Based on what I observe is expectations from Code only Competitions are to Provide similar compute limits to build models and make predictions , this kinds of brings much more simple and novel approaches .
But for this competition if we read Code Requirements it says external data freely and publicly available is allowed including pre trained models , As far I can understand this means we can't build our model elsewhere and incorporate that in our submission we have to build models in Kaggle Competition itself and make predictions .
So am I correct , do we have to build our model Kaggle Kernel environment and in the same kernel environment write code for making submission adhering to guidelines provided here , or do we need to build model elsewhere and incorporate in our submission !
", "messages": [ { "id": "1043794", "postDate": "10/09/2020 08:49:06", "content": "Based on what I observe is expectations from Code only Competitions are to Provide similar compute limits to build models and make predictions , this kinds of brings much more simple and novel approaches .
But for this competition if we read Code Requirements it says external data freely and publicly available is allowed including pre trained models , As far I can understand this means we can't build our model elsewhere and incorporate that in our submission we have to build models in Kaggle Competition itself and make predictions .
So am I correct , do we have to build our model Kaggle Kernel environment and in the same kernel environment write code for making submission adhering to guidelines provided here , or do we need to build model elsewhere and incorporate in our submission !
", "rawMarkdown": "1. Based on what I observe is expectations from Code only Competitions are to Provide similar compute limits to build models and make predictions , this kinds of brings much more simple and novel approaches .\n\n2. But for this competition if we read [Code Requirements](https://www.kaggle.com/c/riiid-test-answer-prediction/overview/code-requirements) it says external data freely and publicly available is allowed including pre trained models , As far I can understand this means we can't build our model elsewhere and incorporate that in our submission we have to build models in Kaggle Competition itself and make predictions . \n\nSo am I correct , do we have to build our model Kaggle Kernel environment and in the same kernel environment write code for making submission adhering to guidelines provided [here ](https://www.kaggle.com/sohier/competition-api-detailed-introduction) , or do we need to build model elsewhere and incorporate in our submission !", "votes": null }, { "id": "1043827", "postDate": "10/09/2020 09:19:38", "content": "\n\nwe can't build our model elsewhere and incorporate that in our submission
\n
This is not true. You can.
\nMany code competition notebooks use models built outside Kaggle.
@rohanrao I agree , For Example , PE Detection is an inference only code competition , where we have to build models outside of our final submission script and incorporate them in the final submission script as our final submission script won't have access to Training data.
\nI wanted to know whether for this competition we can do the same or not that is build models in separate kernels and in final submission script just load the weights of models , by uploading them as data source using .h5 format or pickle format and in final submission script just make predictions ! Can we do this ? I want to clear confusion on this !
You can do it in all competitions.
", "rawMarkdown": "You can do it in all competitions.", "votes": null }, { "id": "1043934", "postDate": "10/09/2020 10:48:17", "content": "Okay I got it cleared now @rohanrao the main constraint in code competitions is during prediction ! Prediction should follow Time constraints !
", "rawMarkdown": "Okay I got it cleared now @rohanrao the main constraint in code competitions is during prediction ! Prediction should follow Time constraints !", "votes": null } ], "comments": [ { "id": 1043827, "author_name": "rohanrao", "author_url": "", "post_date": "10/09/2020 09:19:38", "content": "\n\nwe can't build our model elsewhere and incorporate that in our submission
\n
This is not true. You can.
\nMany code competition notebooks use models built outside Kaggle.
@rohanrao I agree , For Example , PE Detection is an inference only code competition , where we have to build models outside of our final submission script and incorporate them in the final submission script as our final submission script won't have access to Training data.
\nI wanted to know whether for this competition we can do the same or not that is build models in separate kernels and in final submission script just load the weights of models , by uploading them as data source using .h5 format or pickle format and in final submission script just make predictions ! Can we do this ? I want to clear confusion on this !
You can do it in all competitions.
", "votes": null, "replies": [] }, { "id": 1043934, "author_name": "sayedathar11", "author_url": "", "post_date": "10/09/2020 10:48:17", "content": "Okay I got it cleared now @rohanrao the main constraint in code competitions is during prediction ! Prediction should follow Time constraints !
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1043794": "1. Based on what I observe is expectations from Code only Competitions are to Provide similar compute limits to build models and make predictions , this kinds of brings much more simple and novel approaches .\n\n2. But for this competition if we read [Code Requirements](https://www.kaggle.com/c/riiid-test-answer-prediction/overview/code-requirements) it says external data freely and publicly available is allowed including pre trained models , As far I can understand this means we can't build our model elsewhere and incorporate that in our submission we have to build models in Kaggle Competition itself and make predictions . \n\nSo am I correct , do we have to build our model Kaggle Kernel environment and in the same kernel environment write code for making submission adhering to guidelines provided [here ](https://www.kaggle.com/sohier/competition-api-detailed-introduction) , or do we need to build model elsewhere and incorporate in our submission !", "1043827": "> we can't build our model elsewhere and incorporate that in our submission\n\nThis is not true. You can.\nMany code competition notebooks use models built outside Kaggle.", "1043913": "rohanrao I agree , For Example ,[ PE Detection](https://www.kaggle.com/c/rsna-str-pulmonary-embolism-detection) is an inference only code competition , where we have to build models outside of our final submission script and incorporate them in the final submission script as our final submission script won't have access to Training data.\nI wanted to know whether for this competition we can do the same or not that is build models in separate kernels and in final submission script just load the weights of models , by uploading them as data source using .h5 format or pickle format and in final submission script just make predictions ! Can we do this ? I want to clear confusion on this !", "1043931": "You can do it in all competitions.", "1043934": "Okay I got it cleared now @rohanrao the main constraint in code competitions is during prediction ! Prediction should follow Time constraints !" }, "source": "meta" }