{
  "id": 71224,
  "title": "Sampling the Data for higher F1 score",
  "url": "/competitions/quora-insincere-questions-classification/discussion/71224",
  "author_name": "Ashutosh Mishra",
  "post_date": "2018-11-11T15:57:16.758000",
  "votes": 0,
  "comment_count": 0,
  "views": 0,
  "content": "<p>After few experimentation :\n- Oversampling makes the F1 score around 0.8 but the cost is time. The kernel took over 9000 secs for just two embeddings with ensemble of LSTM and 1D-ConvNet layers which is way higher than provided 7200 sec range.\n- I have yet to try <a href=\"https://www.kaggle.com/c/quora-insincere-questions-classification/discussion/71184\">F1 score optimisation mentioned here</a>.\n- If anyone has any idea it would be highly helpful.</p>\n\n<p>Thanks!</p>",
  "messages": [
    {
      "id": 419264,
      "postDate": "2018-11-11T15:57:16.757Z",
      "content": "<p>After few experimentation :\n- Oversampling makes the F1 score around 0.8 but the cost is time. The kernel took over 9000 secs for just two embeddings with ensemble of LSTM and 1D-ConvNet layers which is way higher than provided 7200 sec range.\n- I have yet to try <a href=\"https://www.kaggle.com/c/quora-insincere-questions-classification/discussion/71184\">F1 score optimisation mentioned here</a>.\n- If anyone has any idea it would be highly helpful.</p>\n\n<p>Thanks!</p>",
      "rawMarkdown": "After few experimentation :\n- Oversampling makes the F1 score around 0.8 but the cost is time. The kernel took over 9000 secs for just two embeddings with ensemble of LSTM and 1D-ConvNet layers which is way higher than provided 7200 sec range.\n- I have yet to try [F1 score optimisation mentioned here](https://www.kaggle.com/c/quora-insincere-questions-classification/discussion/71184).\n- If anyone has any idea it would be highly helpful.\n\nThanks!"
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "419264": "After few experimentation :\n- Oversampling makes the F1 score around 0.8 but the cost is time. The kernel took over 9000 secs for just two embeddings with ensemble of LSTM and 1D-ConvNet layers which is way higher than provided 7200 sec range.\n- I have yet to try [F1 score optimisation mentioned here](https://www.kaggle.com/c/quora-insincere-questions-classification/discussion/71184).\n- If anyone has any idea it would be highly helpful.\n\nThanks!"
  }
}