Based on experiments of @lihaoweicvch discussed here.
\nI've prepared training and inference notebook to continue from there.
Training Notebook: https://www.kaggle.com/code/salmanahmedtamu/training-0-65-0-66/notebook
\nInference Notebook: https://www.kaggle.com/code/salmanahmedtamu/new-torch-jit-submission-nf
Note: Please install libraries if they are missing in training notebook.
", "messages": [ { "id": "2789620", "postDate": "05/02/2024 18:31:57", "content": "Based on experiments of @lihaoweicvch discussed here.
\nI've prepared training and inference notebook to continue from there.
Training Notebook: https://www.kaggle.com/code/salmanahmedtamu/training-0-65-0-66/notebook
\nInference Notebook: https://www.kaggle.com/code/salmanahmedtamu/new-torch-jit-submission-nf
Note: Please install libraries if they are missing in training notebook.
", "rawMarkdown": "Based on experiments of @lihaoweicvch discussed [here](https://www.kaggle.com/competitions/birdclef-2024/discussion/497539). \nI've prepared training and inference notebook to continue from there. \n\nTraining Notebook: https://www.kaggle.com/code/salmanahmedtamu/training-0-65-0-66/notebook\nInference Notebook: https://www.kaggle.com/code/salmanahmedtamu/new-torch-jit-submission-nf\n\nNote: Please install libraries if they are missing in training notebook.", "votes": null }, { "id": "2789699", "postDate": "05/02/2024 19:23:10", "content": "Thank you for sharing! For how many epochs did you train this model, 100?
", "rawMarkdown": "Thank you for sharing! For how many epochs did you train this model, 100?", "votes": null }, { "id": "2789799", "postDate": "05/02/2024 20:24:50", "content": "I used the model at 18th epoch for inference notebook.
", "rawMarkdown": "I used the model at 18th epoch for inference notebook.", "votes": null }, { "id": "2789857", "postDate": "05/02/2024 21:25:33", "content": "Did you have these functions out there somewhere? from metrics import calculate_competition_metrics, metrics_to_string, calculate_competition_metrics_no_map
", "rawMarkdown": "Did you have these functions out there somewhere? from metrics import calculate_competition_metrics, metrics_to_string, calculate_competition_metrics_no_map", "votes": null }, { "id": "2790079", "postDate": "05/03/2024 03:12:40", "content": "Yeah theses functions just calculate ROC, CMAP and put values in a string for the logs.
", "rawMarkdown": "Yeah theses functions just calculate ROC, CMAP and put values in a string for the logs.", "votes": null }, { "id": "2790781", "postDate": "05/03/2024 10:32:43", "content": "I am very curious what local scores you get with this setup? I could not see that in your notebook.
", "rawMarkdown": "I am very curious what local scores you get with this setup? I could not see that in your notebook.", "votes": null }, { "id": "2791251", "postDate": "05/03/2024 15:03:58", "content": "Its in the notebook. Please scroll down and you will see 2nd or 3rd cell with the logs.
", "rawMarkdown": "Its in the notebook. Please scroll down and you will see 2nd or 3rd cell with the logs.", "votes": null }, { "id": "2791957", "postDate": "05/04/2024 00:45:12", "content": "Nice work!
", "rawMarkdown": "Nice work!", "votes": null }, { "id": "2792011", "postDate": "05/04/2024 03:20:30", "content": "Nice work! When I try to submit the results, I get a Submission Scoring Error in submission. Do you know why?
", "rawMarkdown": "Nice work! When I try to submit the results, I get a Submission Scoring Error in submission. Do you know why?", "votes": null }, { "id": "2792017", "postDate": "05/04/2024 03:24:56", "content": "The model itself is not shared. If you want to submit this notebook you would need to alter the training notebook to run and train it first.
", "rawMarkdown": "The model itself is not shared. If you want to submit this notebook you would need to alter the training notebook to run and train it first.", "votes": null }, { "id": "2792027", "postDate": "05/04/2024 03:31:23", "content": "I have modified his notebook and trained my own model. But failed at the last step.
", "rawMarkdown": "I have modified his notebook and trained my own model. But failed at the last step.", "votes": null }, { "id": "2804270", "postDate": "05/09/2024 23:21:32", "content": "Thank you for sharing your notebook! I edited your training and inference code to add some functions and ran it; my score was 0.61. Were there any crucial points that helped you achieve a score between 0.65 and 0.66? It seems that the output from the training notebook, which is in .bin format, is not used directly in the inference code that requires a .pt file. I would appreciate it if you could share any tips!
