Hello to all kagglers! If you've made enough progress in this competition, you've probably realized that the use of ProtBERT's embedding vectors is quite adequate for this competition. Therefore, I have published and written a detailed dataset data card that contains the information needed to use these embeddings:
\nhttps://www.kaggle.com/datasets/henriupton/protbert-embeddings-for-cafa5
\nYou can also find my script that allowed me to collect these data. This will allow you to take it back and customize the settings if you need. To do this I used the ProtBERT data card on HuggingFace : https://huggingface.co/Rostlab/prot_bert .
\nFeel free to give me feedback, share your ideas and put a little upvote to this post and dataset to help me.
\nGood luck to all
\nšš„š„
Hello to all kagglers! If you've made enough progress in this competition, you've probably realized that the use of ProtBERT's embedding vectors is quite adequate for this competition. Therefore, I have published and written a detailed dataset data card that contains the information needed to use these embeddings:
\nhttps://www.kaggle.com/datasets/henriupton/protbert-embeddings-for-cafa5
\nYou can also find my script that allowed me to collect these data. This will allow you to take it back and customize the settings if you need. To do this I used the ProtBERT data card on HuggingFace : https://huggingface.co/Rostlab/prot_bert .
\nFeel free to give me feedback, share your ideas and put a little upvote to this post and dataset to help me.
\nGood luck to all
\nšš„š„
Thanks for providing the data.
\nIn case you are not aware, ProtBERT is one of the older pLMs in the field. There are more recent models that perform better. At the same group's page ( https://huggingface.co/Rostlab ) there are more recent ProtT5 models, and also ESM2 by Facebook. In fact, both ProtT5 and ESM2 embeddings applied to this dataset are already available if you search through the posts.
", "rawMarkdown": "Thanks for providing the data.\n\nIn case you are not aware, ProtBERT is one of the older pLMs in the field. There are more recent models that perform better. At the same group's page ( https://huggingface.co/Rostlab ) there are more recent ProtT5 models, and also ESM2 by Facebook. In fact, both ProtT5 and ESM2 embeddings applied to this dataset are already available if you search through the posts.", "votes": null }, { "id": "2247025", "postDate": "05/05/2023 16:17:17", "content": "Thank you very much for these informations! Will definitely check for ProtT5 and ESM2 models. Would be interesting to provide a kind of benchmark of these three models šš¼
", "rawMarkdown": "Thank you very much for these informations! Will definitely check for ProtT5 and ESM2 models. Would be interesting to provide a kind of benchmark of these three models šš¼", "votes": null } ], "comments": [ { "id": 2246845, "author_name": "tilii7", "author_url": "", "post_date": "05/05/2023 14:01:27", "content": "Thanks for providing the data.
\nIn case you are not aware, ProtBERT is one of the older pLMs in the field. There are more recent models that perform better. At the same group's page ( https://huggingface.co/Rostlab ) there are more recent ProtT5 models, and also ESM2 by Facebook. In fact, both ProtT5 and ESM2 embeddings applied to this dataset are already available if you search through the posts.
", "votes": null, "replies": [ { "id": 2247025, "author_name": "henriupton", "author_url": "", "post_date": "05/05/2023 16:17:17", "content": "Thank you very much for these informations! Will definitely check for ProtT5 and ESM2 models. Would be interesting to provide a kind of benchmark of these three models šš¼
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "2246629": "Hello to all kagglers! If you've made enough progress in this competition, you've probably realized that the use of **ProtBERT's embedding vectors** is quite adequate for this competition. Therefore, I have published and written a detailed dataset data card that contains the information needed to use these embeddings: \n\nhttps://www.kaggle.com/datasets/henriupton/protbert-embeddings-for-cafa5\n\nYou can also find my script that allowed me to collect these data. This will allow you to take it back and customize the settings if you need. To do this I used the ProtBERT data card on HuggingFace : https://huggingface.co/Rostlab/prot_bert .\n\nFeel free to give me feedback, share your ideas and put a little upvote to this post and dataset to help me. \n\nGood luck to all \nšš„š„", "2246845": "Thanks for providing the data.\n\nIn case you are not aware, ProtBERT is one of the older pLMs in the field. There are more recent models that perform better. At the same group's page ( https://huggingface.co/Rostlab ) there are more recent ProtT5 models, and also ESM2 by Facebook. In fact, both ProtT5 and ESM2 embeddings applied to this dataset are already available if you search through the posts.", "2247025": "Thank you very much for these informations! Will definitely check for ProtT5 and ESM2 models. Would be interesting to provide a kind of benchmark of these three models šš¼" }, "source": "meta" }