Title
\ndata
\nSNR
\nThis is all I can tell the difference. But still, it's far not clear enough. I really expect to hear the thoughts of everyone especially from the organizers @rodrigotenorio. Thanks in advance!
", "messages": [ { "id": 2031690, "postDate": "2022-11-16T08:03:41.443Z", "content": "Title
\ndata
\nSNR
\nThis is all I can tell the difference. But still, it's far not clear enough. I really expect to hear the thoughts of everyone especially from the organizers @rodrigotenorio. Thanks in advance!
", "rawMarkdown": "**Title**\n- The title last year is **G2Net Gravitational Wave Detection**, and the title this year is **G2Net Detecting Continuous Gravitational Waves**\n\n**data**\n- The raw data last year are 1d waves, whose x-axis is time, and y-axis is the amplitude of each timestamp. With CQT/CWT, the 1d wave can be converted into a 2d image(spectrum), which x-axis is time, y-axis is frequency and the intensity per unit pixel is the amplitude of a specific frequency within a specific timestamp.\n- The raw data are 2d images(spectrum), which x-axis is time, y-axis is frequency and the intensity per unit pixel is the amplitude of a specific frequency within a specific timestamp. (Which is the same as the data after CQT last year in format? I am not sure...)\n\n**SNR**\n- The SNR in the competition this year is much lower, which means that the signal this year is much weaker, resulting in the auc being much lower.\n\nThis is all I can tell the difference. But still, it's far not clear enough. I really expect to hear the thoughts of everyone especially from the organizers @rodrigotenorio. Thanks in advance! ", "votes": 9 }, { "id": 2033833, "postDate": "2022-11-17T15:24:54.960Z", "content": "Physics-wise, last year competition dealt with the waves from compact binary objects, like close orbiting black holes or white dwarfs. This year we face fast spinning asymmetric neutron stars. Last year set-up provides much more explosive waveforms, and much stronger.
\nContinuous GW we are hunting here were never detected so far, so we have a chance to help a real scientific discovery.
Amazing. That's exactly the difference I was expecting to know. Your answer was very enlightening to me. @alexz0
", "rawMarkdown": "Amazing. That's exactly the difference I was expecting to know. Your answer was very enlightening to me. @alexz0 " } ] }, { "id": 2032129, "postDate": "2022-11-16T12:52:46.043Z", "content": "My personal opinion is organizers are expecting something else that only a Spetrogram Image Classifier, when signal is very inside into the noise. I guess we must combine SIC with another data model to achieve 90% or upper results. What model is that? What data we must generate or compute for it? Good questions. I think the true key is there.
\nCheers :)
", "rawMarkdown": "My personal opinion is organizers are expecting something else that only a Spetrogram Image Classifier, when signal is very inside into the noise. I guess we must combine SIC with another data model to achieve 90% or upper results. What model is that? What data we must generate or compute for it? Good questions. I think the true key is there.\n\nCheers :)", "votes": 2, "replies": [ { "id": 2032178, "postDate": "2022-11-16T13:15:46.687Z", "content": "I agree. I just ran an experiment. I generated the same amount of signal and noise, where the D of signals took values from 50 to 100. I found that the trained model barely converged. After running the test set, the probability for all samples was almost 0.5. This shows that only relying on the spectrum for image classification is not enough.
", "rawMarkdown": "I agree. I just ran an experiment. I generated the same amount of signal and noise, where the D of signals took values from 50 to 100. I found that the trained model barely converged. After running the test set, the probability for all samples was almost 0.5. This shows that only relying on the spectrum for image classification is not enough.", "votes": 1 }, { "id": 2032269, "postDate": "2022-11-16T14:19:20.150Z", "content": "Exactly. There is so much collinearity among the samples, so that the Student's Test is never accomplished in data cause there is homogeneity in the variances (noises). In other words, the p-value is pretty high, being p-value >> alpha.
", "rawMarkdown": "Exactly. There is so much collinearity among the samples, so that the Student's Test is never accomplished in data cause there is homogeneity in the variances (noises). In other words, the p-value is pretty high, being p-value >> alpha." }, { "id": 2033138, "postDate": "2022-11-17T03:58:19.703Z", "content": "It's impossible to reach 0.9 auc. The champion of G2Net last year was 0.88, and the data last year is much much more easier than this year.