", "rawMarkdown": "Thank you for sharing your notebook! I edited your training and inference code to add some functions and ran it; my score was 0.61. Were there any crucial points that helped you achieve a score between 0.65 and 0.66? It seems that the output from the training notebook, which is in .bin format, is not used directly in the inference code that requires a .pt file. I would appreciate it if you could share any tips!", "votes": null }, { "id": "2804306", "postDate": "05/10/2024 00:40:11", "content": "Same question here. I can only get LB score 0.62 by using this notebook.
", "rawMarkdown": "Same question here. I can only get LB score 0.62 by using this notebook.", "votes": null }, { "id": "2804324", "postDate": "05/10/2024 01:31:23", "content": "\n\nIt seems that the output from the training notebook, which is in .bin format, is not used directly in the inference code that requires a .pt file.
\n
I think it's converted to torchscript format, like this :
\nmodel_scripted = torch.jit.script(model) # Export to TorchScript
\nmodel_scripted.save('model_scripted_00.pt')
I'm not sure if the author has a different way to do the conversion
", "rawMarkdown": "> It seems that the output from the training notebook, which is in .bin format, is not used directly in the inference code that requires a .pt file.\n\nI think it's converted to torchscript format, like this :\n\nmodel_scripted = torch.jit.script(model) # Export to TorchScript\nmodel_scripted.save('model_scripted_00.pt')\n\nI'm not sure if the author has a different way to do the conversion", "votes": null }, { "id": "2815636", "postDate": "05/16/2024 02:30:12", "content": "Thank you for sharing your notebook! I want to know why you process the output with sigmoid function. Softmax will be all right.
\ne.g.
\nfor row_id_idx, row_id in enumerate(row_ids):\n prediction_dict[str(row_id)].append(output[row_id_idx, :].sigmoid().detach().numpy())\n
This one is a multilabel competition. Each example can have multiple labels. Hence Sigmoid based approaches.
", "rawMarkdown": "This one is a multilabel competition. Each example can have multiple labels. Hence Sigmoid based approaches.", "votes": null } ], "comments": [ { "id": 2789699, "author_name": "snnclsr", "author_url": "", "post_date": "05/02/2024 19:23:10", "content": "Thank you for sharing! For how many epochs did you train this model, 100?
", "votes": null, "replies": [ { "id": 2789799, "author_name": "salmanahmedtamu", "author_url": "", "post_date": "05/02/2024 20:24:50", "content": "I used the model at 18th epoch for inference notebook.
", "votes": null, "replies": [] } ] }, { "id": 2789857, "author_name": "cody11null", "author_url": "", "post_date": "05/02/2024 21:25:33", "content": "Did you have these functions out there somewhere? from metrics import calculate_competition_metrics, metrics_to_string, calculate_competition_metrics_no_map
", "votes": null, "replies": [ { "id": 2790079, "author_name": "salmanahmedtamu", "author_url": "", "post_date": "05/03/2024 03:12:40", "content": "Yeah theses functions just calculate ROC, CMAP and put values in a string for the logs.
", "votes": null, "replies": [] } ] }, { "id": 2790781, "author_name": "tolgakopar", "author_url": "", "post_date": "05/03/2024 10:32:43", "content": "I am very curious what local scores you get with this setup? I could not see that in your notebook.
", "votes": null, "replies": [ { "id": 2791251, "author_name": "salmanahmedtamu", "author_url": "", "post_date": "05/03/2024 15:03:58", "content": "Its in the notebook. Please scroll down and you will see 2nd or 3rd cell with the logs.
", "votes": null, "replies": [] } ] }, { "id": 2791957, "author_name": "lihaoweicvch", "author_url": "", "post_date": "05/04/2024 00:45:12", "content": "Nice work!
", "votes": null, "replies": [] }, { "id": 2792011, "author_name": "shawntung", "author_url": "", "post_date": "05/04/2024 03:20:30", "content": "Nice work! When I try to submit the results, I get a Submission Scoring Error in submission. Do you know why?
", "votes": null, "replies": [ { "id": 2792017, "author_name": "cody11null", "author_url": "", "post_date": "05/04/2024 03:24:56", "content": "The model itself is not shared. If you want to submit this notebook you would need to alter the training notebook to run and train it first.
", "votes": null, "replies": [ { "id": 2792027, "author_name": "shawntung", "author_url": "", "post_date": "05/04/2024 03:31:23", "content": "I have modified his notebook and trained my own model. But failed at the last step.
