{
  "id": 107138,
  "title": "Why test data with scc is not yet available?",
  "url": "/competitions/champs-scalar-coupling/discussion/107138",
  "author_name": "m karthikeyan",
  "post_date": "2019-09-02T14:26:00.139000",
  "votes": 1,
  "comment_count": 0,
  "views": 0,
  "content": "<p>71% evaluation means what about outliers? why? I would be happy to get the test set SCC values.. posted accessible to all. Please let me know the link if available.. (privately available to gold medalists and award winners) We put some efforts, time, computational resources (even if it is laptop to compute 25m entries), but did not get any email from organizers.. irresective of winning/losing.. I did only one submission based on calculation done on 25m entries my resutls was 0.8 R2 .. (for entire dataset for buidling models PLS etc) and for the diverse molecular dataset, without any filters.. if DSO, PSO, SD and FC not given which is critical for SCC .. getting SCC from XYZ without calculating energy , forces, distance is quite surprising.. and magnetic properties calculation depends on these factors (from my Pauling's book)</p>\n\n<p>I am happy to see the results declared in hurry.. would be happy to see the research papers coming out of prediction results (71% of total). if the predicted is directly coorelated with test set values originally available to organizers only, then the relationship with the submitted prediction (y'=m'x+c') can be built and that would correct prediction.</p>\n\n<p>top 99% is very interesting.. the SCC does not belong to one group of compounds (it has 6 types of compounds/SCCs) building 6 models and identify the suitable model based on structural evaluation for the accuracy is not correct as it kills the sprit of competition that should not use extensive computational resources or time. Time and computational complexity of final test prediction should be a criteria for true success it is my personal view. karthi</p>",
  "messages": [
    {
      "id": 615943,
      "postDate": "2019-09-02T14:26:00.140Z",
      "content": "<p>71% evaluation means what about outliers? why? I would be happy to get the test set SCC values.. posted accessible to all. Please let me know the link if available.. (privately available to gold medalists and award winners) We put some efforts, time, computational resources (even if it is laptop to compute 25m entries), but did not get any email from organizers.. irresective of winning/losing.. I did only one submission based on calculation done on 25m entries my resutls was 0.8 R2 .. (for entire dataset for buidling models PLS etc) and for the diverse molecular dataset, without any filters.. if DSO, PSO, SD and FC not given which is critical for SCC .. getting SCC from XYZ without calculating energy , forces, distance is quite surprising.. and magnetic properties calculation depends on these factors (from my Pauling's book)</p>\n\n<p>I am happy to see the results declared in hurry.. would be happy to see the research papers coming out of prediction results (71% of total). if the predicted is directly coorelated with test set values originally available to organizers only, then the relationship with the submitted prediction (y'=m'x+c') can be built and that would correct prediction.</p>\n\n<p>top 99% is very interesting.. the SCC does not belong to one group of compounds (it has 6 types of compounds/SCCs) building 6 models and identify the suitable model based on structural evaluation for the accuracy is not correct as it kills the sprit of competition that should not use extensive computational resources or time. Time and computational complexity of final test prediction should be a criteria for true success it is my personal view. karthi</p>",
      "rawMarkdown": "71% evaluation means what about outliers? why? I would be happy to get the test set SCC values.. posted accessible to all. Please let me know the link if available.. (privately available to gold medalists and award winners) We put some efforts, time, computational resources (even if it is laptop to compute 25m entries), but did not get any email from organizers.. irresective of winning/losing.. I did only one submission based on calculation done on 25m entries my resutls was 0.8 R2 .. (for entire dataset for buidling models PLS etc) and for the diverse molecular dataset, without any filters.. if DSO, PSO, SD and FC not given which is critical for SCC .. getting SCC from XYZ without calculating energy , forces, distance is quite surprising.. and magnetic properties calculation depends on these factors (from my Pauling's book)\n\nI am happy to see the results declared in hurry.. would be happy to see the research papers coming out of prediction results (71% of total). if the predicted is directly coorelated with test set values originally available to organizers only, then the relationship with the submitted prediction (y'=m'x+c') can be built and that would correct prediction.\n\ntop 99% is very interesting.. the SCC does not belong to one group of compounds (it has 6 types of compounds/SCCs) building 6 models and identify the suitable model based on structural evaluation for the accuracy is not correct as it kills the sprit of competition that should not use extensive computational resources or time. Time and computational complexity of final test prediction should be a criteria for true success it is my personal view. karthi",
      "votes": 1
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "615943": "71% evaluation means what about outliers? why? I would be happy to get the test set SCC values.. posted accessible to all. Please let me know the link if available.. (privately available to gold medalists and award winners) We put some efforts, time, computational resources (even if it is laptop to compute 25m entries), but did not get any email from organizers.. irresective of winning/losing.. I did only one submission based on calculation done on 25m entries my resutls was 0.8 R2 .. (for entire dataset for buidling models PLS etc) and for the diverse molecular dataset, without any filters.. if DSO, PSO, SD and FC not given which is critical for SCC .. getting SCC from XYZ without calculating energy , forces, distance is quite surprising.. and magnetic properties calculation depends on these factors (from my Pauling's book)\n\nI am happy to see the results declared in hurry.. would be happy to see the research papers coming out of prediction results (71% of total). if the predicted is directly coorelated with test set values originally available to organizers only, then the relationship with the submitted prediction (y'=m'x+c') can be built and that would correct prediction.\n\ntop 99% is very interesting.. the SCC does not belong to one group of compounds (it has 6 types of compounds/SCCs) building 6 models and identify the suitable model based on structural evaluation for the accuracy is not correct as it kills the sprit of competition that should not use extensive computational resources or time. Time and computational complexity of final test prediction should be a criteria for true success it is my personal view. karthi"
  }
}