks. I'll check that out.
", "votes": null, "replies": [] }, { "id": 458918, "author_name": "happyprancer", "author_url": "", "post_date": "01/20/2019 20:04:57", "content": "I 'fixed' the API by just forking my kernel and it seemed to work. Now I have a workflow for hyperparameter optimization by running each hyperparameter configuration as a kernel version. The API can collect the outputs from each version for local analysis to identify the best performing configuration.
", "votes": null, "replies": [] }, { "id": 484190, "author_name": "franky12", "author_url": "", "post_date": "03/05/2019 17:04:54", "content": "@happyprancer \nCan you explain how you use stickytape ?\nI tried it, but it produces a lot of garbage into a single file...
Don't get why some people are downvoting you. You've raised some completely valid points; the experience with the Kaggle interface has been extremely frustrating recently.
", "votes": null, "replies": [ { "id": 459021, "author_name": "", "author_url": "", "post_date": "01/21/2019 03:12:29", "content": "@Anshul Rai: excellent point! those who downvoted him are the same with kernel issues recently..lol
", "votes": null, "replies": [] }, { "id": 459310, "author_name": "arigion", "author_url": "", "post_date": "01/21/2019 15:07:35", "content": "I don't understand the downvotes either. But frankly speaking I don't mind. I do not write posts for artificial medals but to talk about things that matter to me.
", "votes": null, "replies": [] } ] }, { "id": 459020, "author_name": "soonhwankwon", "author_url": "", "post_date": "01/21/2019 03:10:55", "content": "Yes, it is sometimes painful but freely playing with many sessions of 12 gb memory vga is a pure pleasure. And I'm impressed by their fast response time for new kernel generation also.
", "votes": null, "replies": [] }, { "id": 459127, "author_name": "zikazika", "author_url": "", "post_date": "01/21/2019 08:39:59", "content": "Download Anaconda, use the Jupyter Notebooks there. Upload the code later on. If you are collaborating just download the latest working version from kaggle. Regarding data uploading updating please consult https://codelabs.developers.google.com/codelabs/upload-update-data-kaggle-api/index.html#1
", "votes": null, "replies": [ { "id": 459313, "author_name": "arigion", "author_url": "", "post_date": "01/21/2019 15:16:24", "content": "Actually I use the Atom editor with the Hydrogen plugin (https://atom.io/packages/hydrogen). You can use it like a normal Python IDE or just execute single lines/blocks like in a Jupyter notebook with Hydrogen, but with git integration and refactoring. I don't like Jupyter Notebooks for actual models. They are great for exploring or testing stuff, but I don't really like them for coding. So my workflow is a Jupyter notebook for testing and writing reusable code into a library which I import into my models/notebooks.\nThat's one of the reasons why the kernel only competition annoys me, because I can't use my normal workflow and many tasks are so inconvenient.
", "votes": null, "replies": [] } ] }, { "id": 459192, "author_name": "erikbruin", "author_url": "", "post_date": "01/21/2019 11:20:23", "content": "I think Kaggle kernels work best with Chrome, so I am doing that. The only thing that happens to me once in a while is that changes do not get saved (see a round, red circle that is busy instead of quickly showing \"draft saved). I just keep an eye on that, so that I don't lose much if that happens. Have not had any failed runs lately.
", "votes": null, "replies": [ { "id": 460090, "author_name": "niuddd", "author_url": "", "post_date": "01/23/2019 01:34:08", "content": "How about click the download button to keep your script?
", "votes": null, "replies": [] } ] }, { "id": 459377, "author_name": "madara", "author_url": "", "post_date": "01/21/2019 16:22:35", "content": "my kernel is stuck for more than an hour on a simple one lgbm model, that on my machine run in less than 7 minutes..;\ni'm a niubie, ok, but I doubt that most of real expert analysts can waste their time on this... \"only (inefficient) kernel competition\"...please...
", "votes": null, "replies": [] }, { "id": 459567, "author_name": "icemekaveli", "author_url": "", "post_date": "01/22/2019 02:15:35", "content": "I guess it's just for avoiding cheats or something
", "votes": null, "replies": [] }, { "id": 459770, "author_name": "plarmuseau", "author_url": "", "post_date": "01/22/2019 10:40:54", "content": "i always work online, and it works flawless, except on iPad or old PC with small memory.
