Normalized histogram in MATLAB incorrect? - matlab

I have the following set of data:
X=[4.692
6.328
4.677
6.836
5.032
5.269
5.732
5.083
4.772
4.659
4.564
5.627
4.959
4.631
6.407
4.747
4.920
4.771
5.308
5.200
5.242
4.738
4.758
4.725
4.808
4.618
4.638
7.829
7.702
4.659]; % Sample set
I fitted a Pareto distribution to this using the maximum likelihood method and I obtain the following graph:
Where the following bit of code is what draws the histogram:
[N,edges,bin] = histcounts(X,'BinMethod','auto');
bin_middles=mean([edges(1:end-1);edges(2:end)]);
f_X_sample=N/trapz(bin_middles,N);
bar(bin_middles,f_X_sample,1);;
Am I doing this right? I checked 100 times and the Pareto distribution is indeed optimal, but it seems awfully different from the histogram. Is there an error that may be causing this? Thank you!

I would agree with #tashuhka's comment that you need to think about how you're binning your data.
Imagine the extreme case where you lump everything together into one bin, and then try to fit that single point to a distribution. Your PDF would look nothing like your single square bar. Split into two bins, and now the fit still sucks, but at least one bar is (probably) a little bigger than the other, etc., etc. At the other extreme, every data point has its own bar and the bar graph is nothing but a random forest of bars with only one count.
There are a number of different strategies for choosing an "optimal" bin size that minimizes the number of bins but maximizes the representation of the underlying PDF.
Finally, note that you only have 30 points here, so your other problem may be that you just haven't collected enough data to really nail down the underlying PDF.

Related

scipy.interpolate.griddata slow due to unnecessary data

I have a map with a 600*600 aequidistant x,y grid with associated scalar values.
I have around 1000 x,y coordinates at which I would like to get the bi-linear interpolated map values. Those are randomly placed in an inner center area of the map with arround 400*400 size.
I decided to go with the griddata function with method linear. My understanding is that with linear interpolation I would only need the three nearest grid positions around each coordinate do get the well defined interpolated values. So I would require around 3000 data points of the map to perform the interpolation. The 360k data points are highly unnecessary for this task.
Throwing stupidly the complete map in results in long excecution times of a half minute. Since it's easy to narrow the map already down to the area of interest I could reduce excecution time to nearly 20%.
I am now wondering if I oversaw something in my assumption that I need only the three nearest neighbours for my task. And if not, whether there is a fast solution to filter those 3000 out of the 360k. I assume looping 3000 times over the 360k lines will take longer than to just throw in the inner map.
Edit: I had also a look at the comparisson of the result with 600*600 and the reduced data points. I am actually surprised and concerned about the observation, that the interpolation results differ partly significantly.
So I found out that RegularGridinterpolator is the way to go for me. It's fast and I have a regular grid already.
I tried to sort out my findings with the differences in interpolation value and found griddata to show unexpected behavior for me.
Check out the issue I created for details.
https://github.com/scipy/scipy/issues/17378

