![\"oliver\"](\"https://www.kaggleusercontent.com/kf/7397796/eyJhbGciOiJkaXIiLCJlbmMiOiJBMTI4Q0JDLUhTMjU2In0..VILcCAgDh7swBpC4AlEiUg.dZtGzalQCMPNsWE_c1Jw1Ko9QwmVawtVm3ywCRpTFfcq2IpbylcJxjRV3upf26db-PLyPhtwzEXWJFNNz9PebllcjDznW8Nn5h70PjBmtuyzPA7g_vDhOc3H6KWwZDraW3AHmd5LqHr5bgTVDJVOaFsr1piZ_TMjgYhaGq2Rh5s.0F8imgo8WcWcXtNVDgYB4w/importances.png\")
We've seen hostgal_photoz is the most important feature. But why?
\n\nLooking at the non-zero photoz distributions, it seems every class is evenly distributed over the photoz space with a single exception; thus one would assume there are some good FI's going on. From an astronomy point of view, how exactly is redshift/distance used to interpret photometry light curves?
", "messages": [ { "id": 421305, "postDate": "2018-11-14T22:02:46.980Z", "content": "We've seen hostgal_photoz is the most important feature. But why?
\n\nLooking at the non-zero photoz distributions, it seems every class is evenly distributed over the photoz space with a single exception; thus one would assume there are some good FI's going on. From an astronomy point of view, how exactly is redshift/distance used to interpret photometry light curves?
", "rawMarkdown": "We've seen hostgal_photoz is the most important feature. But why?\n\nLooking at the non-zero photoz distributions, it seems every class is evenly distributed over the photoz space with a single exception; thus one would assume there are some good FI's going on. From an astronomy point of view, how exactly is redshift/distance used to interpret photometry light curves?", "votes": 9 }, { "id": 421439, "postDate": "2018-11-15T02:27:01.587Z", "content": "Objects that are farther away appear to be fainter. Redshift on its own might not be that meaningful, but if you look at how bright an object is as a function of redshift you'll see lots of information.
\n\nFor example, here is a plot of the peak brightness of a Type Ia supernova as a function of redshift:\nhttp://supernova.lbl.gov/union/figures/Union2.1_Hubble_slide.pdf\nDifferent objects will be inherently brighter or fainter, so they'll show up at different places in the brightness-redshift space.
\n\nThe classifiers are picking up on this with things like the median flux, so the combination of that with redshift gives you a lot of information.
", "rawMarkdown": "Objects that are farther away appear to be fainter. Redshift on its own might not be that meaningful, but if you look at how bright an object is as a function of redshift you'll see lots of information.\n\nFor example, here is a plot of the peak brightness of a Type Ia supernova as a function of redshift:\nhttp://supernova.lbl.gov/union/figures/Union2.1_Hubble_slide.pdf\nDifferent objects will be inherently brighter or fainter, so they'll show up at different places in the brightness-redshift space.\n\nThe classifiers are picking up on this with things like the median flux, so the combination of that with redshift gives you a lot of information.", "votes": 10, "replies": [ { "id": 422387, "postDate": "2018-11-16T06:43:06.927Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 422297, "postDate": "2018-11-16T03:09:06.037Z", "content": "Apart from Milky Way/Not Milky Way, photoz still shows up as an important feature in my Extragalactic-Only model as well. The reason for this shows up in Section 4 of the \"Starter Kit\" https://www.kaggle.com/michaelapers/the-plasticc-astronomy-starter-kit. Also a slightly \"math-ier\" explanation is here: https://physics.stackexchange.com/questions/172474/redshift-of-supernova-light-curve. Events that are moving away from us faster will appear to have a longer duration (time dilation).
", "rawMarkdown": "Apart from Milky Way/Not Milky Way, photoz still shows up as an important feature in my Extragalactic-Only model as well. The reason for this shows up in Section 4 of the \"Starter Kit\" https://www.kaggle.com/michaelapers/the-plasticc-astronomy-starter-kit. Also a slightly \"math-ier\" explanation is here: https://physics.stackexchange.com/questions/172474/redshift-of-supernova-light-curve. Events that are moving away from us faster will appear to have a longer duration (time dilation).", "votes": 6 }, { "id": 421522, "postDate": "2018-11-15T05:12:41.673Z", "content": "It distinguishes betwene galactic vs extra galactic sources, i.e. separates with 100% accuracy the classes into two groups.
