![\"image\"](\"http://playagricola.com/Kaggle/ani2.gif\")
\n
![\"image\"](\"http://playagricola.com/Kaggle/ani1.gif\")
\nFrom the histogram below, we see that sc_constant = 30 is small and sc_constant = 70 is large.\n![\"image\"](\"http://playagricola.com/Kaggle/p3382019.png\")
If you're brainstorming for new model features, I posted some molecule animations that may give you some new ideas here.
\n\nFor example from observing animations, we observe that if nitrogen from 1JHN coupling is connected to more nitrogen than the sc_constant is large. And if nitrogen is connected to only 1 more atom (in addition to the original hydrogen) then sc_constant is small. There are many more observable patterns displayed in my kernel/notebook. Enjoy!
These animations are made with matplotlib. Code is here
\n\nFrom the histogram below, we see that sc_constant = 30 is small and sc_constant = 70 is large.\n
If you're brainstorming for new model features, I posted some molecule animations that may give you some new ideas here.
\n\nFor example from observing animations, we observe that if nitrogen from 1JHN coupling is connected to more nitrogen than the sc_constant is large. And if nitrogen is connected to only 1 more atom (in addition to the original hydrogen) then sc_constant is small. There are many more observable patterns displayed in my kernel/notebook. Enjoy!
These animations are made with matplotlib. Code is here
\n\nFrom the histogram below, we see that sc_constant = 30 is small and sc_constant = 70 is large.\n
Seriously, how can you guys not comment on this post? 😂 This is totally new info to me.
", "rawMarkdown": "Seriously, how can you guys not comment on this post? 😂 This is totally new info to me.", "votes": null }, { "id": "604085", "postDate": "08/21/2019 02:37:20", "content": "Thanks Jiwei. I'm late to the party but I started doing some EDA this week. Most people know that 1JHN coupling is correlated with the bond distance between hydrogen and nitrogen. Decreased bond distance is associated with increased coupling constant as seen below.\n![\"image\"](\"http://playagricola.com/Kaggle/p0082019.png\")
\nWhat's interesting is that you can also look at the bond distances to the other atoms that nitrogen is connected to. Below is the next closest atom to nitrogen (besides the original hydrogen). It's type and distance predicts coupling constant.\n![\"image\"](\"http://playagricola.com/Kaggle/p1182019.png\")
\nFurthermore, you can look at the third atom attached to nitrogen\n![\"image\"](\"http://playagricola.com/Kaggle/p2282019.png\")
\nAll of these patterns are observable by humans if you look at lots of molecule animations. I posted more animations and EDA in my kernel/notebook here. After you look at enough rotation molecules you can predict the coupling constant yourself (using your eyes to within plus minus 10) when presented with an unknown HN bond.
@jiweiliu I have spend like a week looking at molecules and I think that given two atoms in a molecule I could estimate scalar coupling constant pretty well by myself (neural net trained :D).\nEDA is like the first thing to do and distances and bonds between atoms are very basic features almost everybody is using here, so that may be the reason why nobody comments on this.\nNevertheless nice EDA :)
", "rawMarkdown": "jiweiliu I have spend like a week looking at molecules and I think that given two atoms in a molecule I could estimate scalar coupling constant pretty well by myself (neural net trained :D).\nEDA is like the first thing to do and distances and bonds between atoms are very basic features almost everybody is using here, so that may be the reason why nobody comments on this.\nNevertheless nice EDA :)", "votes": null }, { "id": "604155", "postDate": "08/21/2019 05:13:39", "content": "I studied your work and I aggree with @jiweiliu @cdeotte . How can you guys not comment on this discussion :) Thanks for sharing.
", "rawMarkdown": "I studied your work and I aggree with @jiweiliu @cdeotte . How can you guys not comment on this discussion :) Thanks for sharing.", "votes": null }, { "id": "604213", "postDate": "08/21/2019 06:43:05", "content": "@cdeotte Right, neighbors of neighbors matter. This is why graph NN are so widely used here.
", "rawMarkdown": "cdeotte Right, neighbors of neighbors matter. This is why graph NN are so widely used here.", "votes": null }, { "id": "604242", "postDate": "08/21/2019 07:20:46", "content": "Awesome.... Impressive Visuals.... Thanks @cdeotte
", "rawMarkdown": "Awesome.... Impressive Visuals.... Thanks @cdeotte", "votes": null }, { "id": "604527", "postDate": "08/21/2019 13:33:14", "content": "Nice and helps lot..\nthanks for sharing @cdeotte
", "rawMarkdown": "Nice and helps lot..\nthanks for sharing @cdeotte", "votes": null }, { "id": "604554", "postDate": "08/21/2019 14:08:43", "content": "\n\n\nSeriously, how can you guys not comment on this post?
