{ "id": 222488, "title": "How much value is TYPE_BEACON adding?", "url": "/competitions/indoor-location-navigation/discussion/222488", "author_name": "higepon", "post_date": "2021-02-27T11:00:45.844000", "votes": 15, "comment_count": 13, "views": 0, "content": "

Hi folks,
\nI'm trying to use TYPE_BEACON data in addition to TYPE_WIFI data. So far I'm getting very small improvement by adding it. Did anyone successfully make it work?

\n

I thought it would give me more fine grained location info/hint to my model. But when I actually took a look at data very closely I wasn't very sure if I could guess location by TYPE_BEACON data as human.

\n

Btw you can quickly filter training data with the following shell script.

\n
for i in `ls`; do grep -E 'WIFI|BEACON|WAYPOINT' $i > filtered_$i; done\n
", "messages": [ { "id": 1219914, "postDate": "2021-02-27T11:00:45.843Z", "content": "

Hi folks,
\nI'm trying to use TYPE_BEACON data in addition to TYPE_WIFI data. So far I'm getting very small improvement by adding it. Did anyone successfully make it work?

\n

I thought it would give me more fine grained location info/hint to my model. But when I actually took a look at data very closely I wasn't very sure if I could guess location by TYPE_BEACON data as human.

\n

Btw you can quickly filter training data with the following shell script.

\n
for i in `ls`; do grep -E 'WIFI|BEACON|WAYPOINT' $i > filtered_$i; done\n
", "rawMarkdown": "Hi folks,\nI'm trying to use TYPE_BEACON data in addition to TYPE_WIFI data. So far I'm getting very small improvement by adding it. Did anyone successfully make it work?\n\nI thought it would give me more fine grained location info/hint to my model. But when I actually took a look at data very closely I wasn't very sure if I could guess location by TYPE_BEACON data as human.\n\n\nBtw you can quickly filter training data with the following shell script.\n```\nfor i in `ls`; do grep -E 'WIFI|BEACON|WAYPOINT' $i > filtered_$i; done\n```\n\n\n", "votes": 14 }, { "id": 1219933, "postDate": "2021-02-27T11:20:35.453Z", "content": "

Thank you for postint it. I tried it and my validation score became worse.

", "rawMarkdown": "Thank you for postint it. I tried it and my validation score became worse. ", "votes": 3, "replies": [ { "id": 1220001, "postDate": "2021-02-27T12:59:40.640Z", "content": "

Thank you Jiwei for your comment.
\nWould you mind sharing what you did?

\n

In my case I tried 2 things.
\n(a) Treat beacon signal value and timestamp exactly the same as wifi features. => CV became worse.
\n(b) Add beacon as additional feature to wifi features. (wifi features are playing main role there) => CV is slightly better.

", "rawMarkdown": "Thank you Jiwei for your comment.\nWould you mind sharing what you did?\n\nIn my case I tried 2 things.\n(a) Treat beacon signal value and timestamp exactly the same as wifi features. => CV became worse.\n(b) Add beacon as additional feature to wifi features. (wifi features are playing main role there) => CV is slightly better.\n\n" }, { "id": 1220359, "postDate": "2021-02-27T22:17:59.087Z", "content": "

Sorry for the late reply.

\n

What I did is similar to the (a) you listed. I added a dedicated GRU layer of beacon features to the RNN and combined its outputs with other layers.

", "rawMarkdown": "Sorry for the late reply.\n\nWhat I did is similar to the (a) you listed. I added a dedicated GRU layer of beacon features to the RNN and combined its outputs with other layers. ", "votes": 2 } ] }, { "id": 1220301, "postDate": "2021-02-27T20:33:10.883Z", "content": "

I'm also trying to utilize TYPE_BEACON data.
\nMy current attempt is building another model using only TYPE_BEACON. If I successfully make the model good enough, I want to make average with my TYPE_WIFI based model.
\nAs for now, the model looks not good at all. I even haven't submitted a prediction with it yet.

