{ "id": 336819, "title": "I became the first person to cross the sample submission.csv", "url": "/competitions/mayo-clinic-strip-ai/discussion/336819", "author_name": "chumajin", "post_date": "2022-07-13T03:07:23.919000", "votes": 43, "comment_count": 9, "views": 0, "content": "

I did it! It took me 4 days to pass this submission…..
\n(It's been a week since this competition opened, but we didn't get a better score than the sample submission.)

\n

\"\"

\n

I share what I noticed.

\n

1. Memory

\n

If you don't use gpu, you have 16 Gb of RAM, but if you use gpu, it's reduced to 13 Gb on the kaggle notebook.

\n

This is one reason why you can't read large images when submitting.
\n(Many people make datasets with gpu turned off)

\n

You will need to use something like that are already discussed on the topics or public notebooks.
\nhttps://www.kaggle.com/competitions/mayo-clinic-strip-ai/discussion/335976

\n

2. patient id

\n

The sample submission has only the predicted value of the patient id.

\n

Since there may be multiple images for each patient id in test.csv, it is necessary to take the average with groupby etc. before submitting.

", "messages": [ { "id": 1853637, "postDate": "2022-07-13T03:07:23.920Z", "content": "

I did it! It took me 4 days to pass this submission…..
\n(It's been a week since this competition opened, but we didn't get a better score than the sample submission.)

\n

\"\"

\n

I share what I noticed.

\n

1. Memory

\n

If you don't use gpu, you have 16 Gb of RAM, but if you use gpu, it's reduced to 13 Gb on the kaggle notebook.

\n

This is one reason why you can't read large images when submitting.
\n(Many people make datasets with gpu turned off)

\n

You will need to use something like that are already discussed on the topics or public notebooks.
\nhttps://www.kaggle.com/competitions/mayo-clinic-strip-ai/discussion/335976

\n

2. patient id

\n

The sample submission has only the predicted value of the patient id.

\n

Since there may be multiple images for each patient id in test.csv, it is necessary to take the average with groupby etc. before submitting.

", "rawMarkdown": "I did it! It took me 4 days to pass this submission.....\n(It's been a week since this competition opened, but we didn't get a better score than the sample submission.)\n\n![](https://raw.githubusercontent.com/chumajin/mayo/main/mayo.jpg)\n\nI share what I noticed.\n\n# 1. Memory\n\nIf you don't use gpu, you have 16 Gb of RAM, but if you use gpu, it's reduced to 13 Gb on the kaggle notebook.\n\nThis is one reason why you can't read large images when submitting.\n(Many people make datasets with gpu turned off)\n\nYou will need to use something like that are already discussed on the topics or public notebooks.\nhttps://www.kaggle.com/competitions/mayo-clinic-strip-ai/discussion/335976\n\n# 2. patient id\n\nThe sample submission has only the predicted value of the patient id.\n\nSince there may be multiple images for each patient id in test.csv, it is necessary to take the average with groupby etc. before submitting.", "votes": 42 }, { "id": 1856142, "postDate": "2022-07-15T06:37:38.543Z", "content": "

Great work @chumajin, many congratulations !

", "rawMarkdown": "Great work @chumajin, many congratulations !", "votes": 1, "replies": [ { "id": 1863057, "postDate": "2022-07-20T06:19:48.860Z", "content": "

@saurabhbagchi Thank you for comment ! I remember you.

", "rawMarkdown": "@saurabhbagchi Thank you for comment ! I remember you.", "votes": 1 } ] }, { "id": 1855741, "postDate": "2022-07-14T21:06:57.650Z", "content": "

very interesting about the RAM shrinking for GPU instance was suspicious about something similar but did not try yet as there is only one submission per day :D
\nabout the multiple parents IT, ai have done a simple average:
\ndf_pred = df_pred.groupby(\"patient_id\").mean()

\n
\n

ref: https://www.kaggle.com/code/jirkaborovec/bloodclots-classif-lightning-flash-inference

\n
", "rawMarkdown": "very interesting about the RAM shrinking for GPU instance was suspicious about something similar but did not try yet as there is only one submission per day :D\nabout the multiple parents IT, ai have done a simple average:\n`df_pred = df_pred.groupby(\"patient_id\").mean()`\n>ref: https://www.kaggle.com/code/jirkaborovec/bloodclots-classif-lightning-flash-inference", "votes": 1, "replies": [ { "id": 1863059, "postDate": "2022-07-20T06:20:33.657Z", "content": "

@jirkaborovec Thank you for comment and example codes!!

", "rawMarkdown": "@jirkaborovec Thank you for comment and example codes!!", "votes": 1 } ] }, { "id": 1858448, "postDate": "2022-07-16T23:32:27.943Z", "content": "

It's amazing dear friend. Congrats'

\n

\"You're a prize contender\" \"Keep trying\" (kaggle programm words :)

", "rawMarkdown": "It's amazing dear friend. Congrats'\n\n\"You're a prize contender\" \"Keep trying\" (kaggle programm words :)", "votes": 2, "replies": [ { "id": 1863056, "postDate": "2022-07-20T06:18:44.070Z", "content": "

@mpwolke Hi bro. Thank you for kaggle algorithm comment 😄

", "rawMarkdown": "@mpwolke Hi bro. Thank you for kaggle algorithm comment 😄", "votes": 2 } ] }, { "id": 1853669, "postDate": "2022-07-13T03:39:01.540Z", "content": "

Did you not experience overfitting on train set, when I did my submit, although CV was .58, It scored 1.7 on the LB, did you clip your submission to avoid massive score punishment in this submission?

