0
$\begingroup$

I have a dataset containing readings from different sensors. Each sensor can provide distance and signal strength but those data is not always available. At each time interval only three sensors can be read. For example there are 5 sensors in total. At time 0 I have readings from sensor 1,2,3 and at time 1 it could be sensor 2,3,4. The readings are equally time-spaced. So the input dimension would be (3,2) at each time instance. The output would be a corrected distance based on both distance and signal strength for each sensor, thus in this case a (3,1) vector.

However I am having difficulty structuring my data and the network since the sequence varies and discontinues from time to time. What are the approaches of organising data like this as inputs? Any insights would be appreciated.

$\endgroup$

2 Answers 2

1
$\begingroup$

From my understanding of your problem description, I believe a similar problem has been tackled by the authors of RAINDROP. You can take a look at that. The caveat is that the authors use a GNN based approach for the same, which according to your statement would mean the sensors would be modelled as nodes in a graph getting irregularly sampled time-varying inputs.

$\endgroup$
1
  • $\begingroup$ Thank you sidharth I will look into that. $\endgroup$
    – Geng Wang
    Oct 19, 2022 at 5:18
0
$\begingroup$

what i wanna know is what are you actually tryn' to archive? this is data from sensor right ?:

T0
[[1,5] - sensor 1
 [2,7] - sensor 2
 [3,6] - sensor 3
]

i am just curious. what will happen if you do this

T1
[[0,0] - sensor 1
 [2,5] - sensor 2
 [3,7] - sensor 3
 [4,6] - sensor 4
 [0,0] - sensor 5
]

i don't know if i interpreted the question correctly

and if possible can you show us the sample data and the model

$\endgroup$
2
  • $\begingroup$ As it’s currently written, your answer is unclear. Please edit to add additional details that will help others understand how this addresses the question asked. You can find more information on how to write good answers in the help center. $\endgroup$
    – Community Bot
    Oct 19, 2022 at 0:05
  • $\begingroup$ Your interpretation is correct and my data format is pretty much what you have posted. I can definitely add 0s when the sensor data is not available but i am not sure whether that will skew the learning result when the sequences have breaking points like this. $\endgroup$
    – Geng Wang
    Oct 19, 2022 at 5:20

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.