", "votes": null, "replies": [] }, { "id": 459772, "author_name": "plarmuseau", "author_url": "", "post_date": "01/22/2019 10:43:05", "content": "i personally think when you have to create your submission online, the competition solution is more levelplaying for everybody and doesnot give a benefit to people having an extreme fast computer available. I don't think competitions have to work on data of 1 terabyte fe, to generate a solution that could be tested on 1Gb of data ?
", "votes": null, "replies": [] }, { "id": 459848, "author_name": "xiaobai1123q", "author_url": "", "post_date": "01/22/2019 13:39:07", "content": "Whatever doesn't kill you simply makes you stranger.
", "votes": null, "replies": [] }, { "id": 460709, "author_name": "artyomp", "author_url": "", "post_date": "01/24/2019 09:23:43", "content": "I sort of agree. But what's stopping you from using functions, IDE and version control? :) I use py files, not notebooks.
", "votes": null, "replies": [] }, { "id": 460885, "author_name": "msf908", "author_url": "", "post_date": "01/24/2019 16:05:17", "content": "This competition actually necessitates using a kernel based competition due to the nature of the data. Their dataset is derived from actual questions posted to Quora. The insincere questions are removed from their site since they violate their site guidelines. As a result one could get good results simply by cross referencing questions in the dataset and whether the question is still up on Quora's website. In addition, it helps reduce the complexity of solutions so it isn't some crazy blend of hundreds of models.
", "votes": null, "replies": [] }, { "id": 461761, "author_name": "georgedittmar", "author_url": "", "post_date": "01/26/2019 23:59:28", "content": "I liked the idea of the kernel competition with the time limits etc and level setting what resources folks can use. however the kernel environment needs improvement. Too many hangs, errors, and just unstable experience made it frustrating. Hopefully these kernel only comps are used as a way to improve the kernel environment.
", "votes": null, "replies": [] }, { "id": 462466, "author_name": "mjbahmani", "author_url": "", "post_date": "01/28/2019 10:40:17", "content": "We should commit the kernel and after a few hours, we need to go back to see if the code has been executed correctly.\nA complete waste of time!
", "votes": null, "replies": [] }, { "id": 463405, "author_name": "inversion", "author_url": "", "post_date": "01/29/2019 23:51:37", "content": "I wanted to jump in here and let everyone know that we very much appreciate this direct feedback. Our product and engineering teams see feedback like this and either work to address them in the short term, or prioritize them on a longer-term product roadmap. Community members who have been with Kaggle over the years know that we do listen and work hard to respond to the pain points our users experience. Stay tuned for some updates on specific issues. Thanks!
", "votes": null, "replies": [ { "id": 622792, "author_name": "appleyuchi", "author_url": "", "post_date": "09/10/2019 05:22:00", "content": "Could you install the GPU-edtion lightgbm please~~~\nI have to install it each time I open the notebook.\nThanks
", "votes": null, "replies": [] } ] }, { "id": 463927, "author_name": "nik100200", "author_url": "", "post_date": "01/30/2019 22:54:56", "content": "There are some hiccups particularly when you are trying to commit code. Sometimes commit hangs and often it is very slow.
", "votes": null, "replies": [] }, { "id": 464269, "author_name": "franky12", "author_url": "", "post_date": "01/31/2019 14:14:00", "content": "Is there any Plugin for Visual Studio Code or PyCharm to edit the Jupyter Notebook locally, but run it remotely on Kaggle?
", "votes": null, "replies": [] }, { "id": 464840, "author_name": "shreyask92", "author_url": "", "post_date": "02/01/2019 16:01:49", "content": "You can always work offline using Jupyter Notebook and upload them later.\nBut I do had the same problems when I did try to commit a kernal with hyperparameter optimization, had to comment the optimization and hardcode these parameters later.
", "votes": null, "replies": [] }, { "id": 465143, "author_name": "amielle", "author_url": "", "post_date": "02/02/2019 12:35:16", "content": "While I do agree that this post has some valid points, kernel only competitions give people like me a chance to participate on a contest without needing to have access to state of the art hardware or (quite expensive) cloud compute.
