\nhttps://www.kaggle.com/code/andradaolteanu/rsna-breast-cancer-eda-pytorch-baseline
\n
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4392121%2F5a527e4e90338e488152ce30ad95ab70%2FChkZ53p.jpg?generation=1676649585421557&alt=media\")
As a beginner, I’d appreciate any guidance on how an attribute such as machine_id (or laterality) can be incorporated into an image classifier. My googling on the topic is raising more questions than answers
\nIf the answer is to use multi-label classification such as “Y 49” to indicate a cancer diagnosis from machine 49, then how do you get a prediction that only considers the cancer diagnosis? Similarly how do you get a prediction that only considers the cancer diagnosis if a single label classification is used?
\nHow should you proceed if you take an ensemble approach, create a separate model for each machine_id and store the trained learned objects in a list? (This seems to be the most logical approach to me)
\nIs it possible to incorporate some form of control structure such as “ if … then” or “select case” so that a particular trained learner object is used depending on the machine_id?
\nMy WIP notebook is available at https://www.kaggle.com/code/julianmacnamara/rsna-breast-cancer-detection-pttoa-v1-2
\nMany thanks in advance
\nAll the best
\nJulian
", "messages": [ { "id": 2148211, "postDate": "2023-02-17T08:39:37.780Z", "content": "As a beginner, I’d appreciate any guidance on how an attribute such as machine_id (or laterality) can be incorporated into an image classifier. My googling on the topic is raising more questions than answers
\nIf the answer is to use multi-label classification such as “Y 49” to indicate a cancer diagnosis from machine 49, then how do you get a prediction that only considers the cancer diagnosis? Similarly how do you get a prediction that only considers the cancer diagnosis if a single label classification is used?
\nHow should you proceed if you take an ensemble approach, create a separate model for each machine_id and store the trained learned objects in a list? (This seems to be the most logical approach to me)
\nIs it possible to incorporate some form of control structure such as “ if … then” or “select case” so that a particular trained learner object is used depending on the machine_id?
\nMy WIP notebook is available at https://www.kaggle.com/code/julianmacnamara/rsna-breast-cancer-detection-pttoa-v1-2
\nMany thanks in advance
\nAll the best
\nJulian
", "rawMarkdown": "As a beginner, I’d appreciate any guidance on how an attribute such as machine_id (or laterality) can be incorporated into an image classifier. My googling on the topic is raising more questions than answers\n\nIf the answer is to use multi-label classification such as “Y 49” to indicate a cancer diagnosis from machine 49, then how do you get a prediction that only considers the cancer diagnosis? Similarly how do you get a prediction that only considers the cancer diagnosis if a single label classification is used?\n\nHow should you proceed if you take an ensemble approach, create a separate model for each machine_id and store the trained learned objects in a list? (This seems to be the most logical approach to me)\n\nIs it possible to incorporate some form of control structure such as “ if ... then” or “select case” so that a particular trained learner object is used depending on the machine_id?\n\nMy WIP notebook is available at https://www.kaggle.com/code/julianmacnamara/rsna-breast-cancer-detection-pttoa-v1-2\n\nMany thanks in advance\n\nAll the best\n\nJulian\n", "votes": 3 }, { "id": 2148645, "postDate": "2023-02-17T15:39:21.877Z", "content": "Referring to my question on taking an ensemble approach, this seems to work
\nHere is an inference example considering several models/nets.
", "rawMarkdown": "[Here](https://www.kaggle.com/code/hengck23/3hr-tensorrt-nextvit-example?scriptVersionId=116312479&cellId=12) is an inference example considering several models/nets.", "votes": 1, "replies": [ { "id": 2148648, "postDate": "2023-02-17T15:42:10.490Z", "content": "Many thanks Antti. I'll check it out. I've also posted a solution that seems to show promise
", "rawMarkdown": "Many thanks Antti. I'll check it out. I've also posted a solution that seems to show promise" } ] }, { "id": 2149731, "postDate": "2023-02-18T16:11:49.753Z", "content": "Hi Julian, I found this notebook useful.