", "rawMarkdown": "I'm also trying to utilize TYPE_BEACON data.\nMy current attempt is building another model using only TYPE_BEACON. If I successfully make the model good enough, I want to make average with my TYPE_WIFI based model.\nAs for now, the model looks not good at all. I even haven't submitted a prediction with it yet. ", "votes": 2, "replies": [ { "id": 1220386, "postDate": "2021-02-27T23:00:52.713Z", "content": "

Thanks for sharing Kouki. Please let me know when you make it work :)

", "rawMarkdown": "Thanks for sharing Kouki. Please let me know when you make it work :)", "votes": 1 }, { "id": 1220390, "postDate": "2021-02-27T23:10:38.887Z", "content": "

Absolutely. 😄

", "rawMarkdown": "Absolutely. 😄", "votes": 1 }, { "id": 1235056, "postDate": "2021-03-11T19:34:44.420Z", "content": "

Now I made another dataset with TYPE_BEACON and the result isn't good.
\nBecause of the nature of Beacon, I thought it should detect the rough position of devices.
\nBut I guess samples of beacons are not that much and many of the beacons are observed with weak signal so it's hard.
\nLike you do, maybe we should use beacon data with other values like wi-fi signals?

", "rawMarkdown": "Now I made another dataset with TYPE_BEACON and the result isn't good.\nBecause of the nature of Beacon, I thought it should detect the rough position of devices.\nBut I guess samples of beacons are not that much and many of the beacons are observed with weak signal so it's hard. \nLike you do, maybe we should use beacon data with other values like wi-fi signals? ", "votes": 1 }, { "id": 1235416, "postDate": "2021-03-12T05:58:21.500Z", "content": "

Thanks for sharing your results. Yeah having Beacon signals in addition to Wifi might work. It didn't work for me though.

", "rawMarkdown": "Thanks for sharing your results. Yeah having Beacon signals in addition to Wifi might work. It didn't work for me though." }, { "id": 1242394, "postDate": "2021-03-17T15:30:27.503Z", "content": "

Hi @kokitanisaka, I was trying to integrate ibeacon data with the wifi data you generated but I am not sure how to do that. It would be great help if you can suggest me some ideas related to this.

", "rawMarkdown": "Hi @kokitanisaka, I was trying to integrate ibeacon data with the wifi data you generated but I am not sure how to do that. It would be great help if you can suggest me some ideas related to this.", "votes": 1 }, { "id": 1242978, "postDate": "2021-03-17T23:48:50.157Z", "content": "

@smritisingh1997
\nTo try to utilize beacon data, I made different kind of dataset. Didn't work well though.
\nTo reply your message, I made a dataset and a notebook public, to express what I tried.
\nI hope it can give you some ideas or insights.

\n

https://www.kaggle.com/kokitanisaka/make-dataset-with-wi-fi-and-beacon
\nhttps://www.kaggle.com/kokitanisaka/unified-ds-wifi-and-beacon

", "rawMarkdown": "@smritisingh1997 \nTo try to utilize beacon data, I made different kind of dataset. Didn't work well though. \nTo reply your message, I made a dataset and a notebook public, to express what I tried.\nI hope it can give you some ideas or insights. \n\nhttps://www.kaggle.com/kokitanisaka/make-dataset-with-wi-fi-and-beacon\nhttps://www.kaggle.com/kokitanisaka/unified-ds-wifi-and-beacon", "votes": 3 }, { "id": 1243169, "postDate": "2021-03-18T04:15:42.367Z", "content": "

Thanks @kokitanisaka for help.

", "rawMarkdown": "Thanks @kokitanisaka for help." } ] }, { "id": 1244119, "postDate": "2021-03-18T17:46:40.060Z", "rawMarkdown": "", "votes": 1, "isDeleted": true }, { "id": 1244113, "postDate": "2021-03-18T17:44:29.227Z", "rawMarkdown": "", "votes": 1, "isDeleted": true } ], "comments": [ { "id": 1219933, "author_name": "Jiwei Liu", "author_url": "", "post_date": "2021-02-27T11:20:35.453000", "content": "

Thank you for postint it. I tried it and my validation score became worse.

", "votes": 3, "replies": [ { "id": 1220001, "author_name": "higepon", "author_url": "", "post_date": "2021-02-27T12:59:40.640000", "content": "

Thank you Jiwei for your comment.
\nWould you mind sharing what you did?

\n

In my case I tried 2 things.
\n(a) Treat beacon signal value and timestamp exactly the same as wifi features. => CV became worse.
\n(b) Add beacon as additional feature to wifi features. (wifi features are playing main role there) => CV is slightly better.

", "votes": 0, "replies": [] }, { "id": 1220359, "author_name": "Jiwei Liu", "author_url": "", "post_date": "2021-02-27T22:17:59.087000", "content": "

Sorry for the late reply.

