Hi , I am new in Kaggle and this is my first time to join Kaggle Competition
\nI have some confuses about the submission.csv
\nThe image folders a & b in test folder are the predictions for submission?
\nIf they are , there are totally 65 images in each subfolders and how do we present the final predictions in submission.csv ?
\nAny senior can provide me some advices or suggestions or samples?
\nThks a lot !!
For both old and new folks the submission process using notebooks can be confusing.
\nThe test data we can see is not the real thing. It's a reasonable bit of data we can use to make sure our prediction code works. It may or may not represent the actual structure of the test data.
\nDON\"T hard code for only 'a' and 'b'.
\nDON'T assume only 65.
At some point in the near future the old pros will have probed the test data and can answer your question with a bit more clarity.
", "rawMarkdown": "For both old and new folks the submission process using notebooks can be confusing.\n\nThe test data we can see is not the real thing. It's a reasonable bit of data we can use to make sure our prediction code works. It may or may not represent the actual structure of the test data.\n\nDON\"T hard code for only 'a' and 'b'.\nDON'T assume only 65.\n\nAt some point in the near future the old pros will have probed the test data and can answer your question with a bit more clarity.", "votes": 2, "replies": [ { "id": 2193033, "postDate": "2023-03-23T04:09:19.830Z", "content": "so that mean we need to submit notebook file (with prepocessing / training / prediction code) but not submit submission.csv ?
\nI have no idea that what what kinds of 'output file ' that we need to submit ?
See some of the shared notebooks that do submissions. In general for this competition you will likely need to train a model in a notebook. Save the model. You will need to do this as the training and prediction time will both run out the clock on your time limit for any model that's going to generate a decent prediction.
\nIn another notebook (or the same notebook) load the model, make predictions on the test data and generate a submission.csv file. This notebook will be run when you 'submit'. There are a couple of ways to 'submit' so it gets a tiny bit confusing.
\nThe 'old' why was to run the prediction notebook, save it and generate a submission.csv file. Than later select the 'submission.csv' file and \"Submit\". The new way is to have the prediction notebook in edit mode and on the right side find the 'Submit'. I am finding the new way a bit confusing as I seem to be leaving the notebook running and burning up my GPU quota.
", "rawMarkdown": "See some of the shared notebooks that do submissions. In general for this competition you will likely need to train a model in a notebook. Save the model. You will need to do this as the training and prediction time will both run out the clock on your time limit for any model that's going to generate a decent prediction.\n\nIn another notebook (or the same notebook) load the model, make predictions on the test data and generate a submission.csv file. This notebook will be run when you 'submit'. There are a couple of ways to 'submit' so it gets a tiny bit confusing.\n\nThe 'old' why was to run the prediction notebook, save it and generate a submission.csv file. Than later select the 'submission.csv' file and \"Submit\". The new way is to have the prediction notebook in edit mode and on the right side find the 'Submit'. I am finding the new way a bit confusing as I seem to be leaving the notebook running and burning up my GPU quota.\n\n" }, { "id": 2193349, "postDate": "2023-03-23T08:24:40.703Z", "content": "i see ! so that mean i have to create a notebook with some codes in this competition which included loading data , prepocessing , training , save and load the trained-weights , predictions and create a submission.csv .
\nis it correct ?
The notebook that generates the submission does not have to include training. I will train any models I create on my local computers. My kaggle notebook will load a trained model and do any data preprocessing and predictions.
", "rawMarkdown": "The notebook that generates the submission does not have to include training. I will train any models I create on my local computers. My kaggle notebook will load a trained model and do any data preprocessing and predictions." }, { "id": 2194462, "postDate": "2023-03-24T00:08:33.147Z", "content": "i see !!! Thanks for your clear explanations to let me understand !!
", "rawMarkdown": "i see !!! Thanks for your clear explanations to let me understand !! " }, { "id": 2204074, "postDate": "2023-03-31T11:06:07.133Z", "content": "Hi there, I found this discussion about submission issues which I think may be the right place to ask.
