{
  "id": 71808,
  "title": "Some specialist on top of LB",
  "url": "/competitions/PLAsTiCC-2018/discussion/71808",
  "author_name": "",
  "post_date": "2018-11-16T18:27:23.536100500Z",
  "votes": 19,
  "comment_count": 14,
  "views": 0,
  "content": "<p>It seems that our current leader is doing a PhD on... Supernova cosmology in Saul Perlmutter's group</p>\n\n<p>See <a href=\"https://www.linkedin.com/in/kyle-boone-28aa0120/\">https://www.linkedin.com/in/kyle-boone-28aa0120/</a></p>\n\n<p>This makes me think that there are magic features ;)  I haven't found/used any so far, but this is motivating!</p>",
  "messages": [
    {
      "id": "422756",
      "postDate": "11/16/2018 18:27:23",
      "content": "<p>It seems that our current leader is doing a PhD on... Supernova cosmology in Saul Perlmutter's group</p>\n\n<p>See <a href=\"https://www.linkedin.com/in/kyle-boone-28aa0120/\">https://www.linkedin.com/in/kyle-boone-28aa0120/</a></p>\n\n<p>This makes me think that there are magic features ;)  I haven't found/used any so far, but this is motivating!</p>",
      "rawMarkdown": "It seems that our current leader is doing a PhD on... Supernova cosmology in Saul Perlmutter's group\n\nSee https://www.linkedin.com/in/kyle-boone-28aa0120/\n\nThis makes me think that there are magic features ;)  I haven't found/used any so far, but this is motivating!",
      "votes": null
    },
    {
      "id": "422783",
      "postDate": "11/16/2018 19:20:04",
      "content": "<p>I think he should give handicap to us!</p>",
      "rawMarkdown": "I think he should give handicap to us!",
      "votes": null
    },
    {
      "id": "422788",
      "postDate": "11/16/2018 19:25:08",
      "content": "<p>Note that you can form teams that include astronomers! We're mostly harmless!\nAnd you are still eligible for the prizes provided it is not one of the astronomers here (without an asterisk by their names):</p>\n\n<p><a href=\"https://www.kaggle.com/c/PLAsTiCC-2018#PLAsTiCC\">https://www.kaggle.com/c/PLAsTiCC-2018#PLAsTiCC</a>'s%20Team</p>\n\n<p>Cheers,</p>\n\n<p>-Gautham for the PLAsTiCC team</p>",
      "rawMarkdown": "Note that you can form teams that include astronomers! We're mostly harmless!\nAnd you are still eligible for the prizes provided it is not one of the astronomers here (without an asterisk by their names):\n\nhttps://www.kaggle.com/c/PLAsTiCC-2018#PLAsTiCC's%20Team\n\nCheers,\n\n-Gautham for the PLAsTiCC team",
      "votes": null
    },
    {
      "id": "422803",
      "postDate": "11/16/2018 19:47:58",
      "content": "<p>Haha I've been found out. I actually haven't worked on classification of transients before this, so this has been a learning experience for me too. It will be interesting to see if domain knowledge helps in the end. I have made some choices in my feature extraction based off of my intuition, but I'm not using any external data or models, so we should all be on the same footing!</p>",
      "rawMarkdown": "Haha I've been found out. I actually haven't worked on classification of transients before this, so this has been a learning experience for me too. It will be interesting to see if domain knowledge helps in the end. I have made some choices in my feature extraction based off of my intuition, but I'm not using any external data or models, so we should all be on the same footing!",
      "votes": null
    },
    {
      "id": "422831",
      "postDate": "11/16/2018 20:59:23",
      "content": "<p>You are a fast learner then ;)</p>",
      "rawMarkdown": "You are a fast learner then ;)",
      "votes": null
    },
    {
      "id": "422859",
      "postDate": "11/16/2018 22:28:55",
      "content": "<p>We can only team if they come and enter the competition ;)</p>",
      "rawMarkdown": "We can only team if they come and enter the competition ;)",
      "votes": null
    },
    {
      "id": "423056",
      "postDate": "11/17/2018 11:08:14",
      "content": "<p>Indeed, what is in his head would be external data for us ! ;)</p>",
      "rawMarkdown": "Indeed, what is in his head would be external data for us ! ;)",
      "votes": null
    },
    {
      "id": "423217",
      "postDate": "11/17/2018 18:10:49",
      "content": "<p>Good motivation! :)</p>",
      "rawMarkdown": "Good motivation! :)",
      "votes": null
    },
    {
      "id": "423310",
      "postDate": "11/17/2018 22:10:08",
      "content": "<p>@CPMP</p>\n\n<p>You have a lot of external data on the internet, really.</p>",
      "rawMarkdown": "CPMP\n\nYou have a lot of external data on the internet, really.",
      "votes": null
    },
    {
      "id": "423456",
      "postDate": "11/18/2018 09:43:56",
