{
  "id": 45766,
  "title": "autoencoder for  categorical vector?",
  "url": "/competitions/favorita-grocery-sales-forecasting/discussion/45766",
  "author_name": "",
  "post_date": "2017-12-15T16:51:31.661312500Z",
  "votes": 1,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Product have unique number, so applying onehot vectors leads to huge vector of 4000 in length. Is it a good idea to use auto encoder to reduce the vector size?</p>",
  "messages": [
    {
      "id": "258173",
      "postDate": "12/15/2017 16:51:31",
      "content": "<p>Product have unique number, so applying onehot vectors leads to huge vector of 4000 in length. Is it a good idea to use auto encoder to reduce the vector size?</p>",
      "rawMarkdown": "Product have unique number, so applying onehot vectors leads to huge vector of 4000 in length. Is it a good idea to use auto encoder to reduce the vector size?",
      "votes": null
    },
    {
      "id": "260612",
      "postDate": "12/20/2017 15:12:13",
      "content": "<p>I am not sure if this is a good idea. I think that you use or not use one-hot for categorical vector do not hurt a lot if these features do have great importance to your model.  Or you can also try this method \"A Preprocessing Scheme for High-Cardinality Categorical Attributes in Classification and Prediction Problems\".</p>",
      "rawMarkdown": "I am not sure if this is a good idea. I think that you use or not use one-hot for categorical vector do not hurt a lot if these features do have great importance to your model.  Or you can also try this method \"A Preprocessing Scheme for High-Cardinality Categorical Attributes in Classification and Prediction Problems\".",
      "votes": null
    },
    {
      "id": "260710",
      "postDate": "12/20/2017 18:57:49",
      "content": "<p>You can use entity embeddings rather than one hot encoding. You may determine the dimensionality as you wish and learn embedding matrix for categorical variables. For reference <a href=\"https://arxiv.org/abs/1604.06737\">https://arxiv.org/abs/1604.06737</a></p>",
      "rawMarkdown": "You can use entity embeddings rather than one hot encoding. You may determine the dimensionality as you wish and learn embedding matrix for categorical variables. For reference https://arxiv.org/abs/1604.06737",
      "votes": null
    },
    {
      "id": "261040",
      "postDate": "12/21/2017 15:35:40",
      "content": "<p>thanks</p>",
      "rawMarkdown": "thanks",
      "votes": null
    },
    {
      "id": "261042",
      "postDate": "12/21/2017 15:35:58",
      "content": "<p>thanks</p>",
      "rawMarkdown": "thanks",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 260612,
      "author_name": "longyin2",
      "author_url": "",
      "post_date": "12/20/2017 15:12:13",
      "content": "<p>I am not sure if this is a good idea. I think that you use or not use one-hot for categorical vector do not hurt a lot if these features do have great importance to your model.  Or you can also try this method \"A Preprocessing Scheme for High-Cardinality Categorical Attributes in Classification and Prediction Problems\".</p>",
      "votes": null,
      "replies": [
        {
          "id": 261040,
          "author_name": "josthom",
          "author_url": "",
          "post_date": "12/21/2017 15:35:40",
          "content": "<p>thanks</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 260710,
      "author_name": "keremt",
      "author_url": "",
      "post_date": "12/20/2017 18:57:49",
      "content": "<p>You can use entity embeddings rather than one hot encoding. You may determine the dimensionality as you wish and learn embedding matrix for categorical variables. For reference <a href=\"https://arxiv.org/abs/1604.06737\">https://arxiv.org/abs/1604.06737</a></p>",
      "votes": null,
      "replies": [
        {
          "id": 261042,
          "author_name": "josthom",
          "author_url": "",
          "post_date": "12/21/2017 15:35:58",
          "content": "<p>thanks</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "258173": "Product have unique number, so applying onehot vectors leads to huge vector of 4000 in length. Is it a good idea to use auto encoder to reduce the vector size?",
    "260612": "I am not sure if this is a good idea. I think that you use or not use one-hot for categorical vector do not hurt a lot if these features do have great importance to your model.  Or you can also try this method \"A Preprocessing Scheme for High-Cardinality Categorical Attributes in Classification and Prediction Problems\".",
    "260710": "You can use entity embeddings rather than one hot encoding. You may determine the dimensionality as you wish and learn embedding matrix for categorical variables. For reference https://arxiv.org/abs/1604.06737",
    "261040": "thanks",
    "261042": "thanks"
  },
  "source": "meta"
}