{ "id": 215445, "title": "Ways of solving problem", "url": "/competitions/indoor-location-navigation/discussion/215445", "author_name": "Devin Anzelmo", "post_date": "2021-01-29T20:51:53.532000", "votes": 102, "comment_count": 14, "views": 0, "content": "

Trying to figure out various ways to approach this task. Here is what I am thinking so far. The github page has useful information on the what variables are available in the data.

\n

The RSSI (for both the beacon and the wifi) values are a measure of signal strength and hence likely correlated with distance from the phone to wifi/beacon access point. Because there are many different signals that the phone picks up there should be a identifiable signature of signal strengths that can be used to estimate the phone position.

\n

I am still not sure if bssid's correspond to single physical access points or not, but they seem like the best bet to identify single sources (would crossing them (tuple (ssid, bssid)) with the ssid yield a more unique identifier for access point?)

\n

In the training set the position of a block of wifi signals can be estimated by finding the temporally nearest waypoint. This could be used as a label for training data for use with regressors/classifiers. The phone acceleration etc could be used to improve upon the labels, but this seems like it would be a lot of work and I would ignore it for an initial solution.

\n

For the test set I would find the temporally nearest wifi block, to each of the prediction time points, and use the classifiers/regressor to estimate its position and floor. Again this could be improved by then applying the phones acceleration etc data to estimate the position offset between when the wifi/beacon signals were recorded and the target timestamp.

\n

This seems like a full initial solution. I am wondering if there is anything I missed with respect to the general approach? I would be interested to know if there is a completely different way of approaching this problem or not.

\n

I still don't understand what the importance of last_seen _timestamp is for the wifi. Anyone catch how this would be useful?

", "messages": [ { "id": 1176850, "postDate": "2021-01-29T20:51:53.533Z", "content": "

Trying to figure out various ways to approach this task. Here is what I am thinking so far. The github page has useful information on the what variables are available in the data.

\n

The RSSI (for both the beacon and the wifi) values are a measure of signal strength and hence likely correlated with distance from the phone to wifi/beacon access point. Because there are many different signals that the phone picks up there should be a identifiable signature of signal strengths that can be used to estimate the phone position.

\n

I am still not sure if bssid's correspond to single physical access points or not, but they seem like the best bet to identify single sources (would crossing them (tuple (ssid, bssid)) with the ssid yield a more unique identifier for access point?)

\n

In the training set the position of a block of wifi signals can be estimated by finding the temporally nearest waypoint. This could be used as a label for training data for use with regressors/classifiers. The phone acceleration etc could be used to improve upon the labels, but this seems like it would be a lot of work and I would ignore it for an initial solution.

\n

For the test set I would find the temporally nearest wifi block, to each of the prediction time points, and use the classifiers/regressor to estimate its position and floor. Again this could be improved by then applying the phones acceleration etc data to estimate the position offset between when the wifi/beacon signals were recorded and the target timestamp.

\n

This seems like a full initial solution. I am wondering if there is anything I missed with respect to the general approach? I would be interested to know if there is a completely different way of approaching this problem or not.

\n

I still don't understand what the importance of last_seen _timestamp is for the wifi. Anyone catch how this would be useful?

", "rawMarkdown": "Trying to figure out various ways to approach this task. Here is what I am thinking so far. The [github page](https://github.com/location-competition/indoor-location-competition-20) has useful information on the what variables are available in the data.\n\nThe [RSSI](https://en.wikipedia.org/wiki/Received_signal_strength_indication) (for both the beacon and the wifi) values are a measure of signal strength and hence likely correlated with distance from the phone to wifi/beacon access point. Because there are many different signals that the phone picks up there should be a identifiable signature of signal strengths that can be used to estimate the phone position.\n\nI am still not sure if [bssid](https://en.wikipedia.org/wiki/Service_set_(802.11_network)#Basic_service_set_identifier_(BSSID)'s correspond to single physical access points or not, but they seem like the best bet to identify single sources (would crossing them (tuple (ssid, bssid)) with the ssid yield a more unique identifier for access point?)\n\nIn the training set the position of a block of wifi signals can be estimated by finding the temporally nearest waypoint. This could be used as a label for training data for use with regressors/classifiers. The phone acceleration etc could be used to improve upon the labels, but this seems like it would be a lot of work and I would ignore it for an initial solution. \n\nFor the test set I would find the temporally nearest wifi block, to each of the prediction time points, and use the classifiers/regressor to estimate its position and floor. Again this could be improved by then applying the phones acceleration etc data to estimate the position offset between when the wifi/beacon signals were recorded and the target timestamp. \n\nThis seems like a full initial solution. I am wondering if there is anything I missed with respect to the general approach? I would be interested to know if there is a completely different way of approaching this problem or not. \n\nI still don't understand what the importance of last_seen _timestamp is for the wifi. Anyone catch how this would be useful?", "votes": 100 }, { "id": 1176932, "postDate": "2021-01-29T22:43:04.857Z", "content": "