      "content": "<blockquote>\n  <p>You have a lot of external data on the internet, really.</p>\n</blockquote>\n\n<p>Yes, but unlabelled.  Only astronomers can identify the classes present in train, then look for additional data for these classes.  They can even try to identify which new sources are sought for in class 99.  I can't.</p>",
      "rawMarkdown": "&gt; You have a lot of external data on the internet, really.\n\nYes, but unlabelled.  Only astronomers can identify the classes present in train, then look for additional data for these classes.  They can even try to identify which new sources are sought for in class 99.  I can't.",
      "votes": null
    },
    {
      "id": "423505",
      "postDate": "11/18/2018 12:22:22",
      "content": "<p>@CPMP</p>\n\n<p><em><strong>Only astronomers can identify the classes...</strong></em></p>\n\n<p>You may be right on this point.</p>",
      "rawMarkdown": "CPMP\n\n***Only astronomers can identify the classes...***\n\nYou may be right on this point.",
      "votes": null
    },
    {
      "id": "423846",
      "postDate": "11/19/2018 06:19:23",
      "content": "<p>Honestly, identifying the classes doesn't really help that much. For example, class 90 seems to be Type Ia supernovae, but there are a lot of weird things in that class that don't look like the simple models that we usually use. You could try to make your own simulations (for example, see SALT2 at <a href=\"http://supernovae.in2p3.fr/salt/doku.php\">http://supernovae.in2p3.fr/salt/doku.php</a> which is a nice linear model of Type Ia supernovae), but you would spend a lot of time tuning it to match the PLAsTiCC dataset and I don't think that you would cover the full range of what is actually in the dataset. I'm not sure about this, but it seems like the contest organizers modeled some peculiar faint Type Ia supernovae (maybe eg: <a href=\"https://en.wikipedia.org/wiki/SN_2002cx\">https://en.wikipedia.org/wiki/SN_2002cx</a>) and added those in to class 90 too. A SALT2 simulated dataset wouldn't even include those.</p>\n\n<p>There is basically no way to reverse engineer what the contest organizers did without spending an inordinate amount of time on this, so your best bet is to try to fit a model to the data that they gave us rather than trying to simulate more of your own.</p>",
      "rawMarkdown": "Honestly, identifying the classes doesn't really help that much. For example, class 90 seems to be Type Ia supernovae, but there are a lot of weird things in that class that don't look like the simple models that we usually use. You could try to make your own simulations (for example, see SALT2 at http://supernovae.in2p3.fr/salt/doku.php which is a nice linear model of Type Ia supernovae), but you would spend a lot of time tuning it to match the PLAsTiCC dataset and I don't think that you would cover the full range of what is actually in the dataset. I'm not sure about this, but it seems like the contest organizers modeled some peculiar faint Type Ia supernovae (maybe eg: https://en.wikipedia.org/wiki/SN_2002cx) and added those in to class 90 too. A SALT2 simulated dataset wouldn't even include those.\n\nThere is basically no way to reverse engineer what the contest organizers did without spending an inordinate amount of time on this, so your best bet is to try to fit a model to the data that they gave us rather than trying to simulate more of your own.",
      "votes": null
    },
    {
      "id": "424009",
      "postDate": "11/19/2018 12:22:44",
      "content": "<p>@Kyle Boone</p>\n\n<p>Thanks a lot for this clarification and for the links.</p>\n\n<p>As I imagine, Kaggle did not reproduce an existing model, but when an informed person confirms the fact, that's allways better.</p>",
      "rawMarkdown": "Kyle Boone\n\nThanks a lot for this clarification and for the links.\n\nAs I imagine, Kaggle did not reproduce an existing model, but when an informed person confirms the fact, that's allways better.",
      "votes": null
    },
    {
      "id": "424114",
      "postDate": "11/19/2018 15:35:17",
      "content": "<p>Thanks for the valuable information! :)</p>",
      "rawMarkdown": "Thanks for the valuable information! :)",
      "votes": null
    },
    {
      "id": "427059",
      "postDate": "11/24/2018 13:12:03",
      "content": "<p><a href=\"/kyleboone\">@kyleboone</a> </p>\n\n<p>Thank you for the info, we also thought 90 is nova. Maybe you can share your opinion what class 52 is, is it supernova Type Ib ? We have a trouble with that class, would be helpful to know what it is at least... </p>",
      "rawMarkdown": "kyleboone \n\nThank you for the info, we also thought 90 is nova. Maybe you can share your opinion what class 52 is, is it supernova Type Ib ? We have a trouble with that class, would be helpful to know what it is at least...",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 422783,