", "votes": null, "replies": [] } ] } ] }, { "id": 2804270, "author_name": "kmatsu01", "author_url": "", "post_date": "05/09/2024 23:21:32", "content": "Thank you for sharing your notebook! I edited your training and inference code to add some functions and ran it; my score was 0.61. Were there any crucial points that helped you achieve a score between 0.65 and 0.66? It seems that the output from the training notebook, which is in .bin format, is not used directly in the inference code that requires a .pt file. I would appreciate it if you could share any tips!
", "votes": null, "replies": [ { "id": 2804306, "author_name": "sjtuwangshuo", "author_url": "", "post_date": "05/10/2024 00:40:11", "content": "Same question here. I can only get LB score 0.62 by using this notebook.
", "votes": null, "replies": [] }, { "id": 2804324, "author_name": "nyleve", "author_url": "", "post_date": "05/10/2024 01:31:23", "content": "\n\nIt seems that the output from the training notebook, which is in .bin format, is not used directly in the inference code that requires a .pt file.
\n
I think it's converted to torchscript format, like this :
\nmodel_scripted = torch.jit.script(model) # Export to TorchScript
\nmodel_scripted.save('model_scripted_00.pt')
I'm not sure if the author has a different way to do the conversion
", "votes": null, "replies": [] } ] }, { "id": 2815636, "author_name": "jonneryr", "author_url": "", "post_date": "05/16/2024 02:30:12", "content": "Thank you for sharing your notebook! I want to know why you process the output with sigmoid function. Softmax will be all right.
\ne.g.
\nfor row_id_idx, row_id in enumerate(row_ids):\n prediction_dict[str(row_id)].append(output[row_id_idx, :].sigmoid().detach().numpy())\n
This one is a multilabel competition. Each example can have multiple labels. Hence Sigmoid based approaches.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "2789620": "Based on experiments of @lihaoweicvch discussed [here](https://www.kaggle.com/competitions/birdclef-2024/discussion/497539). \nI've prepared training and inference notebook to continue from there. \n\nTraining Notebook: https://www.kaggle.com/code/salmanahmedtamu/training-0-65-0-66/notebook\nInference Notebook: https://www.kaggle.com/code/salmanahmedtamu/new-torch-jit-submission-nf\n\nNote: Please install libraries if they are missing in training notebook.", "2789699": "Thank you for sharing! For how many epochs did you train this model, 100?", "2789799": "I used the model at 18th epoch for inference notebook.", "2789857": "Did you have these functions out there somewhere? from metrics import calculate_competition_metrics, metrics_to_string, calculate_competition_metrics_no_map", "2790079": "Yeah theses functions just calculate ROC, CMAP and put values in a string for the logs.", "2790781": "I am very curious what local scores you get with this setup? I could not see that in your notebook.", "2791251": "Its in the notebook. Please scroll down and you will see 2nd or 3rd cell with the logs.", "2791957": "Nice work!", "2792011": "Nice work! When I try to submit the results, I get a Submission Scoring Error in submission. Do you know why?", "2792017": "The model itself is not shared. If you want to submit this notebook you would need to alter the training notebook to run and train it first.", "2792027": "I have modified his notebook and trained my own model. But failed at the last step.", "2804270": "Thank you for sharing your notebook! I edited your training and inference code to add some functions and ran it; my score was 0.61. Were there any crucial points that helped you achieve a score between 0.65 and 0.66? It seems that the output from the training notebook, which is in .bin format, is not used directly in the inference code that requires a .pt file. I would appreciate it if you could share any tips!", "2804306": "Same question here. I can only get LB score 0.62 by using this notebook.", "2804324": "> It seems that the output from the training notebook, which is in .bin format, is not used directly in the inference code that requires a .pt file.\n\nI think it's converted to torchscript format, like this :\n\nmodel_scripted = torch.jit.script(model) # Export to TorchScript\nmodel_scripted.save('model_scripted_00.pt')\n\nI'm not sure if the author has a different way to do the conversion", "2815636": "Thank you for sharing your notebook! I want to know why you process the output with sigmoid function. Softmax will be all right.\ne.g. \n`for row_id_idx, row_id in enumerate(row_ids):\n prediction_dict[str(row_id)].append(output[row_id_idx, :].sigmoid().detach().numpy())\n`", "2815672": "This one is a multilabel competition. Each example can have multiple labels. Hence Sigmoid based approaches." }, "source": "meta" }