The theoretical best score is around -20.7232 as mentionned in the evaluation section. \nHow low will the best score be at the end of the competition? I guess -3.0 will be crossed soon. \nWhat are your thoughts and if so, any rationale behind it?
", "messages": [ { "id": "585399", "postDate": "07/27/2019 11:29:32", "content": "The theoretical best score is around -20.7232 as mentionned in the evaluation section. \nHow low will the best score be at the end of the competition? I guess -3.0 will be crossed soon. \nWhat are your thoughts and if so, any rationale behind it?
", "rawMarkdown": "The theoretical best score is around -20.7232 as mentionned in the evaluation section. \nHow low will the best score be at the end of the competition? I guess -3.0 will be crossed soon. \nWhat are your thoughts and if so, any rationale behind it?", "votes": null }, { "id": "585430", "postDate": "07/27/2019 12:47:11", "content": "If you look at methane dsgdb9nsd_000001 (which is a tetrahedral molecule with four equivalent C–H bonds) you will see that given scalar coupling constants are not equal (e.g. -11.257, -11.2548, -11.2543, -11.2541, -11.2548)
\n\nid,molecule_name,atom_index_0,atom_index_1,type,scalar_coupling_constant\n0,dsgdb9nsd_000001,1,0,1JHC,84.8076\n1,dsgdb9nsd_000001,1,2,2JHH,-11.257\n2,dsgdb9nsd_000001,1,3,2JHH,-11.2548\n3,dsgdb9nsd_000001,1,4,2JHH,-11.2543\n4,dsgdb9nsd_000001,2,0,1JHC,84.8074\n5,dsgdb9nsd_000001,2,3,2JHH,-11.2541\n6,dsgdb9nsd_000001,2,4,2JHH,-11.2548\n7,dsgdb9nsd_000001,3,0,1JHC,84.8093\n8,dsgdb9nsd_000001,3,4,2JHH,-11.2543\n\nThe training data is noisy. |-11.257 - -11.2541| = 0.003 and log(0.003) = -5.8\nSo it's barely possible to get -5.8
Brilliant observation. Thanks for sharing. 👍
", "rawMarkdown": "Brilliant observation. Thanks for sharing. 👍", "votes": null }, { "id": "585588", "postDate": "07/27/2019 17:26:08", "content": "Could it be that \"our\" methane is not ideal tetrahedral? CH distances are not exactly equal...
", "rawMarkdown": "Could it be that \"our\" methane is not ideal tetrahedral? CH distances are not exactly equal...", "votes": null }, { "id": "589500", "postDate": "08/01/2019 01:58:31", "content": ".
", "rawMarkdown": ".", "votes": null }, { "id": "589501", "postDate": "08/01/2019 02:00:28", "content": "@sergeyzlobin for sure\nand such precise info of the difference might boost your score. I don't know how properly to perform that for now. @yassinealouini
", "rawMarkdown": "sergeyzlobin for sure\nand such precise info of the difference might boost your score. I don't know how properly to perform that for now. @yassinealouini", "votes": null }, { "id": "589560", "postDate": "08/01/2019 05:15:19", "content": "\n\n\nCould it be that \"our\" methane is not ideal tetrahedral? CH distances are not exactly equal…
\n
It is the result of a numerical simulation, and the fact that it is not a perfect tetrahedron illustrates the point @pavelgonchar makes. There is a little noise in the data which puts a lower limit on the accuracy of any predictive model that can be created.
", "rawMarkdown": "> Could it be that \"our\" methane is not ideal tetrahedral? CH distances are not exactly equal…\n\nIt is the result of a numerical simulation, and the fact that it is not a perfect tetrahedron illustrates the point @pavelgonchar makes. There is a little noise in the data which puts a lower limit on the accuracy of any predictive model that can be created.", "votes": null }, { "id": "590440", "postDate": "08/02/2019 07:43:26", "content": "Aaand, finally, the -3 threshold has been broken!