Boxplot is broken, only showing one line

so my data centres around different treatments and how they impact the day of germination. image of dodgy boxplot data
A while ago whilst making violin plots in R to show the distribution of when germination occurs according to treatment, I attempted to add a boxplot as a descriptive statistic and was met with only one line.
I contacted many people who simply had no idea what the issue was, I used this same data in another violin plot as part of a bigger data collection with more treatments including this one.
I moved on from this and found it odd, now when I have come to perform stats tests in SPSS, I have the same problem as imaged below. When I try a Mann Whitney U test I am told "cannot compute" due to not having solely two variables, when I try a Kruskal Wallis test I am met with the dodgy boxplot below and I am told pairwise comparisons cannot be done due to less than 3 test fields (i.e. 2).
I am at an absolute loss, I have tried rewriting the data out, copying data labels with 'stratified' 'strat' 's' etc and I have no idea where the problem could lie, if anyone could give me any guidance this would be really appreciated!
Thank you
The dependent variable in question appears to have only values 1, 2, and 3 in the Stratified group. If there is at least one case with a value of 1, at least one case with a value of 3, but most values at 2, then a box plot like you're seeing would be expected. In SPSS, run the EXAMINE procedure (Analyze>Descriptive Statistics>Explore in the menus), specifying the same dependent variable and grouping variable, and asking for percentiles. The box plots should match what you're getting, and in the percentiles table you should see that Tukey's hinges show the same value of 2 for the 25th, 50th, and 75th percentiles.
Tukey's hinges are the basis for the box and the line in box plots. The line is at the median or 50th percentile, and the upper and lower box edges are at the 25th and 75h percentiles, respectively. When all three coincide, you get just a line instead of a box.
There are two types of outlying values identified in box plots in SPSS. Points greater than 1.5 box lengths below or above the box edges are outliers, marked with circles, and points greater than 3 box lengths below or above the box edges are extremes, marked with asterisks. Since the box length here is 0, anything at other values is automatically an extreme.
Pairwise comparisons following a Kruskal-Wallis test are available only when there are at least three groups, since with only two groups the overall or omnibus test has already compared the two groups. I'm not sure what the issue was when trying to run a Mann-Whitney test.

how to approximate time-series data

I'm not sure if this is the right term but I think I want to s̶m̶o̶o̶t̶h̶ ̶a̶n̶d̶/̶o̶r̶ approximate a data set. I have 30 data points as it is presented in the chart below (the red line with dots)
I want to approximate the dataset so it can be described with fewer data points. The black line represents what I want to achieve.
I want to be able to define an approximation level which will control how much the result data set will differ from the original one.
The approximated data set should contain a set of data points which I can connect together using straight lines.
What is the right algorithm or a math function to solve this problem? I don't expect an implementation here, but rather some suggestions where to start.
I wrote my implementation of the approximation algorithm. It works in most of the cases, but there are certain situations in which it returns non-optimal data.
The example below shows three dotted lines. Thin red line is the original dataset, a thick red-black dotted line is generated by my algorithm, the green line is what I'd like to achieve.
var previousValue;
return array.map(function (dataPoint, index, fullArray) {
var approximation = dataPoint;
if (index > 0) {
if (Math.abs(previousValue - value) < tolerance) {
approximation = previousValue;
} else {
previousValue = dataPoint;
}
} else {
previousValue = dataPoint;
}
return approximation;
});
There are two options here:
if the shown "glitch" in the data is significant, meaning that you cannot smooth it.
if all data shown can be approximated and the "glitch" is insignificant
In (1) case, you may consider approximate by templates (e.g. wavelet) or use basic differential analysis to detect and keep the "glitch" (e.g. meshes).
In (2) case, you may use MA, ARIMA to fit, where the "glitch" can be analyzed further through the roots
Okay, point of clarification, are you looking to smooth the data or approximate it? If you are going to smooth the data, by definition, it will get rid of the little bumps and dips in the data series. On the other hand if the goal is to accurately portray all those dips and bumps, then you do NOT want smoothing. I'm going to talk about smoothing, you tell me if you want the other.
Okay, the best way I know to smooth data is to use an alpha value. The equation is Tn+1=(1-α)Tn+αDatan+1. What this means is that you set the portion of the next function point which is affected by your series history and the portion which is affected by the current data point.
Example graph with alpha = .5
Take a look at this data. Here the α=.5. So the function conforms to the data, but not a lot. The one below is the same, but the alpha is .25. So the data is followed even less, but the function is a lot smoother. There is also a third option where α decreases over time. Initially it can be very high, so you quickly follow the data, but then as α decreases over time the trend becomes smoother and stays smooth over time. Finally, you can set a hard limit on the minimum α This will ensure that you will always have some minimum responsiveness to the data.
Example graph with alpha = .25

