A Picture paints a thousand words\n ![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F551520%2F7eb9e4ba8de761822d3ed98f73bf8a29%2FScreen%20Shot%202019-07-04%20at%203.29.57%20PM.png?generation=1562250888459956&alt=media\")
\n\n\nA Picture paints a thousand words\n
\n
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F551520%2F3fa3998aeed3843e07b2040074253d87%2FScreen%20Shot%202019-07-04%20at%203.38.13%20PM.png?generation=1562251118639667&alt=media\")
Well, yes :)\nI wrote about this long ago here: https://www.kaggle.com/artgor/using-meta-features-to-improve-model
", "rawMarkdown": "Well, yes :)\nI wrote about this long ago here: https://www.kaggle.com/artgor/using-meta-features-to-improve-model", "votes": 4, "replies": [ { "id": 568299, "postDate": "2019-07-04T17:11:27.827Z", "content": "You got it way early in the competition that's true, I saw your kernel it is very good. To learn about data I decided to do EDA from scratch again on my own with multiple sets merged this time and some tricks I learnt on the way.
\n\nAnyways interestingly I have found some more interesting insights in my own analysis such as negative correlation with feature pso the Pearson's correlation coefficient is -0.74. May be something in there will work on it in coming days.
Awesome kernel @artgor !
", "rawMarkdown": "Awesome kernel @artgor ! ", "votes": 1 } ] }, { "id": 568251, "postDate": "2019-07-04T15:44:39.783Z", "content": "Am I wrong to say that it is because\nscalar_coupling_constant is a sum of fc sd pso dso\nand the main part there is fc
", "rawMarkdown": "Am I wrong to say that it is because\nscalar_coupling_constant is a sum of fc sd pso dso\nand the main part there is fc", "votes": 3, "replies": [ { "id": 568297, "postDate": "2019-07-04T17:10:13.523Z", "rawMarkdown": "", "votes": 6, "isDeleted": true }, { "id": 568344, "postDate": "2019-07-04T18:33:56.660Z", "content": "I agree. I am still thinking how come the fermi coupling (fc) out of fold feature can improve the score so much. I still see it as predicting the target value and then using it to predict itself again. I am very curious, how it will generalize in the end.
", "rawMarkdown": "I agree. I am still thinking how come the fermi coupling (fc) out of fold feature can improve the score so much. I still see it as predicting the target value and then using it to predict itself again. I am very curious, how it will generalize in the end." }, { "id": 568364, "postDate": "2019-07-04T19:38:16.147Z", "content": "My bet - it's gonna shot you in the knee 🕵️ \nI don't know why people talk so much about this feature.. You just feeding your model with \"distorted\" target, I think it's wrong in general. But maybe I'm wrong tho
", "rawMarkdown": "My bet - it's gonna shot you in the knee 🕵️ \nI don't know why people talk so much about this feature.. You just feeding your model with \"distorted\" target, I think it's wrong in general. But maybe I'm wrong tho", "votes": 3 }, { "id": 568374, "postDate": "2019-07-04T19:49:44.243Z", "content": "Agreed. I don't use it and have the same opinion as you, but the effect on the LB score is interesting.
", "rawMarkdown": "Agreed. I don't use it and have the same opinion as you, but the effect on the LB score is interesting. " } ] }, { "id": 568465, "postDate": "2019-07-05T01:03:18.790Z", "content": "maybe because scalar_coupling_constant = fc + sd + pso + dso ?
scalar_coupling_contributions.csv - The scalar coupling constants in train.csv are a sum of four terms. scalar_coupling_contributions.csv contain all these terms. The first column (molecule_name) are the name of the molecule, the second (atom_index_0) and third column (atom_index_1) are the atom indices of the atom-pair, the fourth column indicates the type of coupling, the fifth column (fc) is the Fermi Contact contribution, the sixth column (sd) is the Spin-dipolar contribution, the seventh column (pso) is the Paramagnetic spin-orbit contribution and the eighth column (dso) is the Diamagnetic spin-orbit contribution.
scalar coupling constant = fc + sd + pso + dso is correct. but the contribution ratio is molecule dependent !
\n\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F113660%2F4a9c9f6d4414fa15ced57661bf7e4d50%2Fcontribution.png?generation=1562308993006731&alt=media\")
About scalar_coupling_contributions, this page describes it a little bit more.