", "rawMarkdown": "Hi Julian, I found this [notebook](https://www.kaggle.com/code/awsaf49/rsna-bcd-efficientnet-tf-tpu-1vm-train) useful. ", "replies": [ { "id": 2150449, "postDate": "2023-02-19T09:06:23.140Z", "content": "Hi
\nIt is both interesting and helpful
\nMany thanks for sharing it with me
\nAll the best
\nJulian
One way is described in this notebook:
\nhttps://www.kaggle.com/code/andradaolteanu/rsna-breast-cancer-eda-pytorch-baseline
\n
Hello Resoul
\nVery interesting
\nMany thanks
\nAll the best
\nJulian
Perhaps this will be helpful to implement auxiliary outputs?
", "rawMarkdown": "Perhaps [this](https://www.kaggle.com/code/vslaykovsky/infer-pytorch-aux-targets-weighted-loss-thres?scriptVersionId=113756289&cellId=15) will be helpful to implement auxiliary outputs?" } ], "comments": [ { "id": 2148645, "author_name": "Julian Macnamara", "author_url": "", "post_date": "2023-02-17T15:39:21.877000", "content": "Referring to my question on taking an ensemble approach, this seems to work
\nHere is an inference example considering several models/nets.
", "votes": 1, "replies": [ { "id": 2148648, "author_name": "Julian Macnamara", "author_url": "", "post_date": "2023-02-17T15:42:10.490000", "content": "Many thanks Antti. I'll check it out. I've also posted a solution that seems to show promise
", "votes": 0, "replies": [] } ] }, { "id": 2149731, "author_name": "bmjdoherty", "author_url": "", "post_date": "2023-02-18T16:11:49.753000", "content": "Hi Julian, I found this notebook useful.
", "votes": 0, "replies": [ { "id": 2150449, "author_name": "Julian Macnamara", "author_url": "", "post_date": "2023-02-19T09:06:23.140000", "content": "Hi
\nIt is both interesting and helpful
\nMany thanks for sharing it with me
\nAll the best
\nJulian
One way is described in this notebook:
\nhttps://www.kaggle.com/code/andradaolteanu/rsna-breast-cancer-eda-pytorch-baseline
\n
Hello Resoul
\nVery interesting
\nMany thanks
\nAll the best
\nJulian
Perhaps this will be helpful to implement auxiliary outputs?
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "2148211": "As a beginner, I’d appreciate any guidance on how an attribute such as machine_id (or laterality) can be incorporated into an image classifier. My googling on the topic is raising more questions than answers\n\nIf the answer is to use multi-label classification such as “Y 49” to indicate a cancer diagnosis from machine 49, then how do you get a prediction that only considers the cancer diagnosis? Similarly how do you get a prediction that only considers the cancer diagnosis if a single label classification is used?\n\nHow should you proceed if you take an ensemble approach, create a separate model for each machine_id and store the trained learned objects in a list? (This seems to be the most logical approach to me)\n\nIs it possible to incorporate some form of control structure such as “ if ... then” or “select case” so that a particular trained learner object is used depending on the machine_id?\n\nMy WIP notebook is available at https://www.kaggle.com/code/julianmacnamara/rsna-breast-cancer-detection-pttoa-v1-2\n\nMany thanks in advance\n\nAll the best\n\nJulian\n", "2148645": "Referring to my question on taking an ensemble approach, this seems to work\n\n- For each machine id of interest extract the prediction\n- Create a df containing all the predictions and calculate the mean. (I'll also explore min and max)\n- Proceed as for a single model\n\n", "2148639": "[Here](https://www.kaggle.com/code/hengck23/3hr-tensorrt-nextvit-example?scriptVersionId=116312479&cellId=12) is an inference example considering several models/nets.", "2149731": "Hi Julian, I found this [notebook](https://www.kaggle.com/code/awsaf49/rsna-bcd-efficientnet-tf-tpu-1vm-train) useful. ", "2148671": "One way is described in this notebook:\nhttps://www.kaggle.com/code/andradaolteanu/rsna-breast-cancer-eda-pytorch-baseline\n", "2148638": "Perhaps [this](https://www.kaggle.com/code/vslaykovsky/infer-pytorch-aux-targets-weighted-loss-thres?scriptVersionId=113756289&cellId=15) will be helpful to implement auxiliary outputs?" } }