![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3355479%2F8f3410920f795cdae36668b59247e3cc%2Fsiim-isici.png?generation=1597734395054691&alt=media\")
In my final submission I was simply combining prediction by weights.
\nI choose my 5 best models and combined their predition by weight got a public LB of 0.9646.
\nI computed the 5 weights by simple dirichlet distribution, choosing randomly.
Few days prior I have used best three;
\nChanging weights some up some down.
\nChoose only 3 and was able to get at last public LB 0f 0.9666.
In my final submission I was simply combining prediction by weights.
\nI choose my 5 best models and combined their predition by weight got a public LB of 0.9646.
\nI computed the 5 weights by simple dirichlet distribution, choosing randomly.
Few days prior I have used best three;
\nChanging weights some up some down.
\nChoose only 3 and was able to get at last public LB 0f 0.9666.
Despite the fact that I fell from the top 7% to the top 12% and lost my medal, my best score was shown by a sub selected via CV. At the same time, public blending submissions which gave me 0.9660+ scores have shown bad results at private LB, usually approximately 0.92. I know that a lot of people got a good score with public ensembles. However, CV is above all. Of course, I'm disappointed about losing my medal but I'm not disappointed about choosing the wrong submission. And many people regret it.
", "rawMarkdown": "Despite the fact that I fell from the top 7% to the top 12% and lost my medal, my best score was shown by a sub selected via CV. At the same time, public blending submissions which gave me 0.9660+ scores have shown bad results at private LB, usually approximately 0.92. I know that a lot of people got a good score with public ensembles. However, CV is above all. Of course, I'm disappointed about losing my medal but I'm not disappointed about choosing the wrong submission. And many people regret it.", "votes": null } ], "comments": [ { "id": 975262, "author_name": "vadimtimakin", "author_url": "", "post_date": "08/18/2020 08:10:36", "content": "Despite the fact that I fell from the top 7% to the top 12% and lost my medal, my best score was shown by a sub selected via CV. At the same time, public blending submissions which gave me 0.9660+ scores have shown bad results at private LB, usually approximately 0.92. I know that a lot of people got a good score with public ensembles. However, CV is above all. Of course, I'm disappointed about losing my medal but I'm not disappointed about choosing the wrong submission. And many people regret it.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "975135": "In my final submission I was simply combining prediction by weights.\nI choose my 5 best models and combined their predition by weight got a public LB of 0.9646.\nI computed the 5 weights by simple dirichlet distribution, choosing randomly.\n\nFew days prior I have used best three; \nChanging weights some up some down.\nChoose only 3 and was able to get at last public **LB** 0f **0.9666.**\n\n* But it turns out submission which was combined of 5 model probabilities has best private **LB** of **0.9430.**\n* Cost me a silver model.\nSaw the results in the morning was really angry at myself.\n* Well learning new everyday, being a newbie has it downsides.\n\n", "975262": "Despite the fact that I fell from the top 7% to the top 12% and lost my medal, my best score was shown by a sub selected via CV. At the same time, public blending submissions which gave me 0.9660+ scores have shown bad results at private LB, usually approximately 0.92. I know that a lot of people got a good score with public ensembles. However, CV is above all. Of course, I'm disappointed about losing my medal but I'm not disappointed about choosing the wrong submission. And many people regret it." }, "source": "meta" }