{
  "id": 350708,
  "title": "[CITEseq CV:0.89,Multiome CV:0.67] Sharing my PyTorch swiss army knife with the community",
  "url": "/competitions/open-problems-multimodal/discussion/350708",
  "author_name": "Vladimir Slaykovskiy",
  "post_date": "2022-09-06T19:37:51.792000",
  "votes": 12,
  "comment_count": 0,
  "views": 0,
  "content": "<p>I want to share my <a href=\"https://www.kaggle.com/code/vslaykovsky/multi-67-cite-89-pytorch-swiss-army-knife\" target=\"_blank\">[Multi CV:.67; CITE: CV:.89] PyTorch Swiss Army Knife</a> solution with you. It includes all the parts you'll ever need for a successfull submission. These include:</p>\n<ul>\n<li>training both Multiome and CITEseq models</li>\n<li>support of both SVD-compressed and raw features</li>\n<li>Kaggle and custom server training</li>\n<li>k-fold training and cross-validation</li>\n<li>correlation score/loss function</li>\n<li>hyperparameter optimization with Optuna</li>\n<li>prediction with k-fold ensemble</li>\n<li>wandb logging to track your metrics</li>\n<li>patching public submissions with your predictions</li>\n<li>preconfigured parameters to train Multiome+SVD, Multiome+Sparse, CITEseq+SVD, CITEseq+Sparse models</li>\n</ul>\n<p>Hope you find this useful!<br>\nAnd don't forget to have fun, competition is not only about medals and prizes!</p>",
  "messages": [
    {
      "id": 1929074,
      "postDate": "2022-09-06T19:37:51.793Z",
      "content": "<p>I want to share my <a href=\"https://www.kaggle.com/code/vslaykovsky/multi-67-cite-89-pytorch-swiss-army-knife\" target=\"_blank\">[Multi CV:.67; CITE: CV:.89] PyTorch Swiss Army Knife</a> solution with you. It includes all the parts you'll ever need for a successfull submission. These include:</p>\n<ul>\n<li>training both Multiome and CITEseq models</li>\n<li>support of both SVD-compressed and raw features</li>\n<li>Kaggle and custom server training</li>\n<li>k-fold training and cross-validation</li>\n<li>correlation score/loss function</li>\n<li>hyperparameter optimization with Optuna</li>\n<li>prediction with k-fold ensemble</li>\n<li>wandb logging to track your metrics</li>\n<li>patching public submissions with your predictions</li>\n<li>preconfigured parameters to train Multiome+SVD, Multiome+Sparse, CITEseq+SVD, CITEseq+Sparse models</li>\n</ul>\n<p>Hope you find this useful!<br>\nAnd don't forget to have fun, competition is not only about medals and prizes!</p>",
      "rawMarkdown": "I want to share my [[Multi CV:.67; CITE: CV:.89] PyTorch Swiss Army Knife](https://www.kaggle.com/code/vslaykovsky/multi-67-cite-89-pytorch-swiss-army-knife) solution with you. It includes all the parts you'll ever need for a successfull submission. These include:\n* training both Multiome and CITEseq models\n* support of both SVD-compressed and raw features\n* Kaggle and custom server training\n* k-fold training and cross-validation\n* correlation score/loss function\n* hyperparameter optimization with Optuna\n* prediction with k-fold ensemble\n* wandb logging to track your metrics\n* patching public submissions with your predictions\n* preconfigured parameters to train Multiome+SVD, Multiome+Sparse, CITEseq+SVD, CITEseq+Sparse models\n\nHope you find this useful!\nAnd don't forget to have fun, competition is not only about medals and prizes!",
      "votes": 12
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1929074": "I want to share my [[Multi CV:.67; CITE: CV:.89] PyTorch Swiss Army Knife](https://www.kaggle.com/code/vslaykovsky/multi-67-cite-89-pytorch-swiss-army-knife) solution with you. It includes all the parts you'll ever need for a successfull submission. These include:\n* training both Multiome and CITEseq models\n* support of both SVD-compressed and raw features\n* Kaggle and custom server training\n* k-fold training and cross-validation\n* correlation score/loss function\n* hyperparameter optimization with Optuna\n* prediction with k-fold ensemble\n* wandb logging to track your metrics\n* patching public submissions with your predictions\n* preconfigured parameters to train Multiome+SVD, Multiome+Sparse, CITEseq+SVD, CITEseq+Sparse models\n\nHope you find this useful!\nAnd don't forget to have fun, competition is not only about medals and prizes!"
  }
}