{
  "id": 305008,
  "title": "Some interesting validations in different conditions",
  "url": "/competitions/g-research-crypto-forecasting/discussion/305008",
  "author_name": "Kirderf",
  "post_date": "2022-02-03T11:27:11.520000",
  "votes": 5,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Below is a 8-month step training/validation with my model, GBM + TA solution as many other I guess.</p>\n<p>One can see that new market conditions are not doing well as we all know, even trained with full data one can see that with lagged features even with slightly bear market in training but with a majority in other market conditions it didn’t do well in bear market. Nothing new of course, just underscores the important when creating and dealing with this data and prediction. The top and winning solutions might handle this well. As we can submit two solutions, I submitted two trained with different market conditions, and just hope for the best and lots of luck 😊</p>\n<p>The columns for past months in training was just for information/find a pattern, the lagged features are not in that complete period for me, one wish. The training and main competition data is of course majority in Bull market condition.<br>\nAnd Good/Extreme bad score is relative to the model test period, not in the competition, there it’s in comparison overall bad I guess :)</p>\n<p><img src=\"https://i.postimg.cc/Dznb7wLs/Sk-rmklipp.jpg\" alt=\"\"></p>",
  "messages": [
    {
      "id": 1674286,
      "postDate": "2022-02-03T11:27:11.520Z",
      "content": "<p>Below is a 8-month step training/validation with my model, GBM + TA solution as many other I guess.</p>\n<p>One can see that new market conditions are not doing well as we all know, even trained with full data one can see that with lagged features even with slightly bear market in training but with a majority in other market conditions it didn’t do well in bear market. Nothing new of course, just underscores the important when creating and dealing with this data and prediction. The top and winning solutions might handle this well. As we can submit two solutions, I submitted two trained with different market conditions, and just hope for the best and lots of luck 😊</p>\n<p>The columns for past months in training was just for information/find a pattern, the lagged features are not in that complete period for me, one wish. The training and main competition data is of course majority in Bull market condition.<br>\nAnd Good/Extreme bad score is relative to the model test period, not in the competition, there it’s in comparison overall bad I guess :)</p>\n<p><img src=\"https://i.postimg.cc/Dznb7wLs/Sk-rmklipp.jpg\" alt=\"\"></p>",
      "rawMarkdown": "Below is a 8-month step training/validation with my model, GBM + TA solution as many other I guess.\n\nOne can see that new market conditions are not doing well as we all know, even trained with full data one can see that with lagged features even with slightly bear market in training but with a majority in other market conditions it didn’t do well in bear market. Nothing new of course, just underscores the important when creating and dealing with this data and prediction. The top and winning solutions might handle this well. As we can submit two solutions, I submitted two trained with different market conditions, and just hope for the best and lots of luck 😊\n\nThe columns for past months in training was just for information/find a pattern, the lagged features are not in that complete period for me, one wish. The training and main competition data is of course majority in Bull market condition.\nAnd Good/Extreme bad score is relative to the model test period, not in the competition, there it’s in comparison overall bad I guess :)\n\n![](https://i.postimg.cc/Dznb7wLs/Sk-rmklipp.jpg)\n\t\t\t\t\t\n",
      "votes": 5
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1674286": "Below is a 8-month step training/validation with my model, GBM + TA solution as many other I guess.\n\nOne can see that new market conditions are not doing well as we all know, even trained with full data one can see that with lagged features even with slightly bear market in training but with a majority in other market conditions it didn’t do well in bear market. Nothing new of course, just underscores the important when creating and dealing with this data and prediction. The top and winning solutions might handle this well. As we can submit two solutions, I submitted two trained with different market conditions, and just hope for the best and lots of luck 😊\n\nThe columns for past months in training was just for information/find a pattern, the lagged features are not in that complete period for me, one wish. The training and main competition data is of course majority in Bull market condition.\nAnd Good/Extreme bad score is relative to the model test period, not in the competition, there it’s in comparison overall bad I guess :)\n\n![](https://i.postimg.cc/Dznb7wLs/Sk-rmklipp.jpg)\n\t\t\t\t\t\n"
  }
}