From what I have read, the magnetic field sensors are affected by local steel, concrete etc. My thought is building a regressor that predicts the location and floor based on the magnetic field data. (x,y,z) I am going to try some kind of one hot encoding for the location, as the magnetic signature should be pretty constant (in my mind) based on the same location - given enough data.

", "rawMarkdown": "From what I have read, the magnetic field sensors are affected by local steel, concrete etc. My thought is building a regressor that predicts the location and floor based on the magnetic field data. (x,y,z) I am going to try some kind of one hot encoding for the location, as the magnetic signature should be pretty constant (in my mind) based on the same location - given enough data.", "votes": 11, "replies": [ { "id": 1176952, "postDate": "2021-01-29T23:14:18.390Z", "content": "

Cool! That is something I did not consider. There would need to be enough variation in the internal structure of the building that it would allow for unique magnetic field signals to form. I would be interested to find out if this works.

", "rawMarkdown": "Cool! That is something I did not consider. There would need to be enough variation in the internal structure of the building that it would allow for unique magnetic field signals to form. I would be interested to find out if this works. ", "votes": 2 } ] }, { "id": 1181481, "postDate": "2021-02-01T23:26:27.780Z", "content": "

This makes me think more about it, thanks

\n

In my experience working with electronic devices, the signal strength is just a reference about the distance, this value is very noise and change if something is between or if you are near metal materials. so this can be taking just like a \"near to …\".
\nI think that the last_seen_timestamp is the refresh time, when the mobile device re validate his connection with the router, for cyber security.

", "rawMarkdown": "This makes me think more about it, thanks\n\nIn my experience working with electronic devices, the signal strength is just a reference about the distance, this value is very noise and change if something is between or if you are near metal materials. so this can be taking just like a \"near to ...\".\nI think that the last_seen_timestamp is the refresh time, when the mobile device re validate his connection with the router, for cyber security. ", "votes": 5 }, { "id": 1176872, "postDate": "2021-01-29T21:13:33.660Z", "content": "

I've also just had a quick look at the data, but I think you are on the right track.

\n
\n

I still don't understand what the importance of last_seen _timestamp is for the wifi

\n
\n

I'm also not 100%, but I believe that this is the last time that the wifi routers responded. I think the time in the left column is the time the phone prints out the list of routers it saw, and the timestamp is when it last saw them (which is strictly earlier than the time it reports on them).

\n

Am I wrong in saying that we are not given the location of the beacons/wifi devices?

\n

If we were it would be a somewhat straightforward triangulation problem, but as far as I can see that information is not available anywhere?

\n

That would mean to turn it into a triangulation problem we would first need to estimate the positions of the beacons/wifi devices, and then we could just triangulate from there.

\n

It would be a lot easier if we knew the location of these devices.

", "rawMarkdown": "I've also just had a quick look at the data, but I think you are on the right track. \n\n> I still don't understand what the importance of last_seen _timestamp is for the wifi\n\nI'm also not 100%, but I believe that this is the last time that the wifi routers responded. I think the time in the left column is the time the phone prints out the list of routers it saw, and the timestamp is when it last saw them (which is strictly earlier than the time it reports on them). \n\nAm I wrong in saying that we are **not** given the location of the beacons/wifi devices? \n\nIf we were it would be a somewhat straightforward triangulation problem, but as far as I can see that information is not available anywhere? \n\nThat would mean to turn it into a triangulation problem we would first need to estimate the positions of the beacons/wifi devices, and then we could just triangulate from there. \n\nIt would be a lot easier if we knew the location of these devices. ", "votes": 4, "replies": [ { "id": 1176893, "postDate": "2021-01-29T21:38:08.007Z", "content": "

Yeah I am pretty sure there is no positional data on physical wifi/beacon devices. I think that the signal does not behave simply. There are interactions with physical stuff (walls etc) and maybe magnetic field which is why we have the data for it. This may make the signal strength/distance relationship more complicated, but I can see how it would make the problem easier if we did have the device positions.