", "rawMarkdown": "Aaand, finally, the -3 threshold has been broken!", "votes": null }, { "id": "590486", "postDate": "08/02/2019 08:31:08", "content": "Congratulations your team broken -3.0 ,can I ask if your team get benifits from processing noise target?
", "rawMarkdown": "Congratulations your team broken -3.0 ,can I ask if your team get benifits from processing noise target?", "votes": null }, { "id": "590546", "postDate": "08/02/2019 10:15:13", "content": "Yes that's a great achievement and a motivation for all of us to strive for better results.
", "rawMarkdown": "Yes that's a great achievement and a motivation for all of us to strive for better results.", "votes": null }, { "id": "590626", "postDate": "08/02/2019 12:13:27", "content": "I think others also have passed this but they don't submit...
", "rawMarkdown": "I think others also have passed this but they don't submit...", "votes": null }, { "id": "590635", "postDate": "08/02/2019 12:23:10", "content": "Have you seen this behavior (in ebay it was called auction sniping, last-second subs) in other competitions?
", "rawMarkdown": "Have you seen this behavior (in ebay it was called auction sniping, last-second subs) in other competitions?", "votes": null }, { "id": "590649", "postDate": "08/02/2019 12:41:50", "content": "@antorsae would you work harder if you knew that others have passed you?
\n\nNo submission is my favorite technique to prevent LB overfitting.
", "rawMarkdown": "antorsae would you work harder if you knew that others have passed you?\n\nNo submission is my favorite technique to prevent LB overfitting.", "votes": null }, { "id": "590763", "postDate": "08/02/2019 15:02:43", "content": "@antorsae Yes I saw it, and I saw it from some of the people in top 10 here ;)
", "rawMarkdown": "antorsae Yes I saw it, and I saw it from some of the people in top 10 here ;)", "votes": null }, { "id": "590770", "postDate": "08/02/2019 15:09:38", "content": ">There is a little noise in the data which puts a lower limit on the accuracy of any predictive model that can be created.
\n\nUnless the predictive model can also account for this breaking of symmetries
", "rawMarkdown": ">There is a little noise in the data which puts a lower limit on the accuracy of any predictive model that can be created.\n\nUnless the predictive model can also account for this breaking of symmetries", "votes": null }, { "id": "590771", "postDate": "08/02/2019 15:12:01", "content": "@antorsae @pavelgonchar I should have started with this: congrats for passing the -3 score! This is motivating ;)
", "rawMarkdown": "antorsae @pavelgonchar I should have started with this: congrats for passing the -3 score! This is motivating ;)", "votes": null }, { "id": "590772", "postDate": "08/02/2019 15:13:21", "content": "@antorsae congrats with your new score. now you need to kidnap @outrunner before he submits. his 23 days break looks dangerous unless he is already kidnapped:)
", "rawMarkdown": "antorsae congrats with your new score. now you need to kidnap @outrunner before he submits. his 23 days break looks dangerous unless he is already kidnapped:)", "votes": null }, { "id": "590863", "postDate": "08/02/2019 18:11:41", "content": "I'd kidnap Jaechang Lim too ... And anyone not having submitted in 5 days among to 10...
", "rawMarkdown": "I'd kidnap Jaechang Lim too ... And anyone not having submitted in 5 days among to 10...", "votes": null }, { "id": "591333", "postDate": "08/03/2019 14:09:39", "content": "> Unless the predictive model can also account for this breaking of symmetries
\n\nHow would this be possible? For instance, how can you find how symmetry is broken int he first molecule (methane)?
", "rawMarkdown": "> Unless the predictive model can also account for this breaking of symmetries\n\nHow would this be possible? For instance, how can you find how symmetry is broken int he first molecule (methane)?", "votes": null }, { "id": "591401", "postDate": "08/03/2019 16:27:01", "content": "@cpmpml , I guess one could argue that if the breaking of symmetry in the input variables is correlated to the breaking of the symmetry in the coupling a model might be able to exploit this correlation.
