![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F801574%2F81147eab13529bfb5bfbf1cbc4c7637d%2Ff.png?generation=1565419691804074&alt=media\")
Averaging ranks of all types:
\n\n
Averaging ranks of all types:
\n\n
I don't think it is a good idea to average feature importance for LGB over all types.\nDifferent types have different important features.
", "rawMarkdown": "I don't think it is a good idea to average feature importance for LGB over all types.\nDifferent types have different important features.", "votes": null }, { "id": "596380", "postDate": "08/10/2019 15:03:26", "content": "What does *milliken_atom_0_qm9* mean ? If it is milliken_charge, is it not available for test data...
", "rawMarkdown": "What does *milliken_atom_0_qm9* mean ? If it is milliken_charge, is it not available for test data...", "votes": null }, { "id": "596729", "postDate": "08/11/2019 07:30:32", "content": "Yes Toshi, no milliencharge available for test...What I did was using the inference values from another model for milliken charge with OOB sample. It doesn't always stay at the top features as the inference model is not that good...More consistent with multiple output NN model, however.
", "rawMarkdown": "Yes Toshi, no milliencharge available for test...What I did was using the inference values from another model for milliken charge with OOB sample. It doesn't always stay at the top features as the inference model is not that good...More consistent with multiple output NN model, however.", "votes": null }, { "id": "596779", "postDate": "08/11/2019 09:32:58", "content": "mulliken charge is avaiable for test in QM9 dataset.
", "rawMarkdown": "mulliken charge is avaiable for test in QM9 dataset.", "votes": null }, { "id": "597301", "postDate": "08/12/2019 06:37:22", "content": "CPMP, yes but does it break the rules ? Can we use QM9 for prediction too ?
\n\n\n", "rawMarkdown": "CPMP, yes but does it break the rules ? Can we use QM9 for prediction too ?\n\n> To address this current competition, we are updating the rules to allow the use of QM7 and QM9 published data, if participants wish to use it. Participants still may not use quantum calculations to obtain predictions.", "votes": null }, { "id": "597455", "postDate": "08/12/2019 11:16:15", "content": "To address this current competition, we are updating the rules to allow the use of QM7 and QM9 published data, if participants wish to use it. Participants still may not use quantum calculations to obtain predictions.
\n
toshi_k, they say we can use the data.
", "rawMarkdown": "toshi_k, they say we can use the data.", "votes": null }, { "id": "597616", "postDate": "08/12/2019 15:07:24", "content": "Are Mulliken charges dependent on temperature or not? I had a feeling that QM9 is computed for a bit different temperature so QM9 properties are not transferable.
\n\nI had also a feeling that CPMP said he won't dig into that and now he suggests it (I mean QM9).
\n\nAnyway, there are kernels which can get you Mulliken charges in an allowed way, just look for them ;)
", "rawMarkdown": "Are Mulliken charges dependent on temperature or not? I had a feeling that QM9 is computed for a bit different temperature so QM9 properties are not transferable.\n\nI had also a feeling that CPMP said he won't dig into that and now he suggests it (I mean QM9).\n\nAnyway, there are kernels which can get you Mulliken charges in an allowed way, just look for them ;)", "votes": null }, { "id": "597806", "postDate": "08/12/2019 18:49:22", "content": "Don't read more into what I wrote than what I wrote. I stated a mere fact.
", "rawMarkdown": "Don't read more into what I wrote than what I wrote. I stated a mere fact.", "votes": null } ], "comments": [ { "id": 596101, "author_name": "artgor", "author_url": "", "post_date": "08/10/2019 06:58:20", "content": "I don't think it is a good idea to average feature importance for LGB over all types.\nDifferent types have different important features.
", "votes": null, "replies": [] }, { "id": 596380, "author_name": "toshik", "author_url": "", "post_date": "08/10/2019 15:03:26", "content": "What does *milliken_atom_0_qm9* mean ? If it is milliken_charge, is it not available for test data...
", "votes": null, "replies": [ { "id": 596729, "author_name": "lpachuong", "author_url": "", "post_date": "08/11/2019 07:30:32", "content": "Yes Toshi, no milliencharge available for test...What I did was using the inference values from another model for milliken charge with OOB sample. It doesn't always stay at the top features as the inference model is not that good...More consistent with multiple output NN model, however.
", "votes": null, "replies": [] }, { "id": 596779, "author_name": "cpmpml", "author_url": "", "post_date": "08/11/2019 09:32:58", "content": "mulliken charge is avaiable for test in QM9 dataset.
", "votes": null, "replies": [] }, { "id": 597301, "author_name": "toshik", "author_url": "", "post_date": "08/12/2019 06:37:22", "content": "CPMP, yes but does it break the rules ? Can we use QM9 for prediction too ?
\n\n\n", "votes": null, "replies": [] }, { "id": 597455, "author_name": "cpmpml", "author_url": "", "post_date": "08/12/2019 11:16:15", "content": "To address this current competition, we are updating the rules to allow the use of QM7 and QM9 published data, if participants wish to use it. Participants still may not use quantum calculations to obtain predictions.
\n
toshi_k, they say we can use the data.
", "votes": null, "replies": [] }, { "id": 597616, "author_name": "somnambwl", "author_url": "", "post_date": "08/12/2019 15:07:24", "content": "Are Mulliken charges dependent on temperature or not? I had a feeling that QM9 is computed for a bit different temperature so QM9 properties are not transferable.
\n\nI had also a feeling that CPMP said he won't dig into that and now he suggests it (I mean QM9).
\n\nAnyway, there are kernels which can get you Mulliken charges in an allowed way, just look for them ;)
", "votes": null, "replies": [] }, { "id": 597806, "author_name": "cpmpml", "author_url": "", "post_date": "08/12/2019 18:49:22", "content": "Don't read more into what I wrote than what I wrote. I stated a mere fact.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "596096": "Averaging ranks of all types:\n\n", "596101": "I don't think it is a good idea to average feature importance for LGB over all types.\nDifferent types have different important features.", "596380": "What does *milliken_atom_0_qm9* mean ? If it is milliken_charge, is it not available for test data...", "596729": "Yes Toshi, no milliencharge available for test...What I did was using the inference values from another model for milliken charge with OOB sample. It doesn't always stay at the top features as the inference model is not that good...More consistent with multiple output NN model, however.", "596779": "mulliken charge is avaiable for test in QM9 dataset.", "597301": "CPMP, yes but does it break the rules ? Can we use QM9 for prediction too ?\n\n> To address this current competition, we are updating the rules to allow the use of QM7 and QM9 published data, if participants wish to use it. Participants still may not use quantum calculations to obtain predictions.", "597455": "toshi_k, they say we can use the data.", "597616": "Are Mulliken charges dependent on temperature or not? I had a feeling that QM9 is computed for a bit different temperature so QM9 properties are not transferable.\n\nI had also a feeling that CPMP said he won't dig into that and now he suggests it (I mean QM9).\n\nAnyway, there are kernels which can get you Mulliken charges in an allowed way, just look for them ;)", "597806": "Don't read more into what I wrote than what I wrote. I stated a mere fact." }, "source": "meta" }