\n

There is more information on the last_seen_timestamp at 11:20 in this video, https://www.youtube.com/watch?reload=9&v=xt3OzMC-XMU. I think he is saying that the wifi with large (or maybe small difference difficult to tell) difference between timestamp, and last_seen_timestamp are more useful in estimating position. I am not sure why this would be the case, but it is something to look at for sure.

", "rawMarkdown": "Yeah I am pretty sure there is no positional data on physical wifi/beacon devices. I think that the signal does not behave simply. There are interactions with physical stuff (walls etc) and maybe magnetic field which is why we have the data for it. This may make the signal strength/distance relationship more complicated, but I can see how it would make the problem easier if we did have the device positions.\n\nThere is more information on the last_seen_timestamp at 11:20 in this video, https://www.youtube.com/watch?reload=9&v=xt3OzMC-XMU. I think he is saying that the wifi with large (or maybe small difference difficult to tell) difference between timestamp, and last_seen_timestamp are more useful in estimating position. I am not sure why this would be the case, but it is something to look at for sure.", "votes": 2 }, { "id": 1176902, "postDate": "2021-01-29T21:54:36.937Z", "content": "

Yeah the quote is:

\n
\n

If you use a wifi (reading) with large timestamp gap… it's very possible you will get a bad result.

\n
\n

So using a wifi device that was seen a long time ago is a bad idea (makes sense)

\n
\n

I think that the signal does not behave simply

\n
\n

Definitely. There are many factors to consider, such as 5GHz and 2.4GHz attenuation rates and material people between the phone and the wifi device.
\nWith enough data you should be able to get a reasonable estimate of where the device is though.

", "rawMarkdown": "Yeah the quote is: \n> If you use a wifi (reading) with large timestamp gap... it's very possible you will get a bad result. \n\nSo using a wifi device that was seen a long time ago is a bad idea (makes sense)\n\n> I think that the signal does not behave simply \n\nDefinitely. There are many factors to consider, such as 5GHz and 2.4GHz attenuation rates and material people between the phone and the wifi device. \nWith enough data you should be able to get a reasonable estimate of where the device is though.", "votes": 7 } ] }, { "id": 1200045, "postDate": "2021-02-14T10:58:21.307Z", "content": "

Hello,
\nI have 2 questions (I asked the organizer in the welcome thread but still got no response) :

\n", "rawMarkdown": "Hello,\nI have 2 questions (I asked the organizer in the welcome thread but still got no response) :\n- how were the ground truth waypoints obtained ?\n- is there any new sites or new floors in test data which are not in train data ? (I have not yet looked at the data)", "votes": 1, "replies": [ { "id": 1200293, "postDate": "2021-02-14T15:06:29.333Z", "content": "

Hi, just quick answers.

\n
\n

how were the ground truth waypoints obtained?

\n
\n

These are in the train text data. There are some lines with TYPE_WAYPOINT.
\nYou can find it in the provided GitHub page, so maybe good to play a bit with these APIs.

\n
\n

is there any new sites or new floors in test data that are not in train data?

\n
\n

As far as I know, all test data sites & floors are covered with the train data.

", "rawMarkdown": "Hi, just quick answers.\n\n>how were the ground truth waypoints obtained?\n\nThese are in the train text data. There are some lines with TYPE_WAYPOINT.\nYou can find it in the provided [GitHub page](https://github.com/location-competition/indoor-location-competition-20), so maybe good to play a bit with these APIs.\n\n>is there any new sites or new floors in test data that are not in train data?\n\nAs far as I know, all test data sites & floors are covered with the train data. ", "votes": 3 }, { "id": 1200461, "postDate": "2021-02-14T17:31:10.320Z", "content": "

Thanks @satokiogiso for your reply :-)
\nMy first question was more about the annotation mechanism/procedure used to produce the ground truth waypoints.
\nI see on the github page \"Location surveyor labeled on the map\" for TYPE_WAYPOINT so I guess the ground truth coordinates were manually determined.