", "votes": null, "replies": [] }, { "id": 465192, "author_name": "huytran", "author_url": "", "post_date": "02/02/2019 15:28:32", "content": "I am unable to select my kernel for final score. Does anyone get this?
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "458779": "This is my first kernel only competition. It sucks. A lot.\n\nThe reason for that is that most of the time Kaggle can't save my changes or committing hangs, whatever. Notebook or pure code. No difference.\n\nCommitting does not work with Chromium at all, but with Firefox (sometimes). On the other hand loading the leaderboard hangs in Firefox and works in Chromium.\n\nIf the kernel runs its much slower than on my computer (which is a quite average model).\nI can't store features properly, they have to be recalculated every time I fork (aka rollback) the kernel.\n\nI'm a developer. I like to use functions for reducing duplicate code, which is not possible with notebooks (you have to copy them into every notebook). I like to have a proper IDE and version control.\n\nI spend more time fighting with the Kaggle website than working on the model.\n(I even gave up on making text bold in this editor. Completely unusable.)\n\nIn my view it makes absolutely no sense to host kernel only competitions with such a platform.\n\nDon't get me wrong. I like Kaggle competitions a lot. But being forced to use such a flaky environment is just frustrating.", "458782": "I think today working experience is quite strange. Almost every time today, my editor cannot \"save draft\" and I have to refresh a lot before I can commit.\n\nMy experience in the other day is better, though.", "458791": "I am facing the issue of kernels loading slow as well as drafts not getting saved. Since the changes made to the UI, it has gone downhill for me.", "458798": "The Kaggle website is slow for me most of the time.", "458799": "I try to change one line. If it can be saved as draft I continue, otherwise I reload the kernel.", "458805": "It is a problem of the last 36 hours, I hope they are going to fix it soon because at the time there is no use for Kernels", "458806": "I totally agree! I have also problem when try add a second dataset to a new kernel...", "458811": "The issue is raised here. I am also facing the same issue. I expect kaggle admins to look into the issue\nhttps://www.kaggle.com/product-feedback/78031#458790", "458818": "Nothing stops you from developing your stuff locally and then only copy the script or notebook to Kaggle when you are done. Alternatively, you can also use Google Colab. I don't see the issue. Kernels can be a bit shaky lately, but I am happy that Kaggle/Google offers a GPU infrastructure basically for free. At least in this format everyone has access to the same infrastructure. That does not mean though that I would not be happy if they would make it more stable.", "458845": "Yes. And I try to submit that code for about 4h now...", "458853": "I have organized my code into a library consisting of multiple different files that I develop locally and keep under version control. When I want to do a kernel commit, I use a python package called stickytape to compile my codebase into a single script that can be copy and pasted into a script kernel. This way, I can autogenerate different variations of the script programatically with different hyperparameter choices and models.\n\nCopy and pasting into the webUI is not a good use of time, but the kaggle API gives me errors whenever I try to push code to a kernel. The README implies that kernels are not fully supported in the API yet.", "458866": "Thats a great idea. Thanks. I'll check that out.", "458867": "Thanks for the link.", "458918": "I 'fixed' the API by just forking my kernel and it seemed to work. Now I have a workflow for hyperparameter optimization by running each hyperparameter configuration as a kernel version. The API can collect the outputs from each version for local analysis to identify the best performing configuration.", "459005": "Don't get why some people are downvoting you. You've raised some completely valid points; the experience with the Kaggle interface has been extremely frustrating recently.", "459020": "Yes, it is sometimes painful but freely playing with many sessions of 12 gb memory vga is a pure pleasure. And I'm impressed by their fast response time for new kernel generation also.", "459021": "Anshul Rai: excellent point! those who downvoted him are the same with kernel issues recently..lol", "459127": "Download Anaconda, use the Jupyter Notebooks there. Upload the code later on. If you are collaborating just download the latest working version from kaggle. Regarding data uploading updating please consult https://codelabs.developers.google.com/codelabs/upload-update-data-kaggle-api/index.html#1", "459192": "I think Kaggle kernels work best with Chrome, so I am doing that. The only thing that happens to me once in a while is that changes do not get saved (see a round, red circle that is busy instead of