Hi!!! anyone can explain me why the mullinken charge computed with the library https://mendeleev.readthedocs.io/en/latest/data.html#mulliken-en differs from the mulliken charge provided in the file mulliken_charge.csv.
", "messages": [ { "id": "555769", "postDate": "06/19/2019 12:11:44", "content": "Hi!!! anyone can explain me why the mullinken charge computed with the library https://mendeleev.readthedocs.io/en/latest/data.html#mulliken-en differs from the mulliken charge provided in the file mulliken_charge.csv.
", "rawMarkdown": "Hi!!! anyone can explain me why the mullinken charge computed with the library https://mendeleev.readthedocs.io/en/latest/data.html#mulliken-en differs from the mulliken charge provided in the file mulliken_charge.csv.", "votes": null }, { "id": "555911", "postDate": "06/19/2019 15:43:25", "content": "Just guessing - but here is the theory I am working on.
\n\nSo if the supplemental data, such as mulliken, is different from published results that tells me that it can be calculated using QM methods.
\n\nIf you had found that in fact the published mulliken values are the same as in train than it would have been a simple (NOT ML) task to find and fill the test data.
", "rawMarkdown": "Just guessing - but here is the theory I am working on.\n\n1. Coupling values for all the molecules we are working with are known and published. To avoid a huge leak the training data we are using has been calculated for an unknown temperature (some including me assume 0K) and unknown other variables.\n2. The change in temperature and other variables will also affect the properties of some/all of the supplement files we have been given. So the same or similar method to calculate the coupling values was used to adjust the supplemental data. \n3. Rather than focus on machine learning some folks want to cheat the system - hence the need to create a realistic, but not public set of data.\n\n\nSo if the supplemental data, such as mulliken, is different from published results that tells me that it can be calculated using QM methods. \n\nIf you had found that in fact the published mulliken values are the same as in train than it would have been a simple (NOT ML) task to find and fill the test data.", "votes": null }, { "id": "555937", "postDate": "06/19/2019 16:08:17", "content": "@pcjimmmy - I think your theory is correct.
\n\nI recently used the mulliken constants generated in another script to supply to the test data, but used the competition-supplied values for training. The CV of the model looked very promising, but my leaderboard score was almost a point higher, so the model dramatically underfit the test data. At least that is what I think is going on.
", "rawMarkdown": "pcjimmmy - I think your theory is correct. \n\nI recently used the mulliken constants generated in another script to supply to the test data, but used the competition-supplied values for training. The CV of the model looked very promising, but my leaderboard score was almost a point higher, so the model dramatically underfit the test data. At least that is what I think is going on.", "votes": null }, { "id": "557431", "postDate": "06/21/2019 06:58:43", "content": "You are comparing Mulliken charges with Mulliken electronegativity.
\n\nMulliken, great chemist (physicist), did a lot of work. So take care what you are looking at when you see \"Mulliken\"
\n\nEdit: Mulliken charges highly depend on the used funcional and basis set, so different QM methods give different results.\nOne example for H2O:\nHF/def2-SVP\nH: 0.2352\nO: -0.4704\nH: 0.2352
\n\nHF/def2-TZVPP\nH: 0.1801\nO: -0.3602\nH: 0.1801
\n\nB3LYP/def2-SVP\nH: 0.1481\nO: -0.2962\nH: 0.1481
\n\nP.s. I hope this QM calculation was okay to give other people more insight.
", "rawMarkdown": "You are comparing Mulliken charges with Mulliken electronegativity.\n\nMulliken, great chemist (physicist), did a lot of work. So take care what you are looking at when you see \"Mulliken\"\n\nEdit: Mulliken charges highly depend on the used funcional and basis set, so different QM methods give different results.\nOne example for H2O:\nHF/def2-SVP\nH: 0.2352\nO: -0.4704\nH: 0.2352\n\nHF/def2-TZVPP\nH: 0.1801\nO: -0.3602\nH: 0.1801\n\nB3LYP/def2-SVP\nH: 0.1481\nO: -0.2962\nH: 0.1481\n\nP.s. I hope this QM calculation was okay to give other people more insight.", "votes": null }, { "id": "557445", "postDate": "06/21/2019 07:08:54", "content": "Because it's not the same thing :)
", "rawMarkdown": "Because it's not the same thing :)", "votes": null }, { "id": "558616", "postDate": "06/22/2019 16:34:32", "content": "Molecules of test.csv isn't present in mulliken_charge.csv or any other additional files.