      "author_name": "mamasinkgs",
      "author_url": "",
      "post_date": "11/16/2018 19:20:04",
      "content": "<p>I think he should give handicap to us!</p>",
      "votes": null,
      "replies": [
        {
          "id": 423056,
          "author_name": "cpmpml",
          "author_url": "",
          "post_date": "11/17/2018 11:08:14",
          "content": "<p>Indeed, what is in his head would be external data for us ! ;)</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 423310,
          "author_name": "pfichou",
          "author_url": "",
          "post_date": "11/17/2018 22:10:08",
          "content": "<p>@CPMP</p>\n\n<p>You have a lot of external data on the internet, really.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 423456,
          "author_name": "cpmpml",
          "author_url": "",
          "post_date": "11/18/2018 09:43:56",
          "content": "<blockquote>\n  <p>You have a lot of external data on the internet, really.</p>\n</blockquote>\n\n<p>Yes, but unlabelled.  Only astronomers can identify the classes present in train, then look for additional data for these classes.  They can even try to identify which new sources are sought for in class 99.  I can't.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 423505,
          "author_name": "pfichou",
          "author_url": "",
          "post_date": "11/18/2018 12:22:22",
          "content": "<p>@CPMP</p>\n\n<p><em><strong>Only astronomers can identify the classes...</strong></em></p>\n\n<p>You may be right on this point.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 423846,
          "author_name": "kyleboone",
          "author_url": "",
          "post_date": "11/19/2018 06:19:23",
          "content": "<p>Honestly, identifying the classes doesn't really help that much. For example, class 90 seems to be Type Ia supernovae, but there are a lot of weird things in that class that don't look like the simple models that we usually use. You could try to make your own simulations (for example, see SALT2 at <a href=\"http://supernovae.in2p3.fr/salt/doku.php\">http://supernovae.in2p3.fr/salt/doku.php</a> which is a nice linear model of Type Ia supernovae), but you would spend a lot of time tuning it to match the PLAsTiCC dataset and I don't think that you would cover the full range of what is actually in the dataset. I'm not sure about this, but it seems like the contest organizers modeled some peculiar faint Type Ia supernovae (maybe eg: <a href=\"https://en.wikipedia.org/wiki/SN_2002cx\">https://en.wikipedia.org/wiki/SN_2002cx</a>) and added those in to class 90 too. A SALT2 simulated dataset wouldn't even include those.</p>\n\n<p>There is basically no way to reverse engineer what the contest organizers did without spending an inordinate amount of time on this, so your best bet is to try to fit a model to the data that they gave us rather than trying to simulate more of your own.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 424009,
          "author_name": "pfichou",
          "author_url": "",
          "post_date": "11/19/2018 12:22:44",
          "content": "<p>@Kyle Boone</p>\n\n<p>Thanks a lot for this clarification and for the links.</p>\n\n<p>As I imagine, Kaggle did not reproduce an existing model, but when an informed person confirms the fact, that's allways better.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 424114,
          "author_name": "mamasinkgs",
          "author_url": "",
          "post_date": "11/19/2018 15:35:17",
          "content": "<p>Thanks for the valuable information! :)</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 427059,
          "author_name": "blondinka",
          "author_url": "",
          "post_date": "11/24/2018 13:12:03",
          "content": "<p><a href=\"/kyleboone\">@kyleboone</a> </p>\n\n<p>Thank you for the info, we also thought 90 is nova. Maybe you can share your opinion what class 52 is, is it supernova Type Ib ? We have a trouble with that class, would be helpful to know what it is at least... </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 422788,
      "author_name": "gsnarayan",
      "author_url": "",
      "post_date": "11/16/2018 19:25:08",
      "content": "<p>Note that you can form teams that include astronomers! We're mostly harmless!\nAnd you are still eligible for the prizes provided it is not one of the astronomers here (without an asterisk by their names):</p>\n\n<p><a href=\"https://www.kaggle.com/c/PLAsTiCC-2018#PLAsTiCC\">https://www.kaggle.com/c/PLAsTiCC-2018#PLAsTiCC</a>'s%20Team</p>\n\n<p>Cheers,</p>\n\n<p>-Gautham for the PLAsTiCC team</p>",
      "votes": null,
