{
  "id": 213166,
  "title": "Why Weights and Biases(experiment tracking) is useful from Kaggle's perspective?",
  "url": "/competitions/cassava-leaf-disease-classification/discussion/213166",
  "author_name": "Ayush Thakur",
  "post_date": "2021-01-21T18:04:41.124000",
  "votes": 6,
  "comment_count": 0,
  "views": 0,
  "content": "<p>I use <a href=\"https://wandb.ai/site\" target=\"_blank\">Weights and Biases</a> at work and for my personal endeavors. It's an amazing experiment tracking tool. After working my way in this competition, it just struck me that an experiment tracking tool is really useful when the kernel is in commit mode. </p>\n<ul>\n<li>One can monitor the metrics live which is not possible when the kernel is running in commit mode.</li>\n<li>Can get a sense of the time left for the Kernel to finish running. </li>\n<li>Metric comparison helps to decide a better model.</li>\n</ul>\n<p>There are many more advantages of baking an experiment tracking tool with your kernel. Happy coding. :) </p>\n<p>By the way, these are the two notebooks I have so far:</p>\n<ul>\n<li><p><a href=\"https://www.kaggle.com/ayuraj/efficientnet-mixup-k-fold-using-tf-and-wandb#%E2%9A%97%EF%B8%8F-Save-Model-for-Inference\" target=\"_blank\">EfficientNet+Mixup+K-Fold using TF and wandb</a></p></li>\n<li><p><a href=\"https://www.kaggle.com/ayuraj/tensorflow-inference-no-tta\" target=\"_blank\">TensorFlow Inference [No-TTA]</a></p></li>\n</ul>\n<p>They are a work in progress but provides a useful starter kernel to train an EfficientNet model using TensorFlow and TPU. </p>\n<p>It's baked with Weights and Biases. </p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2212893%2Fdfd1d853455eec18bbd8d1ae1d882de6%2Fffasf.png?generation=1611252183063224&amp;alt=media\" alt=\"\"><br>\n^ The Weights and Biases dashboard that I was monitoring and this thought struck.</p>",
  "messages": [
    {
      "id": 1163526,
      "postDate": "2021-01-21T18:04:41.123Z",
      "content": "<p>I use <a href=\"https://wandb.ai/site\" target=\"_blank\">Weights and Biases</a> at work and for my personal endeavors. It's an amazing experiment tracking tool. After working my way in this competition, it just struck me that an experiment tracking tool is really useful when the kernel is in commit mode. </p>\n<ul>\n<li>One can monitor the metrics live which is not possible when the kernel is running in commit mode.</li>\n<li>Can get a sense of the time left for the Kernel to finish running. </li>\n<li>Metric comparison helps to decide a better model.</li>\n</ul>\n<p>There are many more advantages of baking an experiment tracking tool with your kernel. Happy coding. :) </p>\n<p>By the way, these are the two notebooks I have so far:</p>\n<ul>\n<li><p><a href=\"https://www.kaggle.com/ayuraj/efficientnet-mixup-k-fold-using-tf-and-wandb#%E2%9A%97%EF%B8%8F-Save-Model-for-Inference\" target=\"_blank\">EfficientNet+Mixup+K-Fold using TF and wandb</a></p></li>\n<li><p><a href=\"https://www.kaggle.com/ayuraj/tensorflow-inference-no-tta\" target=\"_blank\">TensorFlow Inference [No-TTA]</a></p></li>\n</ul>\n<p>They are a work in progress but provides a useful starter kernel to train an EfficientNet model using TensorFlow and TPU. </p>\n<p>It's baked with Weights and Biases. </p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2212893%2Fdfd1d853455eec18bbd8d1ae1d882de6%2Fffasf.png?generation=1611252183063224&amp;alt=media\" alt=\"\"><br>\n^ The Weights and Biases dashboard that I was monitoring and this thought struck.</p>",
      "rawMarkdown": "I use [Weights and Biases](https://wandb.ai/site) at work and for my personal endeavors. It's an amazing experiment tracking tool. After working my way in this competition, it just struck me that an experiment tracking tool is really useful when the kernel is in commit mode. \n\n* One can monitor the metrics live which is not possible when the kernel is running in commit mode.\n* Can get a sense of the time left for the Kernel to finish running. \n* Metric comparison helps to decide a better model.\n\nThere are many more advantages of baking an experiment tracking tool with your kernel. Happy coding. :) \n\n\nBy the way, these are the two notebooks I have so far:\n\n* [EfficientNet+Mixup+K-Fold using TF and wandb](https://www.kaggle.com/ayuraj/efficientnet-mixup-k-fold-using-tf-and-wandb#%E2%9A%97%EF%B8%8F-Save-Model-for-Inference)\n\n* [TensorFlow Inference [No-TTA]](https://www.kaggle.com/ayuraj/tensorflow-inference-no-tta)\n\nThey are a work in progress but provides a useful starter kernel to train an EfficientNet model using TensorFlow and TPU. \n\nIt's baked with Weights and Biases. \n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2212893%2Fdfd1d853455eec18bbd8d1ae1d882de6%2Fffasf.png?generation=1611252183063224&alt=media)\n^ The Weights and Biases dashboard that I was monitoring and this thought struck.\n",
      "votes": 6
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1163526": "I use [Weights and Biases](https://wandb.ai/site) at work and for my personal endeavors. It's an amazing experiment tracking tool. After working my way in this competition, it just struck me that an experiment tracking tool is really useful when the kernel is in commit mode. \n\n* One can monitor the metrics live which is not possible when the kernel is running in commit mode.\n* Can get a sense of the time left for the Kernel to finish running. \n* Metric comparison helps to decide a better model.\n\nThere are many more advantages of baking an experiment tracking tool with your kernel. Happy coding. :) \n\n\nBy the way, these are the two notebooks I have so far:\n\n* [EfficientNet+Mixup+K-Fold using TF and wandb](https://www.kaggle.com/ayuraj/efficientnet-mixup-k-fold-using-tf-and-wandb#%E2%9A%97%EF%B8%8F-Save-Model-for-Inference)\n\n* [TensorFlow Inference [No-TTA]](https://www.kaggle.com/ayuraj/tensorflow-inference-no-tta)\n\nThey are a work in progress but provides a useful starter kernel to train an EfficientNet model using TensorFlow and TPU. \n\nIt's baked with Weights and Biases. \n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2212893%2Fdfd1d853455eec18bbd8d1ae1d882de6%2Fffasf.png?generation=1611252183063224&alt=media)\n^ The Weights and Biases dashboard that I was monitoring and this thought struck.\n"
  }
}