{
  "id": 294790,
  "title": "Different similarity standards",
  "url": "/competitions/cs5785-fall-2021-final/discussion/294790",
  "author_name": "",
  "post_date": "2021-12-12T19:57:05.133509100Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p><a href=\"https://docs.scipy.org/doc/scipy/reference/generated/scipy.spatial.distance.cdist.html\" target=\"_blank\">https://docs.scipy.org/doc/scipy/reference/generated/scipy.spatial.distance.cdist.html</a></p>\n<p>We have tried several different similarity standards to test the difference, cosine similarity is not the only one that works well for the image search engine. Trying different standards might have some contribution to the score (even not really a significant improvement, but it can help to go over some thresholds, such as 29.xx to 30). Hope that helps a little bit.</p>",
  "messages": [
    {
      "id": "1615860",
      "postDate": "12/12/2021 19:57:05",
      "content": "<p><a href=\"https://docs.scipy.org/doc/scipy/reference/generated/scipy.spatial.distance.cdist.html\" target=\"_blank\">https://docs.scipy.org/doc/scipy/reference/generated/scipy.spatial.distance.cdist.html</a></p>\n<p>We have tried several different similarity standards to test the difference, cosine similarity is not the only one that works well for the image search engine. Trying different standards might have some contribution to the score (even not really a significant improvement, but it can help to go over some thresholds, such as 29.xx to 30). Hope that helps a little bit.</p>",
      "rawMarkdown": "https://docs.scipy.org/doc/scipy/reference/generated/scipy.spatial.distance.cdist.html\n\nWe have tried several different similarity standards to test the difference, cosine similarity is not the only one that works well for the image search engine. Trying different standards might have some contribution to the score (even not really a significant improvement, but it can help to go over some thresholds, such as 29.xx to 30). Hope that helps a little bit.",
      "votes": null
    },
    {
      "id": "1616933",
      "postDate": "12/13/2021 18:04:58",
      "content": "<p>Thanks for sharing. Our team also tried different methods to calculate distance. We tried Manhattan, Chebyshev  and etc. Overall, cosine is the one with the best performance.   <a href=\"https://tuhinmukherjee74.medium.com/different-types-of-distances-used-in-machine-learning-8f541bbdce8\" target=\"_blank\">https://tuhinmukherjee74.medium.com/different-types-of-distances-used-in-machine-learning-8f541bbdce8</a></p>",
      "rawMarkdown": "Thanks for sharing. Our team also tried different methods to calculate distance. We tried Manhattan, Chebyshev  and etc. Overall, cosine is the one with the best performance.   https://tuhinmukherjee74.medium.com/different-types-of-distances-used-in-machine-learning-8f541bbdce8",
      "votes": null
    },
    {
      "id": "1617194",
      "postDate": "12/13/2021 23:52:33",
      "content": "<p>Same here. Actually for my models, the testing score of cosine is about 0.1 higher than other distance calculation methods.</p>",
      "rawMarkdown": "Same here. Actually for my models, the testing score of cosine is about 0.1 higher than other distance calculation methods.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1616933,
      "author_name": "aqqqqq",
      "author_url": "",
      "post_date": "12/13/2021 18:04:58",
      "content": "<p>Thanks for sharing. Our team also tried different methods to calculate distance. We tried Manhattan, Chebyshev  and etc. Overall, cosine is the one with the best performance.   <a href=\"https://tuhinmukherjee74.medium.com/different-types-of-distances-used-in-machine-learning-8f541bbdce8\" target=\"_blank\">https://tuhinmukherjee74.medium.com/different-types-of-distances-used-in-machine-learning-8f541bbdce8</a></p>",
      "votes": null,
      "replies": [
        {
          "id": 1617194,
          "author_name": "ml2225",
          "author_url": "",
          "post_date": "12/13/2021 23:52:33",
          "content": "<p>Same here. Actually for my models, the testing score of cosine is about 0.1 higher than other distance calculation methods.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1615860": "https://docs.scipy.org/doc/scipy/reference/generated/scipy.spatial.distance.cdist.html\n\nWe have tried several different similarity standards to test the difference, cosine similarity is not the only one that works well for the image search engine. Trying different standards might have some contribution to the score (even not really a significant improvement, but it can help to go over some thresholds, such as 29.xx to 30). Hope that helps a little bit.",
    "1616933": "Thanks for sharing. Our team also tried different methods to calculate distance. We tried Manhattan, Chebyshev  and etc. Overall, cosine is the one with the best performance.   https://tuhinmukherjee74.medium.com/different-types-of-distances-used-in-machine-learning-8f541bbdce8",
    "1617194": "Same here. Actually for my models, the testing score of cosine is about 0.1 higher than other distance calculation methods."
  },
  "source": "meta"
}