\n\nScalar Coupling \nhttps://bouman.chem.georgetown.edu/nmr/scalar/scalar.htm
", "rawMarkdown": "About scalar_coupling_contributions, this page describes it a little bit more.\n\nScalar Coupling \nhttps://bouman.chem.georgetown.edu/nmr/scalar/scalar.htm", "votes": 1 }, { "id": 568215, "postDate": "2019-07-04T14:51:39.283Z", "content": "\n\n\nA Picture paints a thousand words\n
\n
Well, yes :)\nI wrote about this long ago here: https://www.kaggle.com/artgor/using-meta-features-to-improve-model
", "votes": 4, "replies": [ { "id": 568299, "author_name": "cyberia", "author_url": "", "post_date": "2019-07-04T17:11:27.827000", "content": "You got it way early in the competition that's true, I saw your kernel it is very good. To learn about data I decided to do EDA from scratch again on my own with multiple sets merged this time and some tricks I learnt on the way.
\n\nAnyways interestingly I have found some more interesting insights in my own analysis such as negative correlation with feature pso the Pearson's correlation coefficient is -0.74. May be something in there will work on it in coming days.
Awesome kernel @artgor !
", "votes": 1, "replies": [] } ] }, { "id": 568251, "author_name": "Konstantin Yakovlev", "author_url": "", "post_date": "2019-07-04T15:44:39.783000", "content": "Am I wrong to say that it is because\nscalar_coupling_constant is a sum of fc sd pso dso\nand the main part there is fc
", "votes": 3, "replies": [ { "id": 568297, "author_name": "", "author_url": "", "post_date": "2019-07-04T17:10:13.523000", "content": "", "votes": 6, "replies": [] }, { "id": 568344, "author_name": "somnambWl", "author_url": "", "post_date": "2019-07-04T18:33:56.660000", "content": "I agree. I am still thinking how come the fermi coupling (fc) out of fold feature can improve the score so much. I still see it as predicting the target value and then using it to predict itself again. I am very curious, how it will generalize in the end.
", "votes": 0, "replies": [] }, { "id": 568364, "author_name": "Stanislav Blinov", "author_url": "", "post_date": "2019-07-04T19:38:16.147000", "content": "My bet - it's gonna shot you in the knee 🕵️ \nI don't know why people talk so much about this feature.. You just feeding your model with \"distorted\" target, I think it's wrong in general. But maybe I'm wrong tho
", "votes": 3, "replies": [] }, { "id": 568374, "author_name": "somnambWl", "author_url": "", "post_date": "2019-07-04T19:49:44.243000", "content": "Agreed. I don't use it and have the same opinion as you, but the effect on the LB score is interesting.
", "votes": 0, "replies": [] } ] }, { "id": 568465, "author_name": "Giba", "author_url": "", "post_date": "2019-07-05T01:03:18.790000", "content": "maybe because scalar_coupling_constant = fc + sd + pso + dso ?
scalar_coupling_contributions.csv - The scalar coupling constants in train.csv are a sum of four terms. scalar_coupling_contributions.csv contain all these terms. The first column (molecule_name) are the name of the molecule, the second (atom_index_0) and third column (atom_index_1) are the atom indices of the atom-pair, the fourth column indicates the type of coupling, the fifth column (fc) is the Fermi Contact contribution, the sixth column (sd) is the Spin-dipolar contribution, the seventh column (pso) is the Paramagnetic spin-orbit contribution and the eighth column (dso) is the Diamagnetic spin-orbit contribution.
scalar coupling constant = fc + sd + pso + dso is correct. but the contribution ratio is molecule dependent !
\n\n
About scalar_coupling_contributions, this page describes it a little bit more.
\n\nScalar Coupling \nhttps://bouman.chem.georgetown.edu/nmr/scalar/scalar.htm
", "votes": 1, "replies": [] } ], "raw_markdown_by_id": { "568292": "Well, yes :)\nI wrote about this long ago here: https://www.kaggle.com/artgor/using-meta-features-to-improve-model", "568251": "Am I wrong to say that it is because\nscalar_coupling_constant is a sum of fc sd pso dso\nand the main part there is fc", "568465": "maybe because `scalar_coupling_constant = fc + sd + pso + dso` ?\n\n`scalar_coupling_contributions.csv - The scalar coupling constants in train.csv are a sum of four terms. scalar_coupling_contributions.csv contain all these terms. The first column (molecule_name) are the name of the molecule, the second (atom_index_0) and third column (atom_index_1) are the atom indices of the atom-pair, the fourth column indicates the type of coupling, the fifth column (fc) is the Fermi Contact contribution, the sixth column (sd) is the Spin-dipolar contribution, the seventh column (pso) is the Paramagnetic spin-orbit contribution and the eighth column (dso) is the Diamagnetic spin-orbit contribution.`", "568487": "About scalar_coupling_contributions, this page describes it a little bit more.\n\nScalar Coupling \nhttps://bouman.chem.georgetown.edu/nmr/scalar/scalar.htm", "568215": "> A Picture paints a thousand words\n\n\n\n" } }