\n
Probably because many people have already shared molecule visualization code in notebooks or in the forum, for instance this awesome notebook published early in the competition: https://www.kaggle.com/borisdee/how-to-easy-visualization-of-molecules
\n\n\n\n\nThis is totally new info to me.
\n
You should read all the forum and browse all public kernels IMHO ;)
", "rawMarkdown": "> Seriously, how can you guys not comment on this post? \n\nProbably because many people have already shared molecule visualization code in notebooks or in the forum, for instance this awesome notebook published early in the competition: https://www.kaggle.com/borisdee/how-to-easy-visualization-of-molecules\n\n> This is totally new info to me.\n\nYou should read all the forum and browse all public kernels IMHO ;)", "votes": null }, { "id": "604631", "postDate": "08/21/2019 15:32:56", "content": "Here's a fun molecule. It's dsgdb9nsd_024553. It is a unique closed loop of 9 atoms. Since nitrogen is connected to 2 carbons (in addition to original hydrogen), it has a middle strength sc_constant.\n![\"image\"](\"http://playagricola.com/Kaggle/ani9.gif\")
Seriously, how can you guys not comment on this post? 😂 This is totally new info to me.
", "votes": null, "replies": [ { "id": 604085, "author_name": "cdeotte", "author_url": "", "post_date": "08/21/2019 02:37:20", "content": "Thanks Jiwei. I'm late to the party but I started doing some EDA this week. Most people know that 1JHN coupling is correlated with the bond distance between hydrogen and nitrogen. Decreased bond distance is associated with increased coupling constant as seen below.\n\nWhat's interesting is that you can also look at the bond distances to the other atoms that nitrogen is connected to. Below is the next closest atom to nitrogen (besides the original hydrogen). It's type and distance predicts coupling constant.\n\nFurthermore, you can look at the third atom attached to nitrogen\n\nAll of these patterns are observable by humans if you look at lots of molecule animations. I posted more animations and EDA in my kernel/notebook here. After you look at enough rotation molecules you can predict the coupling constant yourself (using your eyes to within plus minus 10) when presented with an unknown HN bond.
@jiweiliu I have spend like a week looking at molecules and I think that given two atoms in a molecule I could estimate scalar coupling constant pretty well by myself (neural net trained :D).\nEDA is like the first thing to do and distances and bonds between atoms are very basic features almost everybody is using here, so that may be the reason why nobody comments on this.\nNevertheless nice EDA :)
", "votes": null, "replies": [] }, { "id": 604213, "author_name": "cpmpml", "author_url": "", "post_date": "08/21/2019 06:43:05", "content": "@cdeotte Right, neighbors of neighbors matter. This is why graph NN are so widely used here.
", "votes": null, "replies": [] }, { "id": 604554, "author_name": "cpmpml", "author_url": "", "post_date": "08/21/2019 14:08:43", "content": "\n\n\nSeriously, how can you guys not comment on this post?
\n
Probably because many people have already shared molecule visualization code in notebooks or in the forum, for instance this awesome notebook published early in the competition: https://www.kaggle.com/borisdee/how-to-easy-visualization-of-molecules
\n\n\n\n\nThis is totally new info to me.
\n
You should read all the forum and browse all public kernels IMHO ;)
", "votes": null, "replies": [] } ] }, { "id": 604155, "author_name": "utkukubilay", "author_url": "", "post_date": "08/21/2019 05:13:39", "content": "I studied your work and I aggree with @jiweiliu @cdeotte . How can you guys not comment on this discussion :) Thanks for sharing.
", "votes": null, "replies": [] }, { "id": 604242, "author_name": "veeralakrishna", "author_url": "", "post_date": "08/21/2019 07:20:46", "content": "Awesome.... Impressive Visuals.... Thanks @cdeotte
", "votes": null, "replies": [] }, { "id": 604527, "author_name": "niteshfre", "author_url": "", "post_date": "08/21/2019 13:33:14", "content": "Nice and helps lot..\nthanks for sharing @cdeotte
", "votes": null, "replies": [] }, { "id": 604631, "author_name": "cdeotte", "author_url": "", "post_date": "08/21/2019 15:32:56", "content": "Here's a fun molecule. It's dsgdb9nsd_024553. It is a unique closed loop of 9 atoms. Since nitrogen is connected to 2 carbons (in addition to original hydrogen), it has a middle strength sc_constant.\n