LabVIEW
I have several 20,000+ point scatter plots that I'd like to graph with a best fit. Whats the best way to do this?
Your question is not entirely clear. Do you just want to show the best fit or do you want to show both?
The fit is easy. Do you have any kind of model? LabVIEW has many fit models built-in (linear, polynominal, etc.) or you can use levenberg Marquardt with more complex mathematical models. Later just generate your fit function for each x-value.
To actually show the scatter plot, I would recommend to generate a 2D histogram (e.g. 200x200) and display it on an intensity graph of the same dimensions. This solves the problem of overlapping data. If a pixel contains several data points, it will be proportionally brighter.
The fit is easy. Do you have any kind of model? LabVIEW has many fit models built-in (linear, polynominal, etc.) or you can use levenberg Marquardt with more complex mathematical models. Later just generate your fit function for each x-value.
To actually show the scatter plot, I would recommend to generate a 2D histogram (e.g. 200x200) and display it on an intensity graph of the same dimensions. This solves the problem of overlapping data. If a pixel contains several data points, it will be proportionally brighter.
You also need to wire the desired polynomial order. The output will be the polynomial coeficients. (ignore the "best fit output, it will contain way too many data points in your case).
Now simply create an array with a suitable selection of x-values and feed them together with the best fit coeficients to the polynomial evaluation function.
See attched example (LabVIEW 7.0). It generates 10000 noisy xy points then fits them to a polynomial which is then displayed using only 32 points.
Now simply create an array with a suitable selection of x-values and feed them together with the best fit coeficients to the polynomial evaluation function.
See attched example (LabVIEW 7.0). It generates 10000 noisy xy points then fits them to a polynomial which is then displayed using only 32 points.
There is no way to fit your data with any kind of polynomial, it is too discontinuous. Couldn't you just do a smoothing (low pass filtering)?
Example: In the attached image, I have convoluted the raw Y data with a Gaussian of 0.5% width, then decimated the data to 209 points total. Looks pretty good to me ;-). You could also try just a simple running average.
Example: In the attached image, I have convoluted the raw Y data with a Gaussian of 0.5% width, then decimated the data to 209 points total. Looks pretty good to me ;-). You could also try just a simple running average.
Well, here's a very primitive way to decimate the data without any fancy algorithm. For this particular case, it looks pretty good. Each loop makes the array 3x smaller, averaging 3 adjacent elements. 4 iterations seem to be perfect. (I have included your raw data in diagram constants).
For a polynomial fit, the more complex the data, the higher the order polynomial is needed to describe it. If it is a simple "banana" (my example above) you're OK with an order of 2 (up to quadratic), if it is s-shaped, you need at least 3rd order, etc. I tried your data with much higher orders (up to 40!) and it did not work.
For a polynomial fit, the more complex the data, the higher the order polynomial is needed to describe it. If it is a simple "banana" (my example above) you're OK with an order of 2 (up to quadratic), if it is s-shaped, you need at least 3rd order, etc. I tried your data with much higher orders (up to 40!) and it did not work.
Dave,
Since you just started with LabVIEW, I thought I point out some improvements to your file reading and parsing methods. Basically, you do way too much work! You can do the entire thing in one simple loop (see attached VI in LabVIEW 7.1). On my machine, The reading of the data is now about 30x faster (much less than a second)). It is very inefficient to read a file line-by line if you need the entire file anyway. Enjoy!
Since you just started with LabVIEW, I thought I point out some improvements to your file reading and parsing methods. Basically, you do way too much work! You can do the entire thing in one simple loop (see attached VI in LabVIEW 7.1). On my machine, The reading of the data is now about 30x faster (much less than a second)). It is very inefficient to read a file line-by line if you need the entire file anyway. Enjoy!