", "rawMarkdown": "Did you not experience overfitting on train set, when I did my submit, although CV was .58, It scored 1.7 on the LB, did you clip your submission to avoid massive score punishment in this submission?", "votes": 2, "replies": [ { "id": 1853725, "postDate": "2022-07-13T05:20:09.830Z", "content": "

@harshitsheoran Thank you for comment. I just used softmax.

", "rawMarkdown": "@harshitsheoran Thank you for comment. I just used softmax. ", "votes": 1 } ] }, { "id": 1961100, "postDate": "2022-09-29T02:24:14.140Z", "content": "

Thank you for your amazing work!

", "rawMarkdown": "Thank you for your amazing work!" } ], "comments": [ { "id": 1856142, "author_name": "Old Monk", "author_url": "", "post_date": "2022-07-15T06:37:38.543000", "content": "

Great work @chumajin, many congratulations !

", "votes": 1, "replies": [ { "id": 1863057, "author_name": "chumajin", "author_url": "", "post_date": "2022-07-20T06:19:48.860000", "content": "

@saurabhbagchi Thank you for comment ! I remember you.

", "votes": 1, "replies": [] } ] }, { "id": 1855741, "author_name": "Jirka", "author_url": "", "post_date": "2022-07-14T21:06:57.650000", "content": "

very interesting about the RAM shrinking for GPU instance was suspicious about something similar but did not try yet as there is only one submission per day :D
\nabout the multiple parents IT, ai have done a simple average:
\ndf_pred = df_pred.groupby(\"patient_id\").mean()

\n
\n

ref: https://www.kaggle.com/code/jirkaborovec/bloodclots-classif-lightning-flash-inference

\n
", "votes": 1, "replies": [ { "id": 1863059, "author_name": "chumajin", "author_url": "", "post_date": "2022-07-20T06:20:33.657000", "content": "

@jirkaborovec Thank you for comment and example codes!!

", "votes": 1, "replies": [] } ] }, { "id": 1858448, "author_name": "Marília Prata", "author_url": "", "post_date": "2022-07-16T23:32:27.943000", "content": "

It's amazing dear friend. Congrats'

\n

\"You're a prize contender\" \"Keep trying\" (kaggle programm words :)

", "votes": 2, "replies": [ { "id": 1863056, "author_name": "chumajin", "author_url": "", "post_date": "2022-07-20T06:18:44.070000", "content": "

@mpwolke Hi bro. Thank you for kaggle algorithm comment 😄

", "votes": 2, "replies": [] } ] }, { "id": 1853669, "author_name": "Harshit Sheoran", "author_url": "", "post_date": "2022-07-13T03:39:01.540000", "content": "

Did you not experience overfitting on train set, when I did my submit, although CV was .58, It scored 1.7 on the LB, did you clip your submission to avoid massive score punishment in this submission?

", "votes": 2, "replies": [ { "id": 1853725, "author_name": "chumajin", "author_url": "", "post_date": "2022-07-13T05:20:09.830000", "content": "

@harshitsheoran Thank you for comment. I just used softmax.

", "votes": 1, "replies": [] } ] }, { "id": 1961100, "author_name": "Paceiver", "author_url": "", "post_date": "2022-09-29T02:24:14.140000", "content": "

Thank you for your amazing work!

", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1853637": "I did it! It took me 4 days to pass this submission.....\n(It's been a week since this competition opened, but we didn't get a better score than the sample submission.)\n\n![](https://raw.githubusercontent.com/chumajin/mayo/main/mayo.jpg)\n\nI share what I noticed.\n\n# 1. Memory\n\nIf you don't use gpu, you have 16 Gb of RAM, but if you use gpu, it's reduced to 13 Gb on the kaggle notebook.\n\nThis is one reason why you can't read large images when submitting.\n(Many people make datasets with gpu turned off)\n\nYou will need to use something like that are already discussed on the topics or public notebooks.\nhttps://www.kaggle.com/competitions/mayo-clinic-strip-ai/discussion/335976\n\n# 2. patient id\n\nThe sample submission has only the predicted value of the patient id.\n\nSince there may be multiple images for each patient id in test.csv, it is necessary to take the average with groupby etc. before submitting.", "1856142": "Great work @chumajin, many congratulations !", "1855741": "very interesting about the RAM shrinking for GPU instance was suspicious about something similar but did not try yet as there is only one submission per day :D\nabout the multiple parents IT, ai have done a simple average:\n`df_pred = df_pred.groupby(\"patient_id\").mean()`\n>ref: https://www.kaggle.com/code/jirkaborovec/bloodclots-classif-lightning-flash-inference", "1858448": "It's amazing dear friend. Congrats'\n\n\"You're a prize contender\" \"Keep trying\" (kaggle programm words :)", "1853669": "Did you not experience overfitting on train set, when I did my submit, although CV was .58, It scored 1.7 on the LB, did you clip your submission to avoid massive score punishment in this submission?", "1961100": "Thank you for your amazing work!" } }