{ "id": 5774, "title": "Values of compX_inv and compX_rate_percent_diff", "url": "/competitions/expedia-personalized-sort/discussion/5774", "author_name": "", "post_date": "2013-09-16T15:32:46.197Z", "votes": null, "comment_count": 3, "views": 1443, "content": "

1) The definition for compX_inv is as follows:

\n

+1 if competitor X does not have availability in the hotel; 0 if both
Expedia and competitor X have availability; null signifies there is no
competitive data

\n

Question: What does  -1 signify for this feature? A value of -1 is specified in the definition of compX_rate, but not for compX_inv.

\n

2) The definition for compX_rate_percent_diff is as follows:

\n

The absolute percentage difference (if one exists) between Expedia and
competitor X’s price (Expedia’s price the denominator); null signifies
there is no competitive data

\n

Question A: Why are some of the values so crazy large? For example, the max comp2_rate_percent_diff is 1 439 917%

\n

Question B: Is the formula for this feature as is:

\n

| ExpediaPrice - CompXPrice | / ExpediaPrice

", "messages": [ { "id": "30981", "postDate": "09/16/2013 15:32:46", "content": "

1) The definition for compX_inv is as follows:

\n

+1 if competitor X does not have availability in the hotel; 0 if both
Expedia and competitor X have availability; null signifies there is no
competitive data

\n

Question: What does  -1 signify for this feature? A value of -1 is specified in the definition of compX_rate, but not for compX_inv.

\n

2) The definition for compX_rate_percent_diff is as follows:

\n

The absolute percentage difference (if one exists) between Expedia and
competitor X’s price (Expedia’s price the denominator); null signifies
there is no competitive data

\n

Question A: Why are some of the values so crazy large? For example, the max comp2_rate_percent_diff is 1 439 917%

\n

Question B: Is the formula for this feature as is:

\n

| ExpediaPrice - CompXPrice | / ExpediaPrice

", "rawMarkdown": "", "votes": null }, { "id": "30982", "postDate": "09/16/2013 16:13:59", "content": "

1. If compX_inv = -1 then a given hotel should not appear in rankings. Cases where compX_inv = -1 correspond to matching to aggregate scores, see the last post under https://www.kaggle.com/c/expedia-personalized-sort/forums/t/5690/questions-about-search-results.

\n

Assume for example that over a certain period (and for specific check-in and check-out dates) a hotel had no inventory on Expedia (i.e. had scores of -1), except for one day when the search was performed when we had available rooms. Assuming we can not match exactly a hotel, search, check-in and check-out dates, we use a fall-back score, which in this case is -1.

\n

A safe strategy is in such cases would be to set compX_inv = NULL.

\n

2.A. This is a data quality issue. As I wrote before, the scraping mechanism is not always accurate. You should consider such big values as outliers.

\n

2.B. If compx_rate in (-1, 1) then compx_rate_percent_diff = 100*abs(ExpediaPrice - CompXPrice ) / ExpediaPrice else compx_rate_percent_diff = NULL .

\n

 

", "rawMarkdown": "", "votes": null }, { "id": "31216", "postDate": "09/19/2013 19:34:35", "content": "

Just a follow up on this, I noticed many records contain a 0 in the compX_rate column and contain a positive value in the corresponding compX_rate_percent_diff column.  For example, srch_id=67500, site_id=5 has a 0 for comp8_rate and a 60 for comp8_rate_percent_diff.

\n

This shouldn't be possible given that compX_rate = 0 is supposed to represent where a given competitor had a matching price with Expedia, which would imply a 0 percent difference and thus comp8_rate_percent_diff should = 0.

\n

I'm sure this is just some normal data noise, but my question is which should we trust more, the compX_rate column or the compX_rate_percent_diff column?  In the example of srch_id=67500, site_id=5, we can override the compX_rate field with a 1 or override the compX_rate_percent_diff with a 0, but that all depends on which data field we think is more accurate.

\n

Thanks for your help.  And like others have said, thanks for sponsoring this competition! It's a little "outside the box" from other competitions in a fun and interesting way, very cool challenge.

