{ "id": 78058, "title": "Some tips", "url": "/competitions/elo-merchant-category-recommendation/discussion/78058", "author_name": "Finlay", "post_date": "2019-01-19T10:10:55", "votes": 37, "comment_count": 3, "views": 0, "content": "

In historical_transactions/new_merchant_transactions file:

\n\n
    \n

  1. installments -1 or 999 mean nan.

    2. \n

  2. Category 3 is based on how the payment is made.

    3. \n

  3. Category 2 is based on the state_id.

    4. \n

  4. Category 1 is influenced by the city_id.

    5. \n

  5. purchase_amount has outlier data.

    6. \n

  6. etc...

    7. \n
\n\n

Some useful link:

\n\n", "messages": [ { "id": 458306, "postDate": "2019-01-19T10:10:55Z", "content": "

In historical_transactions/new_merchant_transactions file:

\n\n
    \n

  1. installments -1 or 999 mean nan.

    2. \n

  2. Category 3 is based on how the payment is made.

    3. \n

  3. Category 2 is based on the state_id.

    4. \n

  4. Category 1 is influenced by the city_id.

    5. \n

  5. purchase_amount has outlier data.

    6. \n

  6. etc...

    7. \n
\n\n

Some useful link:

\n\n", "rawMarkdown": "In `historical_transactions/new_merchant_transactions` file:\n\n1. `installments` -1 or 999 mean `nan`.\n\n2. `Category 3` is based on how the payment is made.\n\n3. `Category 2` is based on the state_id.\n\n4. `Category 1` is influenced by the city_id.\n\n5. `purchase_amount` has outlier data.\n\n6. etc...\n\nSome useful link:\n\n- https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/75034\n- https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/72993\n- https://www.kaggle.com/tapioca/category-1-2-3-in-transactions", "votes": 36 }, { "id": 461311, "postDate": "2019-01-25T18:34:08.850Z", "content": "

For the purchase amount\nI denormalize it (add min(purchase_amount)+1) in order to make all values positive. Then the sum becomes more representative of a lavish life of a cardholder.\nAlso a ratio to the total amount (say - amount spent on each merchant_category to the total amount spent by the cardholder) becomes meaningful. And no divide by zero problem.

", "rawMarkdown": "For the purchase amount\nI denormalize it (add min(purchase_amount)+1) in order to make all values positive. Then the sum becomes more representative of a lavish life of a cardholder.\nAlso a ratio to the total amount (say - amount spent on each merchant_category to the total amount spent by the cardholder) becomes meaningful. And no divide by zero problem." }, { "id": 459609, "postDate": "2019-01-22T04:41:23.620Z", "content": "

Re: purchase amount\nSome ideas:

\n\n
    \n
  1. transform purchase amount into quantiles and treat it as a categorical variable
    2. \n
  2. cutoff the outliers by setting value of all outliers to a large number at the higher part of the range
    3. \n
", "rawMarkdown": "Re: purchase amount\nSome ideas:\n\n1. transform purchase amount into quantiles and treat it as a categorical variable \n2. cutoff the outliers by setting value of all outliers to a large number at the higher part of the range" }, { "id": 458390, "postDate": "2019-01-19T13:53:17.963Z", "rawMarkdown": "", "votes": 3, "isDeleted": true } ], "comments": [ { "id": 461311, "author_name": "Costa.Medvedev", "author_url": "", "post_date": "2019-01-25T18:34:08.850000", "content": "

For the purchase amount\nI denormalize it (add min(purchase_amount)+1) in order to make all values positive. Then the sum becomes more representative of a lavish life of a cardholder.\nAlso a ratio to the total amount (say - amount spent on each merchant_category to the total amount spent by the cardholder) becomes meaningful. And no divide by zero problem.

", "votes": 0, "replies": [] }, { "id": 459609, "author_name": "sergeman", "author_url": "", "post_date": "2019-01-22T04:41:23.620000", "content": "

Re: purchase amount\nSome ideas:

\n\n
    \n
  1. transform purchase amount into quantiles and treat it as a categorical variable
    2. \n
  2. cutoff the outliers by setting value of all outliers to a large number at the higher part of the range
    3. \n
", "votes": 0, "replies": [] }, { "id": 458390, "author_name": "", "author_url": "", "post_date": "2019-01-19T13:53:17.963000", "content": "", "votes": 3, "replies": [] } ], "raw_markdown_by_id": { "458306": "In `historical_transactions/new_merchant_transactions` file:\n\n1. `installments` -1 or 999 mean `nan`.\n\n2. `Category 3` is based on how the payment is made.\n\n3. `Category 2` is based on the state_id.\n\n4. `Category 1` is influenced by the city_id.\n\n5. `purchase_amount` has outlier data.\n\n6. etc...\n\nSome useful link:\n\n- https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/75034\n- https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/72993\n- https://www.kaggle.com/tapioca/category-1-2-3-in-transactions", "461311": "For the purchase amount\nI denormalize it (add min(purchase_amount)+1) in order to make all values positive. Then the sum becomes more representative of a lavish life of a cardholder.\nAlso a ratio to the total amount (say - amount spent on each merchant_category to the total amount spent by the cardholder) becomes meaningful. And no divide by zero problem.", "459609": "Re: purchase amount\nSome ideas:\n\n1. transform purchase amount into quantiles and treat it as a categorical variable \n2. cutoff the outliers by setting value of all outliers to a large number at the higher part of the range", "458390": "" } }