\n\nEdit. I see @ONODERA gave the same answer.
", "rawMarkdown": "It distinguishes betwene galactic vs extra galactic sources, i.e. separates with 100% accuracy the classes into two groups.\n\nEdit. I see @ONODERA gave the same answer.", "votes": 2 }, { "id": 421415, "postDate": "2018-11-15T01:09:57.687Z", "content": "https://www.kaggle.com/kyleboone/naive-benchmark-galactic-vs-extragalactic
\n\nWe can tell galactic or not with it, and classes aren't overlap
", "rawMarkdown": "https://www.kaggle.com/kyleboone/naive-benchmark-galactic-vs-extragalactic\n\nWe can tell galactic or not with it, and classes aren't overlap", "votes": 2, "replies": [ { "id": 421423, "postDate": "2018-11-15T01:44:03.013Z", "content": "Right?
\n\nBut take a look at this image:
\n\n
It was generated by @Oliver's historic kernel (don't worry, I up-voted it ;-) with the only alteration being it only runs on photoz>0 extragalatic candidates. Even here, hostgal_photoz still ends up being the second most important feature, and not even far off from first. For sure there's more going on than our models interpreting galatic/extragalatic as a boolean. @jakevdp?
The image does not show...
\n\nYes, hostgal_photoz is also useful within extra galactic sources. But ONODERA's answer explains what you asked for in the first place: why is this feature the most important?
", "rawMarkdown": "The image does not show... \n\nYes, hostgal_photoz is also useful within extra galactic sources. But ONODERA's answer explains what you asked for in the first place: why is this feature the most important?", "votes": 1 } ] }, { "id": 423988, "postDate": "2018-11-19T11:41:32.590Z", "content": "For what it's worth, hostgal_photoz is my third most important feature only now, according to lgb.
", "rawMarkdown": "For what it's worth, hostgal_photoz is my third most important feature only now, according to lgb.", "replies": [ { "id": 424343, "postDate": "2018-11-20T00:21:39.977Z", "content": "‵hostgal_photoz‵ seems to have some potential to feature engineering further.
", "rawMarkdown": "‵hostgal_photoz‵ seems to have some potential to feature engineering further. ", "votes": 1 }, { "id": 424393, "postDate": "2018-11-20T03:22:03.687Z", "content": "Indeed. For instance, distmod is one of many features that can be derived from hostgal_photoz.
", "rawMarkdown": "Indeed. For instance, distmod is one of many features that can be derived from hostgal_photoz.", "votes": 1 } ] } ], "comments": [ { "id": 421439, "author_name": "Kyle Boone", "author_url": "", "post_date": "2018-11-15T02:27:01.587000", "content": "Objects that are farther away appear to be fainter. Redshift on its own might not be that meaningful, but if you look at how bright an object is as a function of redshift you'll see lots of information.
\n\nFor example, here is a plot of the peak brightness of a Type Ia supernova as a function of redshift:\nhttp://supernova.lbl.gov/union/figures/Union2.1_Hubble_slide.pdf\nDifferent objects will be inherently brighter or fainter, so they'll show up at different places in the brightness-redshift space.
\n\nThe classifiers are picking up on this with things like the median flux, so the combination of that with redshift gives you a lot of information.
", "votes": 10, "replies": [ { "id": 422387, "author_name": "", "author_url": "", "post_date": "2018-11-16T06:43:06.927000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 422297, "author_name": "Mike Holcomb", "author_url": "", "post_date": "2018-11-16T03:09:06.037000", "content": "Apart from Milky Way/Not Milky Way, photoz still shows up as an important feature in my Extragalactic-Only model as well. The reason for this shows up in Section 4 of the \"Starter Kit\" https://www.kaggle.com/michaelapers/the-plasticc-astronomy-starter-kit. Also a slightly \"math-ier\" explanation is here: https://physics.stackexchange.com/questions/172474/redshift-of-supernova-light-curve. Events that are moving away from us faster will appear to have a longer duration (time dilation).