      "replies": [
        {
          "id": 422859,
          "author_name": "cpmpml",
          "author_url": "",
          "post_date": "11/16/2018 22:28:55",
          "content": "<p>We can only team if they come and enter the competition ;)</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 422803,
      "author_name": "kyleboone",
      "author_url": "",
      "post_date": "11/16/2018 19:47:58",
      "content": "<p>Haha I've been found out. I actually haven't worked on classification of transients before this, so this has been a learning experience for me too. It will be interesting to see if domain knowledge helps in the end. I have made some choices in my feature extraction based off of my intuition, but I'm not using any external data or models, so we should all be on the same footing!</p>",
      "votes": null,
      "replies": [
        {
          "id": 422831,
          "author_name": "cpmpml",
          "author_url": "",
          "post_date": "11/16/2018 20:59:23",
          "content": "<p>You are a fast learner then ;)</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 423217,
      "author_name": "darbin",
      "author_url": "",
      "post_date": "11/17/2018 18:10:49",
      "content": "<p>Good motivation! :)</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "422756": "It seems that our current leader is doing a PhD on... Supernova cosmology in Saul Perlmutter's group\n\nSee https://www.linkedin.com/in/kyle-boone-28aa0120/\n\nThis makes me think that there are magic features ;)  I haven't found/used any so far, but this is motivating!",
    "422783": "I think he should give handicap to us!",
    "422788": "Note that you can form teams that include astronomers! We're mostly harmless!\nAnd you are still eligible for the prizes provided it is not one of the astronomers here (without an asterisk by their names):\n\nhttps://www.kaggle.com/c/PLAsTiCC-2018#PLAsTiCC's%20Team\n\nCheers,\n\n-Gautham for the PLAsTiCC team",
    "422803": "Haha I've been found out. I actually haven't worked on classification of transients before this, so this has been a learning experience for me too. It will be interesting to see if domain knowledge helps in the end. I have made some choices in my feature extraction based off of my intuition, but I'm not using any external data or models, so we should all be on the same footing!",
    "422831": "You are a fast learner then ;)",
    "422859": "We can only team if they come and enter the competition ;)",
    "423056": "Indeed, what is in his head would be external data for us ! ;)",
    "423217": "Good motivation! :)",
    "423310": "CPMP\n\nYou have a lot of external data on the internet, really.",
    "423456": "&gt; You have a lot of external data on the internet, really.\n\nYes, but unlabelled.  Only astronomers can identify the classes present in train, then look for additional data for these classes.  They can even try to identify which new sources are sought for in class 99.  I can't.",
    "423505": "CPMP\n\n***Only astronomers can identify the classes...***\n\nYou may be right on this point.",
    "423846": "Honestly, identifying the classes doesn't really help that much. For example, class 90 seems to be Type Ia supernovae, but there are a lot of weird things in that class that don't look like the simple models that we usually use. You could try to make your own simulations (for example, see SALT2 at http://supernovae.in2p3.fr/salt/doku.php which is a nice linear model of Type Ia supernovae), but you would spend a lot of time tuning it to match the PLAsTiCC dataset and I don't think that you would cover the full range of what is actually in the dataset. I'm not sure about this, but it seems like the contest organizers modeled some peculiar faint Type Ia supernovae (maybe eg: https://en.wikipedia.org/wiki/SN_2002cx) and added those in to class 90 too. A SALT2 simulated dataset wouldn't even include those.\n\nThere is basically no way to reverse engineer what the contest organizers did without spending an inordinate amount of time on this, so your best bet is to try to fit a model to the data that they gave us rather than trying to simulate more of your own.",
    "424009": "Kyle Boone\n\nThanks a lot for this clarification and for the links.\n\nAs I imagine, Kaggle did not reproduce an existing model, but when an informed person confirms the fact, that's allways better.",
    "424114": "Thanks for the valuable information! :)",
    "427059": "kyleboone \n\nThank you for the info, we also thought 90 is nova. Maybe you can share your opinion what class 52 is, is it supernova Type Ib ? We have a trouble with that class, would be helpful to know what it is at least..."
  },
  "source": "meta"
}