\n

What I did is similar to the (a) you listed. I added a dedicated GRU layer of beacon features to the RNN and combined its outputs with other layers.

", "votes": 2, "replies": [] } ] }, { "id": 1220301, "author_name": "Kouki", "author_url": "", "post_date": "2021-02-27T20:33:10.883000", "content": "

I'm also trying to utilize TYPE_BEACON data.
\nMy current attempt is building another model using only TYPE_BEACON. If I successfully make the model good enough, I want to make average with my TYPE_WIFI based model.
\nAs for now, the model looks not good at all. I even haven't submitted a prediction with it yet.

", "votes": 2, "replies": [ { "id": 1220386, "author_name": "higepon", "author_url": "", "post_date": "2021-02-27T23:00:52.713000", "content": "

Thanks for sharing Kouki. Please let me know when you make it work :)

", "votes": 1, "replies": [] }, { "id": 1220390, "author_name": "Kouki", "author_url": "", "post_date": "2021-02-27T23:10:38.887000", "content": "

Absolutely. 😄

", "votes": 1, "replies": [] }, { "id": 1235056, "author_name": "Kouki", "author_url": "", "post_date": "2021-03-11T19:34:44.420000", "content": "

Now I made another dataset with TYPE_BEACON and the result isn't good.
\nBecause of the nature of Beacon, I thought it should detect the rough position of devices.
\nBut I guess samples of beacons are not that much and many of the beacons are observed with weak signal so it's hard.
\nLike you do, maybe we should use beacon data with other values like wi-fi signals?

", "votes": 1, "replies": [] }, { "id": 1235416, "author_name": "higepon", "author_url": "", "post_date": "2021-03-12T05:58:21.500000", "content": "

Thanks for sharing your results. Yeah having Beacon signals in addition to Wifi might work. It didn't work for me though.

", "votes": 0, "replies": [] }, { "id": 1242394, "author_name": "Smriti ", "author_url": "", "post_date": "2021-03-17T15:30:27.503000", "content": "

Hi @kokitanisaka, I was trying to integrate ibeacon data with the wifi data you generated but I am not sure how to do that. It would be great help if you can suggest me some ideas related to this.

", "votes": 1, "replies": [] }, { "id": 1242978, "author_name": "Kouki", "author_url": "", "post_date": "2021-03-17T23:48:50.157000", "content": "

@smritisingh1997
\nTo try to utilize beacon data, I made different kind of dataset. Didn't work well though.
\nTo reply your message, I made a dataset and a notebook public, to express what I tried.
\nI hope it can give you some ideas or insights.

\n

https://www.kaggle.com/kokitanisaka/make-dataset-with-wi-fi-and-beacon
\nhttps://www.kaggle.com/kokitanisaka/unified-ds-wifi-and-beacon

", "votes": 3, "replies": [] }, { "id": 1243169, "author_name": "Smriti ", "author_url": "", "post_date": "2021-03-18T04:15:42.367000", "content": "

Thanks @kokitanisaka for help.

", "votes": 0, "replies": [] } ] }, { "id": 1244119, "author_name": "", "author_url": "", "post_date": "2021-03-18T17:46:40.060000", "content": "", "votes": 1, "replies": [] }, { "id": 1244113, "author_name": "", "author_url": "", "post_date": "2021-03-18T17:44:29.227000", "content": "", "votes": 1, "replies": [] } ], "raw_markdown_by_id": { "1219914": "Hi folks,\nI'm trying to use TYPE_BEACON data in addition to TYPE_WIFI data. So far I'm getting very small improvement by adding it. Did anyone successfully make it work?\n\nI thought it would give me more fine grained location info/hint to my model. But when I actually took a look at data very closely I wasn't very sure if I could guess location by TYPE_BEACON data as human.\n\n\nBtw you can quickly filter training data with the following shell script.\n```\nfor i in `ls`; do grep -E 'WIFI|BEACON|WAYPOINT' $i > filtered_$i; done\n```\n\n\n", "1219933": "Thank you for postint it. I tried it and my validation score became worse. ", "1220301": "I'm also trying to utilize TYPE_BEACON data.\nMy current attempt is building another model using only TYPE_BEACON. If I successfully make the model good enough, I want to make average with my TYPE_WIFI based model.\nAs for now, the model looks not good at all. I even haven't submitted a prediction with it yet. ", "1244119": "", "1244113": "" } }