\nI tried to use the example submission notebook as a first apprach. I generated a model.pt file in training mode and then tried to submit the notebook loading the trained module instead of re-training the model. In a first attempt, the evaluation process of that notebook took too long, so the process failed. I then modified the notebook to send a dummy version of predictions, but then, a new error emerged. When trying to load the model from \"/kaggle/working\" it doesn't find the model.pt file. This is weird, since I can see and list the file while running the notebook. Any idea why this may be happening? Thanks a lot! :)
If you look at the past competitions that use this type of submission process there will be lots of folks asking similar questions. I have found that putting the saved model into a kaggle dataset is the cleanest way. The alternative of adding it to a new notebook by taking it from another notebook seems to get a little messy for me when I start training a bunch of different models and want to have them saved and handy.
\nWithout seeing your actual code I am not good enough to be of much help. The example submission notebook probably not a real good one to try as both the training and the submission processes run for very long times. For the notebook you must train a model, save it and than load it in a notebook and do the predictions. So I would try one of the shared notebooks that have a score, they train and generate submissions at must quicker pace.
\nIf you really like your code, than the best help occurs when you make the code public and open up a new discussion topic asking for some help.
", "rawMarkdown": "If you look at the past competitions that use this type of submission process there will be lots of folks asking similar questions. I have found that putting the saved model into a kaggle dataset is the cleanest way. The alternative of adding it to a new notebook by taking it from another notebook seems to get a little messy for me when I start training a bunch of different models and want to have them saved and handy.\n\nWithout seeing your actual code I am not good enough to be of much help. The example submission notebook probably not a real good one to try as both the training and the submission processes run for very long times. For the notebook you must train a model, save it and than load it in a notebook and do the predictions. So I would try one of the shared notebooks that have a score, they train and generate submissions at must quicker pace.\n\nIf you really like your code, than the best help occurs when you make the code public and open up a new discussion topic asking for some help.", "votes": 1 }, { "id": 2204605, "postDate": "2023-03-31T17:52:25.243Z", "content": "Thanks for the replay. I don't have actual code yet, I just tried to start from the suggested example submision to learn the process. I'll check how to load dataset in a new notebook. But, what is the path you have to provide? The eror I got was because I assumed that the model I had saved was available in /kaggle/working dir but when the submission process re-run the notebook, it didn't find the model there.
", "rawMarkdown": "Thanks for the replay. I don't have actual code yet, I just tried to start from the suggested example submision to learn the process. I'll check how to load dataset in a new notebook. But, what is the path you have to provide? The eror I got was because I assumed that the model I had saved was available in /kaggle/working dir but when the submission process re-run the notebook, it didn't find the model there." }, { "id": 2204753, "postDate": "2023-03-31T22:34:54.697Z", "content": "When you train a model and save it, as you notice it's not going to be there if you re-run the notebook with the intention of just prediction/submission.
\nYou need to save the model to a local file and than upload to a Datasets or save it directly to a Datasets and than Add Data in the prediction model. Once it's added you can copy & paste the address into the prediction notebook.
\nIts not real clean and it's a real pain to figure it out. I think there are three keys to learning ML. I think #1 is the easiest of the bunch.
\n1) learn to code.
\n2) on a local machine learn to install and maintain Python
\n3) Learn how to navigate kaggle.
I train all my models on local machine, so I create a model and upload it to a 'datasets' and than 'add data' to the prediction model. If I happen to train the model on kaggle, I go to the trouble of downloading it to my local machine and than create a Datasets and upload. I have found over the 4 years+ that I have been on the site they often change the \"easy way to add a model from a notebook\" - I got tired of learning the new way each time they made it 'easier'.
\nI just checked YouTube and don't see a fresh video on how to do any of these things.
", "rawMarkdown": "When you train a model and save it, as you notice it's not going to be there if you re-run the notebook with the intention of just prediction/submission.\n\nYou need to save the model to a local file and than upload to a Datasets or save it directly to a Datasets and than Add Data in the prediction model. Once it's added you can copy & paste the address into the prediction notebook.\n\nIts not real clean and it's a real pain to figure it out. I think there are three keys to learning ML. I think #1 is the easiest of the bunch.\n1) learn to code. \n2) on a local machine learn to install and maintain Python \n3) Learn how to navigate kaggle.\n\nI train all my models on local machine, so I create a model and upload it to a 'datasets' and than 'add data' to the prediction model. If I happen to train the model on kaggle, I go to the trouble of downloading it to my local machine and than create a Datasets and upload. I have found over the 4 years+ that I have been on the site they often change the \"easy way to add a model from a notebook\" - I got tired of learning the new way each time they made it 'easier'. \n\nI just checked YouTube and don't see a fresh video on how to do any of these things.\n", "votes": 1 }, { "id": 2205074, "postDate": "2023-04-01T08:15:55.150Z", "content": "Thanks again! that makes sense. I'll try the dataset solution. Points 1 and 2 are fine, just new to kaggle😊
\ncheers!
