{ "id": 521853, "title": "What is the sample weight?", "url": "/competitions/rsna-2024-lumbar-spine-degenerative-classification/discussion/521853", "author_name": "", "post_date": "2024-07-23T09:32:23.224216900Z", "votes": null, "comment_count": 2, "views": 0, "content": "

I might miss something, and come here for clarification.

\n

In the official metric, the condition loss is calculated as

\n
condition_loss = sklearn(\n        y_true=solution,\n        y_pred=submission,\n        sample_weight=solution\n)\n
\n

It means that there is a sample_weight column in the solution, which might take the severity into account. I see the famous notebook from @itsuki9180 use the following sample weight.

\n
 l  labels:\n     ==0: weights.append(1)\n    elif ==1: weights.append(2)\n    elif ==2: weights.append(4)\n    : weights.append(0)\ncv2 = log_loss(labels, y_pred2, =, =weights)\n
\n

I.e. 1 for Normal/Mild, 2 for Moderate and 4 for Severe. Is this sample weight announced somewhere? Or it is set by ourselves to minic the official metric? If it is set by ourselves, do you have a better sample weight?

", "messages": [ { "id": "2932811", "postDate": "07/23/2024 09:32:23", "content": "

I might miss something, and come here for clarification.

\n

In the official metric, the condition loss is calculated as

\n
condition_loss = sklearn(\n        y_true=solution,\n        y_pred=submission,\n        sample_weight=solution\n)\n
\n

It means that there is a sample_weight column in the solution, which might take the severity into account. I see the famous notebook from @itsuki9180 use the following sample weight.

\n
 l  labels:\n     ==0: weights.append(1)\n    elif ==1: weights.append(2)\n    elif ==2: weights.append(4)\n    : weights.append(0)\ncv2 = log_loss(labels, y_pred2, =, =weights)\n
\n

I.e. 1 for Normal/Mild, 2 for Moderate and 4 for Severe. Is this sample weight announced somewhere? Or it is set by ourselves to minic the official metric? If it is set by ourselves, do you have a better sample weight?

", "rawMarkdown": "I might miss something, and come here for clarification.\n\nIn the [official metric](https://www.kaggle.com/code/metric/rsna-lumbar-metric-71549), the condition loss is calculated as \n\n```\ncondition_loss = sklearn.metrics.log_loss(\n y_true=solution.loc[condition_indices, target_levels].values,\n y_pred=submission.loc[condition_indices, target_levels].values,\n sample_weight=solution.loc[condition_indices, 'sample_weight'].values\n)\n```\nIt means that there is a **sample_weight** column in the solution, which might take the severity into account. I see the famous [notebook](https://www.kaggle.com/code/itsuki9180/rsna2024-lsdc-training-baseline) from @itsuki9180 use the following sample weight.\n```\nfor l in labels:\n if l==0: weights.append(1)\n elif l==1: weights.append(2)\n elif l==2: weights.append(4)\n else: weights.append(0)\ncv2 = log_loss(labels, y_pred2, normalize=True, sample_weight=weights)\n```\nI.e. 1 for Normal/Mild, 2 for Moderate and 4 for Severe. Is this sample weight announced somewhere? Or it is set by ourselves to minic the official metric? If it is set by ourselves, do you have a better sample weight?", "votes": null }, { "id": "2932854", "postDate": "07/23/2024 09:51:33", "content": "

Sample weight is given here: Evaluation. For completeness, here is what is given:

\n
\n

The sample weights are as follows:
\n 1 for normal/mild.
\n 2 for moderate.
\n 4 for severe.

\n
", "rawMarkdown": "Sample weight is given here: [Evaluation](https://www.kaggle.com/competitions/rsna-2024-lumbar-spine-degenerative-classification/overview/evaluation). For completeness, here is what is given:\n>The sample weights are as follows:\n1 for normal/mild.\n2 for moderate.\n4 for severe.", "votes": null }, { "id": "2932935", "postDate": "07/23/2024 10:51:49", "content": "

I ignore that information. Really thank you

", "rawMarkdown": "I ignore that information. Really thank you", "votes": null } ], "comments": [ { "id": 2932854, "author_name": "coderrkj", "author_url": "", "post_date": "07/23/2024 09:51:33", "content": "

Sample weight is given here: Evaluation. For completeness, here is what is given:

\n
\n

The sample weights are as follows:
\n 1 for normal/mild.
\n 2 for moderate.
\n 4 for severe.

\n
", "votes": null, "replies": [ { "id": 2932935, "author_name": "baohaoliao", "author_url": "", "post_date": "07/23/2024 10:51:49", "content": "

I ignore that information. Really thank you

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "2932811": "I might miss something, and come here for clarification.\n\nIn the [official metric](https://www.kaggle.com/code/metric/rsna-lumbar-metric-71549), the condition loss is calculated as \n\n```\ncondition_loss = sklearn.metrics.log_loss(\n y_true=solution.loc[condition_indices, target_levels].values,\n y_pred=submission.loc[condition_indices, target_levels].values,\n sample_weight=solution.loc[condition_indices, 'sample_weight'].values\n)\n```\nIt means that there is a **sample_weight** column in the solution, which might take the severity into account. I see the famous [notebook](https://www.kaggle.com/code/itsuki9180/rsna2024-lsdc-training-baseline) from @itsuki9180 use the following sample weight.\n```\nfor l in labels:\n if l==0: weights.append(1)\n elif l==1: weights.append(2)\n elif l==2: weights.append(4)\n else: weights.append(0)\ncv2 = log_loss(labels, y_pred2, normalize=True, sample_weight=weights)\n```\nI.e. 1 for Normal/Mild, 2 for Moderate and 4 for Severe. Is this sample weight announced somewhere? Or it is set by ourselves to minic the official metric? If it is set by ourselves, do you have a better sample weight?", "2932854": "Sample weight is given here: [Evaluation](https://www.kaggle.com/competitions/rsna-2024-lumbar-spine-degenerative-classification/overview/evaluation). For completeness, here is what is given:\n>The sample weights are as follows:\n1 for normal/mild.\n2 for moderate.\n4 for severe.", "2932935": "I ignore that information. Really thank you" }, "source": "meta" }