{ "id": 315305, "title": "How to train the model after EDA ?", "url": "/competitions/happy-whale-and-dolphin/discussion/315305", "author_name": "", "post_date": "2022-03-27T12:34:46.392734400Z", "votes": 2, "comment_count": 3, "views": 0, "content": "

Hi everyone, this is my first competition and i have a very basic question regarding this :-

\n

After the EDA part I don't know how to start the classification of the images as mostly I have used pre-trained models like VGG ,INCEPTION etc. on dataset where the images are classified in particular folder but that's not the case here and the notebooks here where they have trained the models are very complex for me to understand.

\n

So can anyone suggest me how to start the classification part for these types of dataset or can suggest me some resources to kickstart these things as i have general idea on this?

\n

Have a great day !!! 😊

", "messages": [ { "id": "1736540", "postDate": "03/27/2022 12:34:46", "content": "

Hi everyone, this is my first competition and i have a very basic question regarding this :-

\n

After the EDA part I don't know how to start the classification of the images as mostly I have used pre-trained models like VGG ,INCEPTION etc. on dataset where the images are classified in particular folder but that's not the case here and the notebooks here where they have trained the models are very complex for me to understand.

\n

So can anyone suggest me how to start the classification part for these types of dataset or can suggest me some resources to kickstart these things as i have general idea on this?

\n

Have a great day !!! 😊

", "rawMarkdown": "Hi everyone, this is my first competition and i have a very basic question regarding this :-\n\nAfter the EDA part I don't know how to start the classification of the images as mostly I have used pre-trained models like VGG ,INCEPTION etc. on dataset where the images are classified in particular folder but that's not the case here and the notebooks here where they have trained the models are very complex for me to understand.\n\nSo can anyone suggest me how to start the classification part for these types of dataset or can suggest me some resources to kickstart these things as i have general idea on this?\n\nHave a great day !!! 😊", "votes": null }, { "id": "1743657", "postDate": "04/03/2022 07:26:56", "content": "

The first think is to be sure you understand the nature of this competition, this is not a classical classification or regression machine learning problem. You can start reading and understanding some of the good public notebooks.
\nI suggest starting with https://www.kaggle.com/code/andradaolteanu/whales-dolphins-effnet-embedding-cos-distance from @andradaolteanu , a very clean and full of information notebook, and then focus on pre-processing, convolutional architecture and embeddings post processing techniques

", "rawMarkdown": "The first think is to be sure you understand the nature of this competition, this is not a classical classification or regression machine learning problem. You can start reading and understanding some of the good public notebooks.\nI suggest starting with https://www.kaggle.com/code/andradaolteanu/whales-dolphins-effnet-embedding-cos-distance from @andradaolteanu , a very clean and full of information notebook, and then focus on pre-processing, convolutional architecture and embeddings post processing techniques", "votes": null }, { "id": "1758141", "postDate": "04/17/2022 11:10:57", "content": "

Thanks @vladvdv for replying i will definitely check what you have suggested and can you tell me some good resources to learn more about these topics.

", "rawMarkdown": "Thanks @vladvdv for replying i will definitely check what you have suggested and can you tell me some good resources to learn more about these topics.", "votes": null }, { "id": "1758173", "postDate": "04/17/2022 11:54:59", "content": "

@kunalrawat
\nYou can try understanding my public notebook for training
\nhttps://www.kaggle.com/code/vladvdv/simple-vanilla-pytorch-train-notebook-gem-pooling
\nand then, after training, use the inference one
\nhttps://www.kaggle.com/code/vladvdv/simple-vanilla-pytorch-inference-gem-pooling

\n

There are pretty clear. Let me know if you have trouble understanding what I have design

", "rawMarkdown": "kunalrawat \nYou can try understanding my public notebook for training \nhttps://www.kaggle.com/code/vladvdv/simple-vanilla-pytorch-train-notebook-gem-pooling\nand then, after training, use the inference one\nhttps://www.kaggle.com/code/vladvdv/simple-vanilla-pytorch-inference-gem-pooling\n\nThere are pretty clear. Let me know if you have trouble understanding what I have design", "votes": null } ], "comments": [ { "id": 1743657, "author_name": "vladvdv", "author_url": "", "post_date": "04/03/2022 07:26:56", "content": "

The first think is to be sure you understand the nature of this competition, this is not a classical classification or regression machine learning problem. You can start reading and understanding some of the good public notebooks.
\nI suggest starting with https://www.kaggle.com/code/andradaolteanu/whales-dolphins-effnet-embedding-cos-distance from @andradaolteanu , a very clean and full of information notebook, and then focus on pre-processing, convolutional architecture and embeddings post processing techniques

", "votes": null, "replies": [ { "id": 1758141, "author_name": "kunalrawat", "author_url": "", "post_date": "04/17/2022 11:10:57", "content": "

Thanks @vladvdv for replying i will definitely check what you have suggested and can you tell me some good resources to learn more about these topics.

", "votes": null, "replies": [] }, { "id": 1758173, "author_name": "vladvdv", "author_url": "", "post_date": "04/17/2022 11:54:59", "content": "

@kunalrawat
\nYou can try understanding my public notebook for training
\nhttps://www.kaggle.com/code/vladvdv/simple-vanilla-pytorch-train-notebook-gem-pooling
\nand then, after training, use the inference one
\nhttps://www.kaggle.com/code/vladvdv/simple-vanilla-pytorch-inference-gem-pooling

\n

There are pretty clear. Let me know if you have trouble understanding what I have design

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1736540": "Hi everyone, this is my first competition and i have a very basic question regarding this :-\n\nAfter the EDA part I don't know how to start the classification of the images as mostly I have used pre-trained models like VGG ,INCEPTION etc. on dataset where the images are classified in particular folder but that's not the case here and the notebooks here where they have trained the models are very complex for me to understand.\n\nSo can anyone suggest me how to start the classification part for these types of dataset or can suggest me some resources to kickstart these things as i have general idea on this?\n\nHave a great day !!! 😊", "1743657": "The first think is to be sure you understand the nature of this competition, this is not a classical classification or regression machine learning problem. You can start reading and understanding some of the good public notebooks.\nI suggest starting with https://www.kaggle.com/code/andradaolteanu/whales-dolphins-effnet-embedding-cos-distance from @andradaolteanu , a very clean and full of information notebook, and then focus on pre-processing, convolutional architecture and embeddings post processing techniques", "1758141": "Thanks @vladvdv for replying i will definitely check what you have suggested and can you tell me some good resources to learn more about these topics.", "1758173": "kunalrawat \nYou can try understanding my public notebook for training \nhttps://www.kaggle.com/code/vladvdv/simple-vanilla-pytorch-train-notebook-gem-pooling\nand then, after training, use the inference one\nhttps://www.kaggle.com/code/vladvdv/simple-vanilla-pytorch-inference-gem-pooling\n\nThere are pretty clear. Let me know if you have trouble understanding what I have design" }, "source": "meta" }