", "rawMarkdown": "Thanks @satokiogiso for your reply :-)\nMy first question was more about the annotation mechanism/procedure used to produce the ground truth waypoints.\nI see on the github page \"Location surveyor labeled on the map\" for TYPE_WAYPOINT so I guess the ground truth coordinates were manually determined.", "votes": 2 } ] }, { "id": 1186454, "postDate": "2021-02-04T20:37:27.803Z", "content": "

Hi! Thanks for a really nice notebook! After going over it and reading on the Github page, the data type for wifi doesn't seem to be explained. How did you understand or did you simply saw it as the same as the beacon or have I missed something?

", "rawMarkdown": "Hi! Thanks for a really nice notebook! After going over it and reading on the Github page, the data type for wifi doesn't seem to be explained. How did you understand or did you simply saw it as the same as the beacon or have I missed something? ", "votes": 1, "replies": [ { "id": 1186552, "postDate": "2021-02-04T22:17:14.823Z", "content": "

I just saw that the RSSI value was signal strength from watching the video about the competition. I chose wifi first because there are many more wifi signals than the number of beacon signals.

", "rawMarkdown": "I just saw that the RSSI value was signal strength from watching the video about the competition. I chose wifi first because there are many more wifi signals than the number of beacon signals. ", "votes": 4 } ] }, { "id": 1179976, "postDate": "2021-02-01T03:18:07.860Z", "content": "

Nice start!

\n

I have a guess about the last_seen_timestamp. Signal from a single wifi source is probably not detectable at every point in a building, especially in shopping malls. So the last_seen_timestamp perhaps indicates the last time the phone was within the range of a given wifi source.

", "rawMarkdown": "Nice start!\n\nI have a guess about the last_seen_timestamp. Signal from a single wifi source is probably not detectable at every point in a building, especially in shopping malls. So the last_seen_timestamp perhaps indicates the last time the phone was within the range of a given wifi source.", "votes": 2, "replies": [ { "id": 1180854, "postDate": "2021-02-01T14:04:02.110Z", "content": "

Yeah, that makes sense. If the wifi hasn't been detected in a while it makes it a less reliable signal. One thing to try is to filter out wifi signals with longer deltas between the last_seen_timestamp, and current timestamp (or newest last_seen_timestamp in the test set), and see if it improves score.

", "rawMarkdown": "Yeah, that makes sense. If the wifi hasn't been detected in a while it makes it a less reliable signal. One thing to try is to filter out wifi signals with longer deltas between the last_seen_timestamp, and current timestamp (or newest last_seen_timestamp in the test set), and see if it improves score. ", "votes": 2 } ] }, { "id": 1244209, "postDate": "2021-03-18T19:12:25.223Z", "rawMarkdown": "", "votes": 1, "isDeleted": true } ], "comments": [ { "id": 1176932, "author_name": "Ken Miller", "author_url": "", "post_date": "2021-01-29T22:43:04.857000", "content": "

From what I have read, the magnetic field sensors are affected by local steel, concrete etc. My thought is building a regressor that predicts the location and floor based on the magnetic field data. (x,y,z) I am going to try some kind of one hot encoding for the location, as the magnetic signature should be pretty constant (in my mind) based on the same location - given enough data.

", "votes": 11, "replies": [ { "id": 1176952, "author_name": "Devin Anzelmo", "author_url": "", "post_date": "2021-01-29T23:14:18.390000", "content": "

Cool! That is something I did not consider. There would need to be enough variation in the internal structure of the building that it would allow for unique magnetic field signals to form. I would be interested to find out if this works.

", "votes": 2, "replies": [] } ] }, { "id": 1181481, "author_name": "DavidCastillo", "author_url": "", "post_date": "2021-02-01T23:26:27.780000", "content": "

This makes me think more about it, thanks

\n

In my experience working with electronic devices, the signal strength is just a reference about the distance, this value is very noise and change if something is between or if you are near metal materials. so this can be taking just like a \"near to …\".
\nI think that the last_seen_timestamp is the refresh time, when the mobile device re validate his connection with the router, for cyber security.

", "votes": 5, "replies": [] }, { "id": 1176872, "author_name": "fnands", "author_url": "", "post_date": "2021-01-29T21:13:33.660000", "content": "

I've also just had a quick look at the data, but I think you are on the right track.

\n
\n

I still don't understand what the importance of last_seen _timestamp is for the wifi

\n
\n

I'm also not 100%, but I believe that this is the last time that the wifi routers responded. I think the time in the left column is the time the phone prints out the list of routers it saw, and the timestamp is when it last saw them (which is strictly earlier than the time it reports on them).

\n

Am I wrong in saying that we are not given the location of the beacons/wifi devices?