\n\n\n\n\nFor instance, how can you find how symmetry is broken int he first molecule
\n
e.g. if the distances in methane vary in the same way the couplings do.
\n\nIs that what you meant @borisdee ?
", "rawMarkdown": "cpmpml , I guess one could argue that if the breaking of symmetry in the input variables is correlated to the breaking of the symmetry in the coupling a model might be able to exploit this correlation. \n\n> For instance, how can you find how symmetry is broken int he first molecule\n\ne.g. if the distances in methane vary in the same way the couplings do. \n\nIs that what you meant @borisdee ?", "votes": null }, { "id": "591788", "postDate": "08/04/2019 08:55:32", "content": "There is another way to roughly estimate the upper bound using significant digits. Target's precision is less than 0.00005. log(0.00005)=-9.9
", "rawMarkdown": "There is another way to roughly estimate the upper bound using significant digits. Target's precision is less than 0.00005. log(0.00005)=-9.9", "votes": null }, { "id": "592374", "postDate": "08/05/2019 07:59:55", "content": "Looks @outrunner has broken his chains
", "rawMarkdown": "Looks @outrunner has broken his chains", "votes": null } ], "comments": [ { "id": 585430, "author_name": "pavelgonchar", "author_url": "", "post_date": "07/27/2019 12:47:11", "content": "If you look at methane dsgdb9nsd_000001 (which is a tetrahedral molecule with four equivalent C–H bonds) you will see that given scalar coupling constants are not equal (e.g. -11.257, -11.2548, -11.2543, -11.2541, -11.2548)
\n\nid,molecule_name,atom_index_0,atom_index_1,type,scalar_coupling_constant\n0,dsgdb9nsd_000001,1,0,1JHC,84.8076\n1,dsgdb9nsd_000001,1,2,2JHH,-11.257\n2,dsgdb9nsd_000001,1,3,2JHH,-11.2548\n3,dsgdb9nsd_000001,1,4,2JHH,-11.2543\n4,dsgdb9nsd_000001,2,0,1JHC,84.8074\n5,dsgdb9nsd_000001,2,3,2JHH,-11.2541\n6,dsgdb9nsd_000001,2,4,2JHH,-11.2548\n7,dsgdb9nsd_000001,3,0,1JHC,84.8093\n8,dsgdb9nsd_000001,3,4,2JHH,-11.2543\n\nThe training data is noisy. |-11.257 - -11.2541| = 0.003 and log(0.003) = -5.8\nSo it's barely possible to get -5.8
Brilliant observation. Thanks for sharing. 👍
", "votes": null, "replies": [] }, { "id": 585588, "author_name": "sergeyzlobin", "author_url": "", "post_date": "07/27/2019 17:26:08", "content": "Could it be that \"our\" methane is not ideal tetrahedral? CH distances are not exactly equal...
", "votes": null, "replies": [] }, { "id": 589500, "author_name": "muhakabartay", "author_url": "", "post_date": "08/01/2019 01:58:31", "content": ".
", "votes": null, "replies": [] }, { "id": 589501, "author_name": "muhakabartay", "author_url": "", "post_date": "08/01/2019 02:00:28", "content": "@sergeyzlobin for sure\nand such precise info of the difference might boost your score. I don't know how properly to perform that for now. @yassinealouini
", "votes": null, "replies": [] }, { "id": 589560, "author_name": "cpmpml", "author_url": "", "post_date": "08/01/2019 05:15:19", "content": "\n\n\nCould it be that \"our\" methane is not ideal tetrahedral? CH distances are not exactly equal…
\n
It is the result of a numerical simulation, and the fact that it is not a perfect tetrahedron illustrates the point @pavelgonchar makes. There is a little noise in the data which puts a lower limit on the accuracy of any predictive model that can be created.
", "votes": null, "replies": [] }, { "id": 590486, "author_name": "senkin13", "author_url": "", "post_date": "08/02/2019 08:31:08", "content": "Congratulations your team broken -3.0 ,can I ask if your team get benifits from processing noise target?
