![\"Comparison](\"https://i.postimg.cc/vHC1TvQf/Teams-Road.png\")
Hey fellows, first of all congrats for this great competition you have fight. It has been quite interesting for watching and learning from your different approaches.
\n\nI wanted to see how different has been the road for the top teams, so I created a dendogram to visualize their road, by creating a table with their score by day. The color is the score, the rows are the days and the columns are the teams.
\n\nInterestingly, several of the top 10 teams are part of the same cluster (cluster 1). They entered the competition around June (they have worked hard!), they had a boost around July 10th and then another boost in the middle of August. On the other hand, Hybrid team has made a difference during August reaching the 2nd place. The current first team, Quantum Uncertainty, left others behind from the middle of June.
\n\n
Looking forward to watch the final! \nGood work and good luck ;)
", "messages": [ { "id": "609199", "postDate": "08/27/2019 13:46:46", "content": "Hey fellows, first of all congrats for this great competition you have fight. It has been quite interesting for watching and learning from your different approaches.
\n\nI wanted to see how different has been the road for the top teams, so I created a dendogram to visualize their road, by creating a table with their score by day. The color is the score, the rows are the days and the columns are the teams.
\n\nInterestingly, several of the top 10 teams are part of the same cluster (cluster 1). They entered the competition around June (they have worked hard!), they had a boost around July 10th and then another boost in the middle of August. On the other hand, Hybrid team has made a difference during August reaching the 2nd place. The current first team, Quantum Uncertainty, left others behind from the middle of June.
\n\n
Looking forward to watch the final! \nGood work and good luck ;)
", "rawMarkdown": "Hey fellows, first of all congrats for this great competition you have fight. It has been quite interesting for watching and learning from your different approaches. \n\nI wanted to see how different has been the road for the top teams, so I created a dendogram to visualize their road, by creating a table with their score by day. The color is the score, the rows are the days and the columns are the teams. \n\nInterestingly, several of the top 10 teams are part of the same cluster (cluster 1). They entered the competition around June (they have worked hard!), they had a boost around July 10th and then another boost in the middle of August. On the other hand, Hybrid team has made a difference during August reaching the 2nd place. The current first team, Quantum Uncertainty, left others behind from the middle of June.\n\n\n\nLooking forward to watch the final! \nGood work and good luck ;)", "votes": null }, { "id": "609226", "postDate": "08/27/2019 14:13:13", "content": "Very cool! I really like the idea of using a heatmap to track team's progress. What package did you use to create this?
\n\nI'd like to see a version where teams are sorted in their current LB position. And maybe have the color scaling start at 0 instead of +3 - or use a different color palette) so you can see more variance in top scores.
\n\nI've been updating the LB tracking notebook here. https://www.kaggle.com/robikscube/the-race-to-predict-molecular-properties and will continue to do so until it ends tomorrow.
\n\nI'm excited to read some of the winning solutions write-ups to see how they did it!
", "rawMarkdown": "Very cool! I really like the idea of using a heatmap to track team's progress. What package did you use to create this?\n\nI'd like to see a version where teams are sorted in their current LB position. And maybe have the color scaling start at 0 instead of +3 - or use a different color palette) so you can see more variance in top scores.\n\nI've been updating the LB tracking notebook here. https://www.kaggle.com/robikscube/the-race-to-predict-molecular-properties and will continue to do so until it ends tomorrow.\n\nI'm excited to read some of the winning solutions write-ups to see how they did it!", "votes": null }, { "id": "609286", "postDate": "08/27/2019 15:02:52", "content": "Hey @robikscube I used the seaborn clustermap. Pretty handy.. you can cluster rows, columns, using different linkage methods, different transformations, etc. I didn't order them by ranking because I wanted to visualize the clusters according to teams with similar road patterns (so I clustered the columns, but not the rows (date)).\nI modified the scale and include the line of code to generate it ;)
\n\n![\"Teams](\"https://i.postimg.cc/Ls8qCkMb/Teams-Road2.png\")
Great Work.....!! @carlospk
", "rawMarkdown": "Great Work.....!! @carlospk", "votes": null } ], "comments": [ { "id": 609226, "author_name": "robikscube", "author_url": "", "post_date": "08/27/2019 14:13:13", "content": "Very cool! I really like the idea of using a heatmap to track team's progress. What package did you use to create this?
\n\nI'd like to see a version where teams are sorted in their current LB position. And maybe have the color scaling start at 0 instead of +3 - or use a different color palette) so you can see more variance in top scores.
\n\nI've been updating the LB tracking notebook here. https://www.kaggle.com/robikscube/the-race-to-predict-molecular-properties and will continue to do so until it ends tomorrow.
\n\nI'm excited to read some of the winning solutions write-ups to see how they did it!
", "votes": null, "replies": [ { "id": 609286, "author_name": "carlospk", "author_url": "", "post_date": "08/27/2019 15:02:52", "content": "Hey @robikscube I used the seaborn clustermap. Pretty handy.. you can cluster rows, columns, using different linkage methods, different transformations, etc. I didn't order them by ranking because I wanted to visualize the clusters according to teams with similar road patterns (so I clustered the columns, but not the rows (date)).\nI modified the scale and include the line of code to generate it ;)
\n\n
Great Work.....!! @carlospk
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "609199": "Hey fellows, first of all congrats for this great competition you have fight. It has been quite interesting for watching and learning from your different approaches. \n\nI wanted to see how different has been the road for the top teams, so I created a dendogram to visualize their road, by creating a table with their score by day. The color is the score, the rows are the days and the columns are the teams. \n\nInterestingly, several of the top 10 teams are part of the same cluster (cluster 1). They entered the competition around June (they have worked hard!), they had a boost around July 10th and then another boost in the middle of August. On the other hand, Hybrid team has made a difference during August reaching the 2nd place. The current first team, Quantum Uncertainty, left others behind from the middle of June.\n\n\n\nLooking forward to watch the final! \nGood work and good luck ;)", "609226": "Very cool! I really like the idea of using a heatmap to track team's progress. What package did you use to create this?\n\nI'd like to see a version where teams are sorted in their current LB position. And maybe have the color scaling start at 0 instead of +3 - or use a different color palette) so you can see more variance in top scores.\n\nI've been updating the LB tracking notebook here. https://www.kaggle.com/robikscube/the-race-to-predict-molecular-properties and will continue to do so until it ends tomorrow.\n\nI'm excited to read some of the winning solutions write-ups to see how they did it!", "609286": "Hey @robikscube I used the seaborn clustermap. Pretty handy.. you can cluster rows, columns, using different linkage methods, different transformations, etc. I didn't order them by ranking because I wanted to visualize the clusters according to teams with similar road patterns (so I clustered the columns, but not the rows (date)).\nI modified the scale and include the line of code to generate it ;)\n\n", "609407": "Great Work.....!! @carlospk" }, "source": "meta" }