", "rawMarkdown": "", "votes": null }, { "id": "31239", "postDate": "09/20/2013 07:58:28", "content": "

I can see two possible reasons for this unexpected behavior:

\n

- this is a data quality issue as compX_rate, ExpediaPrice and CompXPrice (the latter two giving rise to compx_rate_percent_diff) are taken as they are from our data warehouse, without any cleaning,

\n

- it it because we match to aggregate scores. More precisely, I can imagine this happening when compX_rate often change signs and correspond to different values of compX_rate_percent_diff; then the average of compX_rate over a certain period can be 0, but compX_rate_percent_diff can be different than 0.

\n

 

\n

It is worth stressing that the main reason we decided to use fall back aggregated scores is because the exact matching over hotel, search, check-in and check-out dates gave low coverage. The price we have to pay for a higher coverage is that the signal in these columns is more diluted.

\n

 

\n

 

\n

X

", "rawMarkdown": "", "votes": null } ], "comments": [ { "id": 30982, "author_name": "adamwoz", "author_url": "", "post_date": "09/16/2013 16:13:59", "content": "

1. If compX_inv = -1 then a given hotel should not appear in rankings. Cases where compX_inv = -1 correspond to matching to aggregate scores, see the last post under https://www.kaggle.com/c/expedia-personalized-sort/forums/t/5690/questions-about-search-results.

\n

Assume for example that over a certain period (and for specific check-in and check-out dates) a hotel had no inventory on Expedia (i.e. had scores of -1), except for one day when the search was performed when we had available rooms. Assuming we can not match exactly a hotel, search, check-in and check-out dates, we use a fall-back score, which in this case is -1.

\n

A safe strategy is in such cases would be to set compX_inv = NULL.

\n

2.A. This is a data quality issue. As I wrote before, the scraping mechanism is not always accurate. You should consider such big values as outliers.

\n

2.B. If compx_rate in (-1, 1) then compx_rate_percent_diff = 100*abs(ExpediaPrice - CompXPrice ) / ExpediaPrice else compx_rate_percent_diff = NULL .

\n

 

", "votes": null, "replies": [] }, { "id": 31216, "author_name": "bryangregory", "author_url": "", "post_date": "09/19/2013 19:34:35", "content": "

Just a follow up on this, I noticed many records contain a 0 in the compX_rate column and contain a positive value in the corresponding compX_rate_percent_diff column.  For example, srch_id=67500, site_id=5 has a 0 for comp8_rate and a 60 for comp8_rate_percent_diff.

\n

This shouldn't be possible given that compX_rate = 0 is supposed to represent where a given competitor had a matching price with Expedia, which would imply a 0 percent difference and thus comp8_rate_percent_diff should = 0.

\n

I'm sure this is just some normal data noise, but my question is which should we trust more, the compX_rate column or the compX_rate_percent_diff column?  In the example of srch_id=67500, site_id=5, we can override the compX_rate field with a 1 or override the compX_rate_percent_diff with a 0, but that all depends on which data field we think is more accurate.

\n

Thanks for your help.  And like others have said, thanks for sponsoring this competition! It's a little "outside the box" from other competitions in a fun and interesting way, very cool challenge.

", "votes": null, "replies": [] }, { "id": 31239, "author_name": "adamwoz", "author_url": "", "post_date": "09/20/2013 07:58:28", "content": "

I can see two possible reasons for this unexpected behavior:

\n

- this is a data quality issue as compX_rate, ExpediaPrice and CompXPrice (the latter two giving rise to compx_rate_percent_diff) are taken as they are from our data warehouse, without any cleaning,

\n

- it it because we match to aggregate scores. More precisely, I can imagine this happening when compX_rate often change signs and correspond to different values of compX_rate_percent_diff; then the average of compX_rate over a certain period can be 0, but compX_rate_percent_diff can be different than 0.

\n

 

\n

It is worth stressing that the main reason we decided to use fall back aggregated scores is because the exact matching over hotel, search, check-in and check-out dates gave low coverage. The price we have to pay for a higher coverage is that the signal in these columns is more diluted.

\n

 

\n

 

\n

X

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "30981": "", "30982": "", "31216": "", "31239": "" }, "source": "meta" }