\n

If we were it would be a somewhat straightforward triangulation problem, but as far as I can see that information is not available anywhere?

\n

That would mean to turn it into a triangulation problem we would first need to estimate the positions of the beacons/wifi devices, and then we could just triangulate from there.

\n

It would be a lot easier if we knew the location of these devices.

", "votes": 4, "replies": [ { "id": 1176893, "author_name": "Devin Anzelmo", "author_url": "", "post_date": "2021-01-29T21:38:08.007000", "content": "

Yeah I am pretty sure there is no positional data on physical wifi/beacon devices. I think that the signal does not behave simply. There are interactions with physical stuff (walls etc) and maybe magnetic field which is why we have the data for it. This may make the signal strength/distance relationship more complicated, but I can see how it would make the problem easier if we did have the device positions.

\n

There is more information on the last_seen_timestamp at 11:20 in this video, https://www.youtube.com/watch?reload=9&v=xt3OzMC-XMU. I think he is saying that the wifi with large (or maybe small difference difficult to tell) difference between timestamp, and last_seen_timestamp are more useful in estimating position. I am not sure why this would be the case, but it is something to look at for sure.

", "votes": 2, "replies": [] }, { "id": 1176902, "author_name": "fnands", "author_url": "", "post_date": "2021-01-29T21:54:36.937000", "content": "

Yeah the quote is:

\n
\n

If you use a wifi (reading) with large timestamp gap… it's very possible you will get a bad result.

\n
\n

So using a wifi device that was seen a long time ago is a bad idea (makes sense)

\n
\n

I think that the signal does not behave simply

\n
\n

Definitely. There are many factors to consider, such as 5GHz and 2.4GHz attenuation rates and material people between the phone and the wifi device.
\nWith enough data you should be able to get a reasonable estimate of where the device is though.

", "votes": 7, "replies": [] } ] }, { "id": 1200045, "author_name": "ISMAX", "author_url": "", "post_date": "2021-02-14T10:58:21.307000", "content": "

Hello,
\nI have 2 questions (I asked the organizer in the welcome thread but still got no response) :

\n", "votes": 1, "replies": [ { "id": 1200293, "author_name": "sog", "author_url": "", "post_date": "2021-02-14T15:06:29.333000", "content": "

Hi, just quick answers.

\n
\n

how were the ground truth waypoints obtained?

\n
\n

These are in the train text data. There are some lines with TYPE_WAYPOINT.
\nYou can find it in the provided GitHub page, so maybe good to play a bit with these APIs.

\n
\n

is there any new sites or new floors in test data that are not in train data?

\n
\n

As far as I know, all test data sites & floors are covered with the train data.

", "votes": 3, "replies": [] }, { "id": 1200461, "author_name": "ISMAX", "author_url": "", "post_date": "2021-02-14T17:31:10.320000", "content": "

Thanks @satokiogiso for your reply :-)
\nMy first question was more about the annotation mechanism/procedure used to produce the ground truth waypoints.
\nI see on the github page \"Location surveyor labeled on the map\" for TYPE_WAYPOINT so I guess the ground truth coordinates were manually determined.

", "votes": 2, "replies": [] } ] }, { "id": 1186454, "author_name": "Johan", "author_url": "", "post_date": "2021-02-04T20:37:27.803000", "content": "

Hi! Thanks for a really nice notebook! After going over it and reading on the Github page, the data type for wifi doesn't seem to be explained. How did you understand or did you simply saw it as the same as the beacon or have I missed something?

", "votes": 1, "replies": [ { "id": 1186552, "author_name": "Devin Anzelmo", "author_url": "", "post_date": "2021-02-04T22:17:14.823000", "content": "

I just saw that the RSSI value was signal strength from watching the video about the competition. I chose wifi first because there are many more wifi signals than the number of beacon signals.

", "votes": 4, "replies": [] } ] }, { "id": 1179976, "author_name": "Tolga", "author_url": "", "post_date": "2021-02-01T03:18:07.860000", "content": "

Nice start!

\n

I have a guess about the last_seen_timestamp. Signal from a single wifi source is probably not detectable at every point in a building, especially in shopping malls. So the last_seen_timestamp perhaps indicates the last time the phone was within the range of a given wifi source.