", "votes": null, "replies": [] }, { "id": 590770, "author_name": "borisdee", "author_url": "", "post_date": "08/02/2019 15:09:38", "content": ">There is a little noise in the data which puts a lower limit on the accuracy of any predictive model that can be created.
\n\nUnless the predictive model can also account for this breaking of symmetries
", "votes": null, "replies": [] }, { "id": 591333, "author_name": "cpmpml", "author_url": "", "post_date": "08/03/2019 14:09:39", "content": "> Unless the predictive model can also account for this breaking of symmetries
\n\nHow would this be possible? For instance, how can you find how symmetry is broken int he first molecule (methane)?
", "votes": null, "replies": [] }, { "id": 591401, "author_name": "fnands", "author_url": "", "post_date": "08/03/2019 16:27:01", "content": "@cpmpml , I guess one could argue that if the breaking of symmetry in the input variables is correlated to the breaking of the symmetry in the coupling a model might be able to exploit this correlation.
\n\n\n\n\nFor instance, how can you find how symmetry is broken int he first molecule
\n
e.g. if the distances in methane vary in the same way the couplings do.
\n\nIs that what you meant @borisdee ?
", "votes": null, "replies": [] }, { "id": 591788, "author_name": "pavelgonchar", "author_url": "", "post_date": "08/04/2019 08:55:32", "content": "There is another way to roughly estimate the upper bound using significant digits. Target's precision is less than 0.00005. log(0.00005)=-9.9
", "votes": null, "replies": [] } ] }, { "id": 590440, "author_name": "fnands", "author_url": "", "post_date": "08/02/2019 07:43:26", "content": "Aaand, finally, the -3 threshold has been broken!
", "votes": null, "replies": [ { "id": 590546, "author_name": "yassinealouini", "author_url": "", "post_date": "08/02/2019 10:15:13", "content": "Yes that's a great achievement and a motivation for all of us to strive for better results.
", "votes": null, "replies": [] }, { "id": 590626, "author_name": "cpmpml", "author_url": "", "post_date": "08/02/2019 12:13:27", "content": "I think others also have passed this but they don't submit...
", "votes": null, "replies": [] }, { "id": 590635, "author_name": "antorsae", "author_url": "", "post_date": "08/02/2019 12:23:10", "content": "Have you seen this behavior (in ebay it was called auction sniping, last-second subs) in other competitions?
", "votes": null, "replies": [] }, { "id": 590649, "author_name": "christofhenkel", "author_url": "", "post_date": "08/02/2019 12:41:50", "content": "@antorsae would you work harder if you knew that others have passed you?
\n\nNo submission is my favorite technique to prevent LB overfitting.
", "votes": null, "replies": [] }, { "id": 590763, "author_name": "cpmpml", "author_url": "", "post_date": "08/02/2019 15:02:43", "content": "@antorsae Yes I saw it, and I saw it from some of the people in top 10 here ;)
", "votes": null, "replies": [] }, { "id": 590771, "author_name": "cpmpml", "author_url": "", "post_date": "08/02/2019 15:12:01", "content": "@antorsae @pavelgonchar I should have started with this: congrats for passing the -3 score! This is motivating ;)
", "votes": null, "replies": [] }, { "id": 590772, "author_name": "aerdem4", "author_url": "", "post_date": "08/02/2019 15:13:21", "content": "@antorsae congrats with your new score. now you need to kidnap @outrunner before he submits. his 23 days break looks dangerous unless he is already kidnapped:)
", "votes": null, "replies": [] }, { "id": 590863, "author_name": "cpmpml", "author_url": "", "post_date": "08/02/2019 18:11:41", "content": "I'd kidnap Jaechang Lim too ... And anyone not having submitted in 5 days among to 10...
