how to deploy sensor nodes randomly in a matrix 100x100

1 次查看（过去 30 天）

显示更早的评论

Debalina Pal 2018-3-27

1
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/390939-how-to-deploy-sensor-nodes-randomly-in-a-matrix-100x100

评论： Walter Roberson 2018-5-18

采纳的回答： Jan

how to deploy sensor nodes randomly in a matrix 100x100 ?

采纳的回答

Jan 2018-3-27

1
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/390939-how-to-deploy-sensor-nodes-randomly-in-a-matrix-100x100#answer_312208

在 MATLAB Online 中打开

M        = zeros(100, 100);
Index    = randperm(100*100, 100);   % Linear index
M(Index) = 1;

Now you have 100 ones in a zero matrix. If "deploy nodes in a matrix" means something else, please be so kind and explain the details.

52 个评论
显示 50更早的评论隐藏 50更早的评论

Walter Roberson 2018-4-3

"yes I have clusters"

I asked, "is it possible that you have one of these .csv files per node ?" and you said "not at all". Your description of your clusters gives no indication that you have multiple measurements of the same underlying data: you have timeseries measurements of different measurements at best.

It is not meaningful to talk about the reliability of the exact maximum and minimum temperature measurements by comparing single readings from different parts of a country against each other. Not even in tiny Liechtenstein, which has mountains in it. In the city I live in, there are two weather official weather stations, one downtown and one at the airport, 10 km away, and the two routinely have different readings (downtown is at the meeting of two important rivers.) It is true that you could probably get away with four stations for Vatican City (less than 1/2 square km), and possibly Monaco (2 square km), but by the time you get to Nauru (21 square km) then placement starts to matter, as the temperature varies across the island. Hmmm -- Nauru might be just about right, if you put stations at the cardinal points, then there might be enough overlap in effective area that you could use two adjacent readings to "sanity check" any one reading.

Debalina Pal 2018-4-5

network structure.png

Thanks a lot for helping me like this.This helps me to clear my doubts.Thank you so much. Now my question is : If We deploy different types sensor nodes in a certain area. These sensors can sense Temperature, Precipitation, Wind speed of that certain area. These sensors collect data from the environment. There is a concept of combiner(i.e. Temperature Combiner, Precipitation Combiner and Wind speed combiner)which are fusing the sensing data from the same kind of sensors and makes a local decision using Bayesian Fusion. And all the combiners are finally fused by Dempster Shafer Fusion for getting global decision. If we follow this type of conditional independence (Directed acyclic graph) (which is attached with this) then what types of inference we can get by using Bayesian fusion and dempster shafer fusion? here,I cannot understand that what probability values I put for those events.Kindly help me to solve this.Thanks again.

Walter Roberson 2018-4-6

The probabilities you should put in would depend upon your available training data.

The csv file you attached before is not at all suited for this kind of operation, as that file only recorded min and max temperatures over 2 months, 3 months, or 4 months, with no information about conditions during the time.

Anyhow, I do not know anything about Dempster Shafer Fusion, and do not know much about data fusion.

I notice that your network is incomplete, in that it predicts that flooding can only occur during cloudy and partly cloudy weather.

Although it is true that some flooding is triggered by heavy rain, the heavy rain could simply be upstream of the flood site. Consider, for example, rain induced by clouds lifting over mountains, then the clouds and mountains could be east of the location where flooding is being recorded, with the rain deposited on the western side mountains by east-traveling clouds and the rain then flowing down the western side of the mountains to flood something further west.

Where I live, the majority of flooding is caused by spring snow melt hundreds of miles south of us (mostly) or west of us (sometimes) (sometimes both at the same time.) That snow melt goes faster on warm sunny days -- and even if rain contributed, the rain could be well over 100 miles away to the south in a different geographic region, with our prevailing weather coming from the west. Oh yes, and that kind of flooding goes on all night for a couple of weeks... the local weather conditions do not stop it (but local heavy rain can make it worse.)

Walter Roberson 2018-4-6

Weather information changes noticeably throughout a single day, especially temperature. But potentially hourly measurements are not required -- one could hypothesize that perhaps only every 2 or 3 or 6 hours might be good enough.

It depends on the level of hypotheses that you want to make. Suppose that you had the hypothesis that floods happen on the day after a rainy day, but only if the temperature on the day of the flood hits 32.8C. That is a valid hypothesis, but is it usefully predictive? Now suppose the hypothesis were that floods happen on the day after a rainy day, but only if the temperature by 10:00 has hit at least 31.2C already: then if you reach 10:00 on the day after a rainy day and the temperature has hit 31.2C already, then you can know to take flood protection steps that same day. The finer the data you have, the more you can explore causation instead of just correlation. One could hypothesize, for example, that rain previous + 31.2C might somehow trigger the floods, and that the additional humidity from the floods might then trigger higher temperatures.

The less data you have, the cruder your available hypotheses.

Walter Roberson 2018-4-23

Then what are the hypotheses?

In order to use naive Bayes you need "events" to measure the occurrence of and then use to calculate conditional probability. A list of temperatures is not an event. Whether there is rain or not is a potential event, but there is no inherent nonzero amount of rain that is an "event" separate from some other nonzero amount of rain. "Not a cloud in the sky" is a potential event. "Partly cloudy" is a poor event, since it applies whenever there is even just one lone cloud at 10 km up. "Partly cloudy" in the sense of the sun being blocked or reduced from time to time needs to be considered in conjunction with the angle to the Sun, since a cloud way off on the horizon that blocks the setting sun might have to do with a weather system that cannot affect you.

Overcast? Another poor event. High thin clouds of ice crystal can cause overcast and that is a very very different weather situation than low heavy clouds that would burst into rain seemingly if a bird flew through them and farted.

A list of pressures — those are not events either.

Now, "temperature exceeds 32.8 degrees", that is an event. So is temperature is between 32.8 and 33.1. So is temperature is between 31.2 and 32.9. Notice the overlap:

boundaries on conditions may be arbitrary
boundaries may overlap
events do not need to be independent to be considered as events.

The arbitrary nature of boundaries for events means that YOU have to decide what boundaries to use. You have to create hypotheses, such as that such and such a range of temperatures is significant. At the time you frame the hypotheses you do not need to have any evidence yet that the range truly is significant: that is what your naive Bayes calculation will tell you.

It is valid to make hypotheses at different scales on the same data, such as binning at 0.2 degrees and also binning at 5 degrees. This would permit you to find effects in groups and zoom down to find the most important subgroups, and it would also allow you to look at a weak effect and zoom out to larger intervals to see if the effect holds (you were looking too narrowly) or perhaps to discover that at larger scale it just washes out, suggesting that it might have been statistically insignificant.

The more events you partition out of the same data, the longer your computation takes, being proportional to the number of events per attribute, all multiplied together.b

But naive Bayes itself is not concerned with how you determine the events: it just needs the multidimensional histogram of the statistics.

Jan 2018-5-18

Please stop this discussion here. The original question to distribute elements in a matrix has been solved already. Attaching 47 further comments about a different topic is not useful. Debalina Pal, please open a new thread.