", "rawMarkdown": "It's impossible to reach 0.9 auc. The champion of G2Net last year was 0.88, and the data last year is much much more easier than this year." }, { "id": 2033371, "postDate": "2022-11-17T08:01:04.457Z", "content": "Sorry, I didn't explain myself well. When I said \"0.9% or upper results\", I wanted to say upper results to actually SIC high results. No matter if its from 0.9 or 0.85 or 0.8.
", "rawMarkdown": "Sorry, I didn't explain myself well. When I said \"0.9% or upper results\", I wanted to say upper results to actually SIC high results. No matter if its from 0.9 or 0.85 or 0.8." }, { "id": 2035893, "postDate": "2022-11-19T09:28:36.130Z", "content": "@zjwxdu Is your test set the generated data, I did pretty much the same job as you but got a 0.9+ cv
", "rawMarkdown": "@zjwxdu Is your test set the generated data, I did pretty much the same job as you but got a 0.9+ cv" }, { "id": 2035916, "postDate": "2022-11-19T09:57:30.497Z", "content": "It is strange… On one occasion, I generate some noise examples in band=0.5, And the frequency shape is 900. So I do a fixed slice [0:360, :], I also found the cv is extremely high. I take the random slice [randint: randint + 360], the cv is back to normal status…
\nAnd I also found that the distribution of my predicted result (in submit test set) has hardly negative judgment. It maybe the signal with high depth factor is easy to confuse with noise.
Physics-wise, last year competition dealt with the waves from compact binary objects, like close orbiting black holes or white dwarfs. This year we face fast spinning asymmetric neutron stars. Last year set-up provides much more explosive waveforms, and much stronger.
\nContinuous GW we are hunting here were never detected so far, so we have a chance to help a real scientific discovery.
Amazing. That's exactly the difference I was expecting to know. Your answer was very enlightening to me. @alexz0
", "votes": 0, "replies": [] } ] }, { "id": 2032129, "author_name": "Zollkron", "author_url": "", "post_date": "2022-11-16T12:52:46.043000", "content": "My personal opinion is organizers are expecting something else that only a Spetrogram Image Classifier, when signal is very inside into the noise. I guess we must combine SIC with another data model to achieve 90% or upper results. What model is that? What data we must generate or compute for it? Good questions. I think the true key is there.
\nCheers :)
", "votes": 2, "replies": [ { "id": 2032178, "author_name": "Jiawei Zhang", "author_url": "", "post_date": "2022-11-16T13:15:46.687000", "content": "I agree. I just ran an experiment. I generated the same amount of signal and noise, where the D of signals took values from 50 to 100. I found that the trained model barely converged. After running the test set, the probability for all samples was almost 0.5. This shows that only relying on the spectrum for image classification is not enough.
", "votes": 1, "replies": [] }, { "id": 2032269, "author_name": "Zollkron", "author_url": "", "post_date": "2022-11-16T14:19:20.150000", "content": "Exactly. There is so much collinearity among the samples, so that the Student's Test is never accomplished in data cause there is homogeneity in the variances (noises). In other words, the p-value is pretty high, being p-value >> alpha.
", "votes": 0, "replies": [] }, { "id": 2033138, "author_name": "ForcewithMe", "author_url": "", "post_date": "2022-11-17T03:58:19.703000", "content": "It's impossible to reach 0.9 auc. The champion of G2Net last year was 0.88, and the data last year is much much more easier than this year.
", "votes": 0, "replies": [] }, { "id": 2033371, "author_name": "Zollkron", "author_url": "", "post_date": "2022-11-17T08:01:04.457000", "content": "Sorry, I didn't explain myself well. When I said \"0.9% or upper results\", I wanted to say upper results to actually SIC high results. No matter if its from 0.9 or 0.85 or 0.8.
", "votes": 0, "replies": [] }, { "id": 2035893, "author_name": "zh", "author_url": "", "post_date": "2022-11-19T09:28:36.130000", "content": "@zjwxdu Is your test set the generated data, I did pretty much the same job as you but got a 0.9+ cv
", "votes": 0, "replies": [] }, { "id": 2035916, "author_name": "Jiawei Zhang", "author_url": "", "post_date": "2022-11-19T09:57:30.497000", "content": "It is strange… On one occasion, I generate some noise examples in band=0.5, And the frequency shape is 900. So I do a fixed slice [0:360, :], I also found the cv is extremely high. I take the random slice [randint: randint + 360], the cv is back to normal status…
\nAnd I also found that the distribution of my predicted result (in submit test set) has hardly negative judgment. It maybe the signal with high depth factor is easy to confuse with noise.