", "votes": 6, "replies": [] }, { "id": 421522, "author_name": "CPMP", "author_url": "", "post_date": "2018-11-15T05:12:41.673000", "content": "It distinguishes betwene galactic vs extra galactic sources, i.e. separates with 100% accuracy the classes into two groups.
\n\nEdit. I see @ONODERA gave the same answer.
", "votes": 2, "replies": [] }, { "id": 421415, "author_name": "ONODERA", "author_url": "", "post_date": "2018-11-15T01:09:57.687000", "content": "https://www.kaggle.com/kyleboone/naive-benchmark-galactic-vs-extragalactic
\n\nWe can tell galactic or not with it, and classes aren't overlap
", "votes": 2, "replies": [ { "id": 421423, "author_name": "عثمان", "author_url": "", "post_date": "2018-11-15T01:44:03.013000", "content": "Right?
\n\nBut take a look at this image:
\n\n
It was generated by @Oliver's historic kernel (don't worry, I up-voted it ;-) with the only alteration being it only runs on photoz>0 extragalatic candidates. Even here, hostgal_photoz still ends up being the second most important feature, and not even far off from first. For sure there's more going on than our models interpreting galatic/extragalatic as a boolean. @jakevdp?
The image does not show...
\n\nYes, hostgal_photoz is also useful within extra galactic sources. But ONODERA's answer explains what you asked for in the first place: why is this feature the most important?
", "votes": 1, "replies": [] } ] }, { "id": 423988, "author_name": "CPMP", "author_url": "", "post_date": "2018-11-19T11:41:32.590000", "content": "For what it's worth, hostgal_photoz is my third most important feature only now, according to lgb.
", "votes": 0, "replies": [ { "id": 424343, "author_name": "Chia-Ta Tsai", "author_url": "", "post_date": "2018-11-20T00:21:39.977000", "content": "‵hostgal_photoz‵ seems to have some potential to feature engineering further.
", "votes": 1, "replies": [] }, { "id": 424393, "author_name": "CPMP", "author_url": "", "post_date": "2018-11-20T03:22:03.687000", "content": "Indeed. For instance, distmod is one of many features that can be derived from hostgal_photoz.
", "votes": 1, "replies": [] } ] } ], "raw_markdown_by_id": { "421305": "We've seen hostgal_photoz is the most important feature. But why?\n\nLooking at the non-zero photoz distributions, it seems every class is evenly distributed over the photoz space with a single exception; thus one would assume there are some good FI's going on. From an astronomy point of view, how exactly is redshift/distance used to interpret photometry light curves?", "421439": "Objects that are farther away appear to be fainter. Redshift on its own might not be that meaningful, but if you look at how bright an object is as a function of redshift you'll see lots of information.\n\nFor example, here is a plot of the peak brightness of a Type Ia supernova as a function of redshift:\nhttp://supernova.lbl.gov/union/figures/Union2.1_Hubble_slide.pdf\nDifferent objects will be inherently brighter or fainter, so they'll show up at different places in the brightness-redshift space.\n\nThe classifiers are picking up on this with things like the median flux, so the combination of that with redshift gives you a lot of information.", "422297": "Apart from Milky Way/Not Milky Way, photoz still shows up as an important feature in my Extragalactic-Only model as well. The reason for this shows up in Section 4 of the \"Starter Kit\" https://www.kaggle.com/michaelapers/the-plasticc-astronomy-starter-kit. Also a slightly \"math-ier\" explanation is here: https://physics.stackexchange.com/questions/172474/redshift-of-supernova-light-curve. Events that are moving away from us faster will appear to have a longer duration (time dilation).", "421522": "It distinguishes betwene galactic vs extra galactic sources, i.e. separates with 100% accuracy the classes into two groups.\n\nEdit. I see @ONODERA gave the same answer.", "421415": "https://www.kaggle.com/kyleboone/naive-benchmark-galactic-vs-extragalactic\n\nWe can tell galactic or not with it, and classes aren't overlap", "423988": "For what it's worth, hostgal_photoz is my third most important feature only now, according to lgb." } }