", "votes": 2, "replies": [ { "id": 1180854, "author_name": "Devin Anzelmo", "author_url": "", "post_date": "2021-02-01T14:04:02.110000", "content": "

Yeah, that makes sense. If the wifi hasn't been detected in a while it makes it a less reliable signal. One thing to try is to filter out wifi signals with longer deltas between the last_seen_timestamp, and current timestamp (or newest last_seen_timestamp in the test set), and see if it improves score.

", "votes": 2, "replies": [] } ] }, { "id": 1244209, "author_name": "", "author_url": "", "post_date": "2021-03-18T19:12:25.223000", "content": "", "votes": 1, "replies": [] } ], "raw_markdown_by_id": { "1176850": "Trying to figure out various ways to approach this task. Here is what I am thinking so far. The [github page](https://github.com/location-competition/indoor-location-competition-20) has useful information on the what variables are available in the data.\n\nThe [RSSI](https://en.wikipedia.org/wiki/Received_signal_strength_indication) (for both the beacon and the wifi) values are a measure of signal strength and hence likely correlated with distance from the phone to wifi/beacon access point. Because there are many different signals that the phone picks up there should be a identifiable signature of signal strengths that can be used to estimate the phone position.\n\nI am still not sure if [bssid](https://en.wikipedia.org/wiki/Service_set_(802.11_network)#Basic_service_set_identifier_(BSSID)'s correspond to single physical access points or not, but they seem like the best bet to identify single sources (would crossing them (tuple (ssid, bssid)) with the ssid yield a more unique identifier for access point?)\n\nIn the training set the position of a block of wifi signals can be estimated by finding the temporally nearest waypoint. This could be used as a label for training data for use with regressors/classifiers. The phone acceleration etc could be used to improve upon the labels, but this seems like it would be a lot of work and I would ignore it for an initial solution. \n\nFor the test set I would find the temporally nearest wifi block, to each of the prediction time points, and use the classifiers/regressor to estimate its position and floor. Again this could be improved by then applying the phones acceleration etc data to estimate the position offset between when the wifi/beacon signals were recorded and the target timestamp. \n\nThis seems like a full initial solution. I am wondering if there is anything I missed with respect to the general approach? I would be interested to know if there is a completely different way of approaching this problem or not. \n\nI still don't understand what the importance of last_seen _timestamp is for the wifi. Anyone catch how this would be useful?", "1176932": "From what I have read, the magnetic field sensors are affected by local steel, concrete etc. My thought is building a regressor that predicts the location and floor based on the magnetic field data. (x,y,z) I am going to try some kind of one hot encoding for the location, as the magnetic signature should be pretty constant (in my mind) based on the same location - given enough data.", "1181481": "This makes me think more about it, thanks\n\nIn my experience working with electronic devices, the signal strength is just a reference about the distance, this value is very noise and change if something is between or if you are near metal materials. so this can be taking just like a \"near to ...\".\nI think that the last_seen_timestamp is the refresh time, when the mobile device re validate his connection with the router, for cyber security. ", "1176872": "I've also just had a quick look at the data, but I think you are on the right track. \n\n> I still don't understand what the importance of last_seen _timestamp is for the wifi\n\nI'm also not 100%, but I believe that this is the last time that the wifi routers responded. I think the time in the left column is the time the phone prints out the list of routers it saw, and the timestamp is when it last saw them (which is strictly earlier than the time it reports on them). \n\nAm I wrong in saying that we are **not** given the location of the beacons/wifi devices? \n\nIf we were it would be a somewhat straightforward triangulation problem, but as far as I can see that information is not available anywhere? \n\nThat would mean to turn it into a triangulation problem we would first need to estimate the positions of the beacons/wifi devices, and then we could just triangulate from there. \n\nIt would be a lot easier if we knew the location of these devices. ", "1200045": "Hello,\nI have 2 questions (I asked the organizer in the welcome thread but still got no response) :\n- how were the ground truth waypoints obtained ?\n- is there any new sites or new floors in test data which are not in train data ? (I have not yet looked at the data)", "1186454": "Hi! Thanks for a really nice notebook! After going over it and reading on the Github page, the data type for wifi doesn't seem to be explained. How did you understand or did you simply saw it as the same as the beacon or have I missed something? ", "1179976": "Nice start!\n\nI have a guess about the last_seen_timestamp. Signal from a single wifi source is probably not detectable at every point in a building, especially in shopping malls. So the last_seen_timestamp perhaps indicates the last time the phone was within the range of a given wifi source.", "1244209": "" } }