Hi there. I never used this example notebook for submission. It definitely took too long to evaluate. I used other public example with the caution to save the trained model in a local file and upload it as a dataset as suggested above. Check the log of the submission to try to gain some information on why it failed. One possible cause is it went beyond the 9h gpu limit. Sorry I can't help any further.
", "rawMarkdown": "Hi there. I never used this example notebook for submission. It definitely took too long to evaluate. I used other public example with the caution to save the trained model in a local file and upload it as a dataset as suggested above. Check the log of the submission to try to gain some information on why it failed. One possible cause is it went beyond the 9h gpu limit. Sorry I can't help any further." } ] } ] }, { "id": 2192952, "postDate": "2023-03-23T02:16:45.627Z", "content": "Hi , I am new in Kaggle and this is my first time to join Kaggle Competition
\nI have some confuses about the submission.csv
\nThe image folders a & b in test folder are the predictions for submission?
\nIf they are , there are totally 65 images in each subfolders and how do we present the final predictions in submission.csv ?
\nAny senior can provide me some advices or suggestions or samples?
\nThks a lot !!
For both old and new folks the submission process using notebooks can be confusing.
\nThe test data we can see is not the real thing. It's a reasonable bit of data we can use to make sure our prediction code works. It may or may not represent the actual structure of the test data.
\nDON\"T hard code for only 'a' and 'b'.
\nDON'T assume only 65.
At some point in the near future the old pros will have probed the test data and can answer your question with a bit more clarity.
", "votes": 2, "replies": [ { "id": 2193033, "author_name": "chusheng chen", "author_url": "", "post_date": "2023-03-23T04:09:19.830000", "content": "so that mean we need to submit notebook file (with prepocessing / training / prediction code) but not submit submission.csv ?
\nI have no idea that what what kinds of 'output file ' that we need to submit ?
See some of the shared notebooks that do submissions. In general for this competition you will likely need to train a model in a notebook. Save the model. You will need to do this as the training and prediction time will both run out the clock on your time limit for any model that's going to generate a decent prediction.
\nIn another notebook (or the same notebook) load the model, make predictions on the test data and generate a submission.csv file. This notebook will be run when you 'submit'. There are a couple of ways to 'submit' so it gets a tiny bit confusing.
\nThe 'old' why was to run the prediction notebook, save it and generate a submission.csv file. Than later select the 'submission.csv' file and \"Submit\". The new way is to have the prediction notebook in edit mode and on the right side find the 'Submit'. I am finding the new way a bit confusing as I seem to be leaving the notebook running and burning up my GPU quota.
", "votes": 0, "replies": [] }, { "id": 2193349, "author_name": "chusheng chen", "author_url": "", "post_date": "2023-03-23T08:24:40.703000", "content": "i see ! so that mean i have to create a notebook with some codes in this competition which included loading data , prepocessing , training , save and load the trained-weights , predictions and create a submission.csv .
\nis it correct ?
The notebook that generates the submission does not have to include training. I will train any models I create on my local computers. My kaggle notebook will load a trained model and do any data preprocessing and predictions.
", "votes": 0, "replies": [] }, { "id": 2194462, "author_name": "chusheng chen", "author_url": "", "post_date": "2023-03-24T00:08:33.147000", "content": "i see !!! Thanks for your clear explanations to let me understand !!
", "votes": 0, "replies": [] }, { "id": 2204074, "author_name": "cpobes", "author_url": "", "post_date": "2023-03-31T11:06:07.133000", "content": "Hi there, I found this discussion about submission issues which I think may be the right place to ask.
