![\"tsne\"](\"https://storage.googleapis.com/kaggle-forum-message-attachments/506285/12842/tsne%20lanl.jpg\")
Check out my TSNE representation of multiple statistical features on train. The colorbar on the right corresponds to the target variable (time to failure)\n
Check out my TSNE representation of multiple statistical features on train. The colorbar on the right corresponds to the target variable (time to failure)\n
It might also be a question mark! Have you specifically looked for data points clustering around the point (0,-200)?
", "rawMarkdown": "It might also be a question mark! Have you specifically looked for data points clustering around the point (0,-200)?", "votes": null }, { "id": "506294", "postDate": "04/03/2019 09:25:17", "content": "nice kernel,thanks for sharing
", "rawMarkdown": "nice kernel,thanks for sharing", "votes": null }, { "id": "506880", "postDate": "04/04/2019 01:40:05", "content": "A significant source of error in this competition is 'mini-quakes' that seem to reset the physical state of the system without triggering TTF==0. These can be seen clearly in some of the longer earthquake periods in train.csv and I've yet to make a model that can distinguish them from actual quakes, leading to a 'double dip' in the predictions. I would hazard that the central overlap region around (0, 0) is a consequence of this. Nice work!
", "rawMarkdown": "A significant source of error in this competition is 'mini-quakes' that seem to reset the physical state of the system without triggering TTF==0. These can be seen clearly in some of the longer earthquake periods in train.csv and I've yet to make a model that can distinguish them from actual quakes, leading to a 'double dip' in the predictions. I would hazard that the central overlap region around (0, 0) is a consequence of this. Nice work!", "votes": null }, { "id": "507049", "postDate": "04/04/2019 08:01:06", "content": "Nice image. Can you give a legend for this image and expain the image a little more (the title is a bit strange to me...why 42?). Where can we find the code for this image ?
", "rawMarkdown": "Nice image. Can you give a legend for this image and expain the image a little more (the title is a bit strange to me...why 42?). Where can we find the code for this image ?", "votes": null }, { "id": "507144", "postDate": "04/04/2019 10:52:10", "content": "42 is just the answer to life, the universe and everything, google it :)
\n\nHow I made this picture\n1. Cut the train data into segments with length equals test sample length\n2. For each train and test segment calculated simple statistics like min, max, mean, etc.\n3. Fit TSNE with two components on union of train and test stats from 2.\n4. Using fitted TSNE transformed train stats from 2.\n5. Draw scatterplot of step 4 output with colorbar corresponded to target variable (from step 1)
", "rawMarkdown": "42 is just the answer to life, the universe and everything, google it :)\n\nHow I made this picture\n1. Cut the train data into segments with length equals test sample length\n2. For each train and test segment calculated simple statistics like min, max, mean, etc.\n3. Fit TSNE with two components on union of train and test stats from 2.\n4. Using fitted TSNE transformed train stats from 2.\n5. Draw scatterplot of step 4 output with colorbar corresponded to target variable (from step 1)", "votes": null }, { "id": "531547", "postDate": "05/15/2019 05:45:10", "content": "![\"\"](\"https://upload.wikimedia.org/wikipedia/commons/thumb/5/52/Hubble2005-01-barred-spiral-galaxy-NGC1300.jpg/1920px-Hubble2005-01-barred-spiral-galaxy-NGC1300.jpg\")
It might also be a question mark! Have you specifically looked for data points clustering around the point (0,-200)?
", "votes": null, "replies": [] }, { "id": 506294, "author_name": "econdata", "author_url": "", "post_date": "04/03/2019 09:25:17", "content": "nice kernel,thanks for sharing
", "votes": null, "replies": [] }, { "id": 506880, "author_name": "bigironsphere", "author_url": "", "post_date": "04/04/2019 01:40:05", "content": "A significant source of error in this competition is 'mini-quakes' that seem to reset the physical state of the system without triggering TTF==0. These can be seen clearly in some of the longer earthquake periods in train.csv and I've yet to make a model that can distinguish them from actual quakes, leading to a 'double dip' in the predictions. I would hazard that the central overlap region around (0, 0) is a consequence of this. Nice work!
", "votes": null, "replies": [] }, { "id": 507049, "author_name": "hmcranbercourt", "author_url": "", "post_date": "04/04/2019 08:01:06", "content": "Nice image. Can you give a legend for this image and expain the image a little more (the title is a bit strange to me...why 42?). Where can we find the code for this image ?
", "votes": null, "replies": [ { "id": 507144, "author_name": "uselessskills", "author_url": "", "post_date": "04/04/2019 10:52:10", "content": "42 is just the answer to life, the universe and everything, google it :)
\n\nHow I made this picture\n1. Cut the train data into segments with length equals test sample length\n2. For each train and test segment calculated simple statistics like min, max, mean, etc.\n3. Fit TSNE with two components on union of train and test stats from 2.\n4. Using fitted TSNE transformed train stats from 2.\n5. Draw scatterplot of step 4 output with colorbar corresponded to target variable (from step 1)
", "votes": null, "replies": [] } ] }, { "id": 531547, "author_name": "slivka83", "author_url": "", "post_date": "05/15/2019 05:45:10", "content": "