quickly showing \"draft saved). I just keep an eye on that, so that I don't lose much if that happens. Have not had any failed runs lately.", "459285": "arigion Today, my kernel experience returns back to normal (i mean 'good') ... Maybe you can give our kernel a 2nd chance :)", "459310": "I don't understand the downvotes either. But frankly speaking I don't mind. I do not write posts for artificial medals but to talk about things that matter to me.", "459313": "Actually I use the Atom editor with the Hydrogen plugin (https://atom.io/packages/hydrogen). You can use it like a normal Python IDE or just execute single lines/blocks like in a Jupyter notebook with Hydrogen, but with git integration and refactoring. I don't like Jupyter Notebooks for actual models. They are great for exploring or testing stuff, but I don't really like them for coding. So my workflow is a Jupyter notebook for testing and writing reusable code into a library which I import into my models/notebooks.\nThat's one of the reasons why the kernel only competition annoys me, because I can't use my normal workflow and many tasks are so inconvenient.", "459377": "my kernel is stuck for more than an hour on a simple one lgbm model, that on my machine run in less than 7 minutes..;\ni'm a niubie, ok, but I doubt that most of real expert analysts can waste their time on this... \"only (inefficient) kernel competition\"...please...", "459460": "Yes, you're right, At least in this way everyone has access to the same infrastructure.", "459567": "I guess it's just for avoiding cheats or something", "459770": "i always work online, and it works flawless, except on iPad or old PC with small memory.", "459772": "i personally think when you have to create your submission online, the competition solution is more levelplaying for everybody and doesnot give a benefit to people having an extreme fast computer available. I don't think competitions have to work on data of 1 terabyte fe, to generate a solution that could be tested on 1Gb of data ?", "459848": "Whatever doesn't kill you simply makes you stranger.", "460090": "How about click the download button to keep your script?", "460709": "I sort of agree. But what's stopping you from using functions, IDE and version control? :) I use py files, not notebooks.", "460885": "This competition actually necessitates using a kernel based competition due to the nature of the data. Their dataset is derived from actual questions posted to Quora. The insincere questions are removed from their site since they violate their site guidelines. As a result one could get good results simply by cross referencing questions in the dataset and whether the question is still up on Quora's website. In addition, it helps reduce the complexity of solutions so it isn't some crazy blend of hundreds of models.", "461761": "I liked the idea of the kernel competition with the time limits etc and level setting what resources folks can use. however the kernel environment needs improvement. Too many hangs, errors, and just unstable experience made it frustrating. Hopefully these kernel only comps are used as a way to improve the kernel environment.", "462466": "We should commit the kernel and after a few hours, we need to go back to see if the code has been executed correctly.\nA complete waste of time!", "463405": "I wanted to jump in here and let everyone know that we very much appreciate this direct feedback. Our product and engineering teams see feedback like this and either work to address them in the short term, or prioritize them on a longer-term product roadmap. Community members who have been with Kaggle over the years know that we do listen and work hard to respond to the pain points our users experience. Stay tuned for some updates on specific issues. Thanks!", "463927": "There are some hiccups particularly when you are trying to commit code. Sometimes commit hangs and often it is very slow.", "464269": "Is there any **Plugin** for **Visual Studio Code** or **PyCharm** to edit the Jupyter Notebook locally, but run it remotely on Kaggle?", "464840": "You can always work offline using Jupyter Notebook and upload them later.\nBut I do had the same problems when I did try to commit a kernal with hyperparameter optimization, had to comment the optimization and hardcode these parameters later.", "465143": "While I do agree that this post has some valid points, kernel only competitions give people like me a chance to participate on a contest without needing to have access to state of the art hardware or (quite expensive) cloud compute.", "465192": "I am unable to select my kernel for final score. Does anyone get this?", "484190": "happyprancer \nCan you explain how you use `stickytape` ?\nI tried it, but it produces a lot of garbage into a single file...", "622792": "Could you install the GPU-edtion lightgbm please~~~\nI have to install it each time I open the notebook.\nThanks", "1200749": "I just came back after quitting for this exact reason." }, "source": "meta" }