", "votes": null, "replies": [] }, { "id": 592374, "author_name": "antorsae", "author_url": "", "post_date": "08/05/2019 07:59:55", "content": "Looks @outrunner has broken his chains
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "585399": "The theoretical best score is around -20.7232 as mentionned in the evaluation section. \nHow low will the best score be at the end of the competition? I guess -3.0 will be crossed soon. \nWhat are your thoughts and if so, any rationale behind it?", "585430": "If you look at methane dsgdb9nsd_000001 (which is a tetrahedral molecule with four equivalent C–H bonds) you will see that given scalar coupling constants are not equal (e.g. -11.257, -11.2548, -11.2543, -11.2541, -11.2548)\n\n`id,molecule_name,atom_index_0,atom_index_1,type,scalar_coupling_constant\n0,dsgdb9nsd_000001,1,0,1JHC,84.8076\n1,dsgdb9nsd_000001,1,2,2JHH,-11.257\n2,dsgdb9nsd_000001,1,3,2JHH,-11.2548\n3,dsgdb9nsd_000001,1,4,2JHH,-11.2543\n4,dsgdb9nsd_000001,2,0,1JHC,84.8074\n5,dsgdb9nsd_000001,2,3,2JHH,-11.2541\n6,dsgdb9nsd_000001,2,4,2JHH,-11.2548\n7,dsgdb9nsd_000001,3,0,1JHC,84.8093\n8,dsgdb9nsd_000001,3,4,2JHH,-11.2543\n`\nThe training data is noisy. |-11.257 - -11.2541| = 0.003 and log(0.003) = -5.8\nSo it's barely possible to get -5.8", "585456": "Brilliant observation. Thanks for sharing. 👍", "585588": "Could it be that \"our\" methane is not ideal tetrahedral? CH distances are not exactly equal...", "589500": ".", "589501": "sergeyzlobin for sure\nand such precise info of the difference might boost your score. I don't know how properly to perform that for now. @yassinealouini", "589560": "> Could it be that \"our\" methane is not ideal tetrahedral? CH distances are not exactly equal…\n\nIt is the result of a numerical simulation, and the fact that it is not a perfect tetrahedron illustrates the point @pavelgonchar makes. There is a little noise in the data which puts a lower limit on the accuracy of any predictive model that can be created.", "590440": "Aaand, finally, the -3 threshold has been broken!", "590486": "Congratulations your team broken -3.0 ,can I ask if your team get benifits from processing noise target?", "590546": "Yes that's a great achievement and a motivation for all of us to strive for better results.", "590626": "I think others also have passed this but they don't submit...", "590635": "Have you seen this behavior (in ebay it was called auction sniping, last-second subs) in other competitions?", "590649": "antorsae would you work harder if you knew that others have passed you?\n\nNo submission is my favorite technique to prevent LB overfitting.", "590763": "antorsae Yes I saw it, and I saw it from some of the people in top 10 here ;)", "590770": ">There is a little noise in the data which puts a lower limit on the accuracy of any predictive model that can be created.\n\nUnless the predictive model can also account for this breaking of symmetries", "590771": "antorsae @pavelgonchar I should have started with this: congrats for passing the -3 score! This is motivating ;)", "590772": "antorsae congrats with your new score. now you need to kidnap @outrunner before he submits. his 23 days break looks dangerous unless he is already kidnapped:)", "590863": "I'd kidnap Jaechang Lim too ... And anyone not having submitted in 5 days among to 10...", "591333": "> Unless the predictive model can also account for this breaking of symmetries\n\nHow would this be possible? For instance, how can you find how symmetry is broken int he first molecule (methane)?", "591401": "cpmpml , I guess one could argue that if the breaking of symmetry in the input variables is correlated to the breaking of the symmetry in the coupling a model might be able to exploit this correlation. \n\n> For instance, how can you find how symmetry is broken int he first molecule\n\ne.g. if the distances in methane vary in the same way the couplings do. \n\nIs that what you meant @borisdee ?", "591788": "There is another way to roughly estimate the upper bound using significant digits. Target's precision is less than 0.00005. log(0.00005)=-9.9", "592374": "Looks @outrunner has broken his chains" }, "source": "meta" }