MATLAB: Using CONVN for moving average on Matrix

I'm looking for a bit of guidance on using CONVN to calculate moving averages in one dimension on a 3d matrix. I'm getting a little caught up on the flipping of the kernel under the hood and am hoping someone might be able to clarify the behaviour for me.
A similar post that still has me a bit confused is here:
CONVN example about flipping
The Problem:
I have daily river and weather flow data for a watershed at different source locations.
So the matrix is as so,
dim 1 (the rows) represent each site
dim 2 (the columns) represent the date
dim 3 (the pages) represent the different type of measurement (river height, flow, rainfall, etc.)
The goal is to try and use CONVN to take a 21 day moving average at each site, for each observation point for each variable.
As I understand it, I should just be able to use a a kernel such as:
ker = ones(1,21) ./ 21.;
mat = randn(150,365*10,4);
avgmat = convn(mat,ker,'valid');
I tried playing around and created another kernel which should also work (I think) and set ker2 as:
ker2 = [zeros(1,21); ker; zeros(1,21)];
avgmat2 = convn(mat,ker2,'valid');
The question:
The results don't quite match and I'm wondering if I have the dimensions incorrect here for the kernel. Any guidance is greatly appreciated.
Judging from the context of your question, you have a 3D matrix and you want to find the moving average of each row independently over all 3D slices. The code above should work (the first case). However, the valid flag returns a matrix whose size is valid in terms of the boundaries of the convolution. Take a look at the first point of the post that you linked to for more details.
Specifically, the first 21 entries for each row will be missing due to the valid flag. It's only when you get to the 22nd entry of each row does the convolution kernel become completely contained inside a row of the matrix and it's from that point where you get valid results (no pun intended). If you'd like to see these entries at the boundaries, then you'll need to use the 'same' flag if you want to maintain the same size matrix as the input or the 'full' flag (which is default) which gives you the size of the output starting from the most extreme outer edges, but bear in mind that the moving average will be done with a bunch of zeroes and so the first 21 entries wouldn't be what you expect anyway.
However, if I'm interpreting what you are asking, then the valid flag is what you want, but bear in mind that you will have 21 entries missing to accommodate for the edge cases. All in all, your code should work, but be careful on how you interpret the results.
BTW, you have a symmetric kernel, and so flipping should have no effect on the convolution output. What you have specified is a standard moving averaging kernel, and so convolution should work in finding the moving average as you expect.
Good luck!

trainning neural network

I have a picture.1200*1175 pixel.I want to train a net(mlp or hopfield) to learn a specific part of it(201*111pixel) to save its weight to use in a new net(with the same previous feature)only without train it to find that specific part.now there are this questions :what kind of nets is useful;mlp or hopfield,if mlp;the number of hidden layers;the trainlm function is unuseful because "out of memory" error.I convert the picture to a binary image,is it useful?
What exactly do you need the solution to do? Find an object with an image (like "Where's Waldo"?). Will the target object always be the same size and orientation? Might it look different because of lighting changes?
If you just need to find a fixed pattern of pixels within a larger image, I suggest using a straightforward correlation measure, such as crosscorrelation to find it efficiently.
If you need to contend with any of the issues mentioned above, then there are two basic solutions: 1. Build a model using examples of the object in different poses, scalings, etc. so that the model will recognize any of them, or 2. Develop a way to normalize the patch of pixels being examined, to minimize the effect of those distortions (like Hu's invariant moments). If nothing else, yuo'll want to perform some sort of data reduction to get the number of inputs down. Technically, you could also try a model which is invariant to rotations, etc., but I don't know how well those work. I suspect that they are more tempermental than traditional approaches.
I found AdaBoost to be helpful in picking out only important bits of an image. That, and resizing the image to something very tiny (like 40x30) using a Gaussian filter will speed it up and put weight on more of an area of the photo rather than on a tiny insignificant pixel.