Whatever steps are needed would be good to know, or if it's not possible and only a full notebook is possible, with probably folder locations changed.
\nEdit: The code requirements say, \"Submissions to this competition must be made through Notebooks.\" so it seems notebooks are the only way.
", "messages": [ { "id": 1262910, "postDate": "2021-04-04T21:32:45.287Z", "content": "Whatever steps are needed would be good to know, or if it's not possible and only a full notebook is possible, with probably folder locations changed.
\nEdit: The code requirements say, \"Submissions to this competition must be made through Notebooks.\" so it seems notebooks are the only way.
", "rawMarkdown": "Whatever steps are needed would be good to know, or if it's not possible and only a full notebook is possible, with probably folder locations changed.\n\nEdit: The code requirements say, \"Submissions to this competition must be made through Notebooks.\" so it seems notebooks are the only way.", "votes": 2 }, { "id": 1263172, "postDate": "2021-04-05T06:34:56.307Z", "content": "You can create a final submissions CSV file and upload that.
\nThere is a sample submission file in the dataset, that is how the columns are needed to be formatted.
\nYes(if you are using python) just upload your notebook in kaggle then upload a submission.csv(from local system) as data then create a new cell and load submission.csv though pandas and then export using submission.to_csv()(index=False) in output folder . Then delete that cell , after that save your notebook along with the outputs . Then simply submit from the notebook. If you have different modules or requirements you can use pip freeze to generate a requirements.txt file . And add a cell as the first cell on the notebook and download the requirements.txt file with pip ( dont delete that cell).
\nNote :-
\nI am not saying retrain your model just run the cells specified
", "rawMarkdown": "Yes(if you are using python) just upload your notebook in kaggle then upload a submission.csv(from local system) as data then create a new cell and load submission.csv though pandas and then export using submission.to_csv()(index=False) in output folder . Then delete that cell , after that save your notebook along with the outputs . Then simply submit from the notebook. If you have different modules or requirements you can use pip freeze to generate a requirements.txt file . And add a cell as the first cell on the notebook and download the requirements.txt file with pip ( dont delete that cell).\n\nNote :-\n\nI am not saying retrain your model just run the cells specified", "replies": [ { "id": 1267621, "postDate": "2021-04-08T16:43:31.847Z", "content": "Yes you could train a model locally, then upload it to make predictions of their test files.
", "rawMarkdown": "Yes you could train a model locally, then upload it to make predictions of their test files." }, { "id": 1268041, "postDate": "2021-04-09T05:03:36.660Z", "content": "You can't upload the model as h5 file , all you can do is upload the model file as data and then call load model and write the code for submission.csv . It is completely okay it will work .
", "rawMarkdown": "You can't upload the model as h5 file , all you can do is upload the model file as data and then call load model and write the code for submission.csv . It is completely okay it will work ." } ] }, { "id": 1263708, "postDate": "2021-04-05T16:09:06.380Z", "content": "I see how to submit a csv, but to get the test images to test with, it says,
\ntest_images - The test set images. This competition has a hidden test set: only three images are provided here as samples while the remaining 5,000 images will be available to your notebook once it is submitted.
\nSo it sounds like one has to have a notebook that gets the test images from the same folder to test with, that they themselves would have to run. This might make sense since feasibly one could just go through all the test pictures and label them themselves (I don't know if this is a common practice on kaggle, and maybe people who run the competetion would want to see the code in any case). Or is there a way to get the test images? Because I submitted results for the 3 test images, but I don't see more test images.
\nIf the only way is to submit a whole notebook, then I guess running locally would just give a way to test out models before submitting, without using up kaggle time limits.
", "rawMarkdown": "I see how to submit a csv, but to get the test images to test with, it says, \n\ntest_images - The test set images. This competition has a hidden test set: only three images are provided here as samples while the remaining 5,000 images will be available to your notebook once it is submitted.\n\nSo it sounds like one has to have a notebook that gets the test images from the same folder to test with, that they themselves would have to run. This might make sense since feasibly one could just go through all the test pictures and label them themselves (I don't know if this is a common practice on kaggle, and maybe people who run the competetion would want to see the code in any case). Or is there a way to get the test images? Because I submitted results for the 3 test images, but I don't see more test images.\n\nIf the only way is to submit a whole notebook, then I guess running locally would just give a way to test out models before submitting, without using up kaggle time limits. " }, { "id": 1263040, "postDate": "2021-04-05T03:31:53.963Z", "content": "I think only the final csv file needs to be submitted.
", "rawMarkdown": "I think only the final csv file needs to be submitted. ", "replies": [ { "id": 1267477, "postDate": "2021-04-08T14:47:17.520Z", "content": "No , i think they test your model on 5000 images it cant calculate performance over only 3 images . So it takes your model and runs model.predict on 5000 images so it needs your model not submission.csv . You may also ask then why we need submission.csv well because you need submission.csv over the 5000 images and expects you write the code for that. This submission.csv is then scored.