", "rawMarkdown": "Molecules of test.csv isn't present in mulliken_charge.csv or any other additional files.", "votes": null }, { "id": "561460", "postDate": "06/26/2019 13:14:00", "content": "Quite alright @mrhermann
", "rawMarkdown": "Quite alright @mrhermann", "votes": null } ], "comments": [ { "id": 555911, "author_name": "pcjimmmy", "author_url": "", "post_date": "06/19/2019 15:43:25", "content": "Just guessing - but here is the theory I am working on.
\n\nSo if the supplemental data, such as mulliken, is different from published results that tells me that it can be calculated using QM methods.
\n\nIf you had found that in fact the published mulliken values are the same as in train than it would have been a simple (NOT ML) task to find and fill the test data.
", "votes": null, "replies": [ { "id": 555937, "author_name": "wjholst", "author_url": "", "post_date": "06/19/2019 16:08:17", "content": "@pcjimmmy - I think your theory is correct.
\n\nI recently used the mulliken constants generated in another script to supply to the test data, but used the competition-supplied values for training. The CV of the model looked very promising, but my leaderboard score was almost a point higher, so the model dramatically underfit the test data. At least that is what I think is going on.
", "votes": null, "replies": [] } ] }, { "id": 557431, "author_name": "mrhermann", "author_url": "", "post_date": "06/21/2019 06:58:43", "content": "You are comparing Mulliken charges with Mulliken electronegativity.
\n\nMulliken, great chemist (physicist), did a lot of work. So take care what you are looking at when you see \"Mulliken\"
\n\nEdit: Mulliken charges highly depend on the used funcional and basis set, so different QM methods give different results.\nOne example for H2O:\nHF/def2-SVP\nH: 0.2352\nO: -0.4704\nH: 0.2352
\n\nHF/def2-TZVPP\nH: 0.1801\nO: -0.3602\nH: 0.1801
\n\nB3LYP/def2-SVP\nH: 0.1481\nO: -0.2962\nH: 0.1481
\n\nP.s. I hope this QM calculation was okay to give other people more insight.
", "votes": null, "replies": [ { "id": 561460, "author_name": "larsbratholm", "author_url": "", "post_date": "06/26/2019 13:14:00", "content": "Quite alright @mrhermann
", "votes": null, "replies": [] } ] }, { "id": 557445, "author_name": "borisdee", "author_url": "", "post_date": "06/21/2019 07:08:54", "content": "Because it's not the same thing :)
", "votes": null, "replies": [] }, { "id": 558616, "author_name": "imansoor", "author_url": "", "post_date": "06/22/2019 16:34:32", "content": "Molecules of test.csv isn't present in mulliken_charge.csv or any other additional files.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "555769": "Hi!!! anyone can explain me why the mullinken charge computed with the library https://mendeleev.readthedocs.io/en/latest/data.html#mulliken-en differs from the mulliken charge provided in the file mulliken_charge.csv.", "555911": "Just guessing - but here is the theory I am working on.\n\n1. Coupling values for all the molecules we are working with are known and published. To avoid a huge leak the training data we are using has been calculated for an unknown temperature (some including me assume 0K) and unknown other variables.\n2. The change in temperature and other variables will also affect the properties of some/all of the supplement files we have been given. So the same or similar method to calculate the coupling values was used to adjust the supplemental data. \n3. Rather than focus on machine learning some folks want to cheat the system - hence the need to create a realistic, but not public set of data.\n\n\nSo if the supplemental data, such as mulliken, is different from published results that tells me that it can be calculated using QM methods. \n\nIf you had found that in fact the published mulliken values are the same as in train than it would have been a simple (NOT ML) task to find and fill the test data.", "555937": "pcjimmmy - I think your theory is correct. \n\nI recently used the mulliken constants generated in another script to supply to the test data, but used the competition-supplied values for training. The CV of the model looked very promising, but my leaderboard score was almost a point higher, so the model dramatically underfit the test data. At least that is what I think is going on.", "557431": "You are comparing Mulliken charges with Mulliken electronegativity.\n\nMulliken, great chemist (physicist), did a lot of work. So take care what you are looking at when you see \"Mulliken\"\n\nEdit: Mulliken charges highly depend on the used funcional and basis set, so different QM methods give different results.\nOne example for H2O:\nHF/def2-SVP\nH: 0.2352\nO: -0.4704\nH: 0.2352\n\nHF/def2-TZVPP\nH: 0.1801\nO: -0.3602\nH: 0.1801\n\nB3LYP/def2-SVP\nH: 0.1481\nO: -0.2962\nH: 0.1481\n\nP.s. I hope this QM calculation was okay to give other people more insight.", "557445": "Because it's not the same thing :)", "558616": "Molecules of test.csv isn't present in mulliken_charge.csv or any other additional files.", "561460": "Quite alright @mrhermann" }, "source": "meta" }