\nI tried to use the example submission notebook as a first apprach. I generated a model.pt file in training mode and then tried to submit the notebook loading the trained module instead of re-training the model. In a first attempt, the evaluation process of that notebook took too long, so the process failed. I then modified the notebook to send a dummy version of predictions, but then, a new error emerged. When trying to load the model from \"/kaggle/working\" it doesn't find the model.pt file. This is weird, since I can see and list the file while running the notebook. Any idea why this may be happening? Thanks a lot! :)
If you look at the past competitions that use this type of submission process there will be lots of folks asking similar questions. I have found that putting the saved model into a kaggle dataset is the cleanest way. The alternative of adding it to a new notebook by taking it from another notebook seems to get a little messy for me when I start training a bunch of different models and want to have them saved and handy.
\nWithout seeing your actual code I am not good enough to be of much help. The example submission notebook probably not a real good one to try as both the training and the submission processes run for very long times. For the notebook you must train a model, save it and than load it in a notebook and do the predictions. So I would try one of the shared notebooks that have a score, they train and generate submissions at must quicker pace.
\nIf you really like your code, than the best help occurs when you make the code public and open up a new discussion topic asking for some help.
", "votes": 1, "replies": [] }, { "id": 2204605, "author_name": "cpobes", "author_url": "", "post_date": "2023-03-31T17:52:25.243000", "content": "Thanks for the replay. I don't have actual code yet, I just tried to start from the suggested example submision to learn the process. I'll check how to load dataset in a new notebook. But, what is the path you have to provide? The eror I got was because I assumed that the model I had saved was available in /kaggle/working dir but when the submission process re-run the notebook, it didn't find the model there.
", "votes": 0, "replies": [] }, { "id": 2204753, "author_name": "PC Jimmmy", "author_url": "", "post_date": "2023-03-31T22:34:54.697000", "content": "When you train a model and save it, as you notice it's not going to be there if you re-run the notebook with the intention of just prediction/submission.
\nYou need to save the model to a local file and than upload to a Datasets or save it directly to a Datasets and than Add Data in the prediction model. Once it's added you can copy & paste the address into the prediction notebook.
\nIts not real clean and it's a real pain to figure it out. I think there are three keys to learning ML. I think #1 is the easiest of the bunch.
\n1) learn to code.
\n2) on a local machine learn to install and maintain Python
\n3) Learn how to navigate kaggle.
I train all my models on local machine, so I create a model and upload it to a 'datasets' and than 'add data' to the prediction model. If I happen to train the model on kaggle, I go to the trouble of downloading it to my local machine and than create a Datasets and upload. I have found over the 4 years+ that I have been on the site they often change the \"easy way to add a model from a notebook\" - I got tired of learning the new way each time they made it 'easier'.
\nI just checked YouTube and don't see a fresh video on how to do any of these things.
", "votes": 1, "replies": [] }, { "id": 2205074, "author_name": "cpobes", "author_url": "", "post_date": "2023-04-01T08:15:55.150000", "content": "Thanks again! that makes sense. I'll try the dataset solution. Points 1 and 2 are fine, just new to kaggle😊
\ncheers!
Hi there. I never used this example notebook for submission. It definitely took too long to evaluate. I used other public example with the caution to save the trained model in a local file and upload it as a dataset as suggested above. Check the log of the submission to try to gain some information on why it failed. One possible cause is it went beyond the 9h gpu limit. Sorry I can't help any further.
", "votes": 0, "replies": [] } ] } ] }, { "id": 2193031, "author_name": "", "author_url": "", "post_date": "2023-03-23T04:09:03.203000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "2193009": "For both old and new folks the submission process using notebooks can be confusing.\n\nThe test data we can see is not the real thing. It's a reasonable bit of data we can use to make sure our prediction code works. It may or may not represent the actual structure of the test data.\n\nDON\"T hard code for only 'a' and 'b'.\nDON'T assume only 65.\n\nAt some point in the near future the old pros will have probed the test data and can answer your question with a bit more clarity.", "2192952": "Hi , I am new in Kaggle and this is my first time to join Kaggle Competition\nI have some confuses about the submission.csv\nThe image folders a & b in test folder are the predictions for submission?\nIf they are , there are totally 65 images in each subfolders and how do we present the final predictions in submission.csv ?\nAny senior can provide me some advices or suggestions or samples?\nThks a lot !! ", "2193031": "" } }