", "rawMarkdown": "No , i think they test your model on 5000 images it cant calculate performance over only 3 images . So it takes your model and runs model.predict on 5000 images so it needs your model not submission.csv . You may also ask then why we need submission.csv well because you need submission.csv over the 5000 images and expects you write the code for that. This submission.csv is then scored.", "votes": 1 } ] }, { "id": 1263422, "postDate": "2021-04-05T12:01:59.293Z", "rawMarkdown": "", "votes": -1, "isDeleted": true } ], "comments": [ { "id": 1263172, "author_name": "Aravind Naidu", "author_url": "", "post_date": "2021-04-05T06:34:56.307000", "content": "You can create a final submissions CSV file and upload that.
\nThere is a sample submission file in the dataset, that is how the columns are needed to be formatted.
\nYes(if you are using python) just upload your notebook in kaggle then upload a submission.csv(from local system) as data then create a new cell and load submission.csv though pandas and then export using submission.to_csv()(index=False) in output folder . Then delete that cell , after that save your notebook along with the outputs . Then simply submit from the notebook. If you have different modules or requirements you can use pip freeze to generate a requirements.txt file . And add a cell as the first cell on the notebook and download the requirements.txt file with pip ( dont delete that cell).
\nNote :-
\nI am not saying retrain your model just run the cells specified
", "votes": 0, "replies": [ { "id": 1267621, "author_name": "Greg B", "author_url": "", "post_date": "2021-04-08T16:43:31.847000", "content": "Yes you could train a model locally, then upload it to make predictions of their test files.
", "votes": 0, "replies": [] }, { "id": 1268041, "author_name": "Sayantan Mazumdar", "author_url": "", "post_date": "2021-04-09T05:03:36.660000", "content": "You can't upload the model as h5 file , all you can do is upload the model file as data and then call load model and write the code for submission.csv . It is completely okay it will work .
", "votes": 0, "replies": [] } ] }, { "id": 1263708, "author_name": "Greg B", "author_url": "", "post_date": "2021-04-05T16:09:06.380000", "content": "I see how to submit a csv, but to get the test images to test with, it says,
\ntest_images - The test set images. This competition has a hidden test set: only three images are provided here as samples while the remaining 5,000 images will be available to your notebook once it is submitted.
\nSo it sounds like one has to have a notebook that gets the test images from the same folder to test with, that they themselves would have to run. This might make sense since feasibly one could just go through all the test pictures and label them themselves (I don't know if this is a common practice on kaggle, and maybe people who run the competetion would want to see the code in any case). Or is there a way to get the test images? Because I submitted results for the 3 test images, but I don't see more test images.
\nIf the only way is to submit a whole notebook, then I guess running locally would just give a way to test out models before submitting, without using up kaggle time limits.
", "votes": 0, "replies": [] }, { "id": 1263040, "author_name": "Roshan Raj", "author_url": "", "post_date": "2021-04-05T03:31:53.963000", "content": "I think only the final csv file needs to be submitted.
", "votes": 0, "replies": [ { "id": 1267477, "author_name": "Sayantan Mazumdar", "author_url": "", "post_date": "2021-04-08T14:47:17.520000", "content": "No , i think they test your model on 5000 images it cant calculate performance over only 3 images . So it takes your model and runs model.predict on 5000 images so it needs your model not submission.csv . You may also ask then why we need submission.csv well because you need submission.csv over the 5000 images and expects you write the code for that. This submission.csv is then scored.
", "votes": 1, "replies": [] } ] }, { "id": 1263422, "author_name": "", "author_url": "", "post_date": "2021-04-05T12:01:59.293000", "content": "", "votes": -1, "replies": [] } ], "raw_markdown_by_id": { "1262910": "Whatever steps are needed would be good to know, or if it's not possible and only a full notebook is possible, with probably folder locations changed.\n\nEdit: The code requirements say, \"Submissions to this competition must be made through Notebooks.\" so it seems notebooks are the only way.", "1263172": "You can create a final submissions CSV file and upload that.\n\nThere is a sample submission file in the dataset, that is how the columns are needed to be formatted. \n1. Make a data frame in that format.\n2. Then save it to a CSV file with the name of your choice in a local location. \nwith : df.to_csv('path/name.csv')\n3. Then upload the same in the submissions space.", "1267489": "Yes(if you are using python) just upload your notebook in kaggle then upload a submission.csv(from local system) as data then create a new cell and load submission.csv though pandas and then export using submission.to_csv()(index=False) in output folder . Then delete that cell , after that save your notebook along with the outputs . Then simply submit from the notebook. If you have different modules or requirements you can use pip freeze to generate a requirements.txt file . And add a cell as the first cell on the notebook and download the requirements.txt file with pip ( dont delete that cell).\n\nNote :-\n\nI am not saying retrain your model just run the cells specified", "1263708": "I see how to submit a csv, but to get the test images to test with, it says, \n\ntest_images - The test set images. This competition has a hidden test set: only three images are provided here as samples while the remaining 5,000 images will be available to your notebook once it is submitted.\n\nSo it sounds like one has to have a notebook that gets the test images from the same folder to test with, that they themselves would have to run. This might make sense since feasibly one could just go through all the test pictures and label them themselves (I don't know if this is a common practice on kaggle, and maybe people who run the competetion would want to see the code in any case). Or is there a way to get the test images? Because I submitted results for the 3 test images, but I don't see more test images.\n\nIf the only way is to submit a whole notebook, then I guess running locally would just give a way to test out models before submitting, without using up kaggle time limits. ", "1263040": "I think only the final csv file needs to be submitted. ", "1263422": "" } }