Plotting in 2D and 3D

I have 7500 rows and 30 columns. I want to plot and rotate 2 and 3 variables against each other and get regression fits for multiple populations noticeable within this data.

The plots I have been able to generate look very shoddy. I can e-mail examples to anyone that can help.

Thanks

[Amit]

Can you post you plot here? You can save you plot as image and upload here using 'image' button

[Rookshana Trollope]

<<

>>

[Rookshana Trollope]

Hi Amit please let me know if you have received both diagrams. The matlab one and the screen shot.

Thanks Shana

[Rookshana Trollope]

[Rookshana Trollope]

I am also including a matlab diagram with a point maker instead of circles.

[Amit]

I see the plot. You're plotting 7500 data points and 29 variables. I believe you have issue in the screenshot image?

Answer 1

[Walter Roberson]

Use the 'fill' option when you call scatter().

And notice that you are using a different aspect ratio for the plots then what is being used in the other program. The sample is notably wider compared to its height, which is going to have the effect of stretching the plot. You can adjust your axis Position property to get different aspect ratios.

[Rookshana Trollope]

Ok I will try this thanks.

[Rookshana Trollope]

I have checked and yes I do have the fill option, but it does not look like it ?. Even if I change the axis and the height. It still does not look like over 7000 data points.

Answer 2

[Rookshana Trollope]

Herewith another plot and also does not look like more than 7000.

[Amit]

Its there a possibility there are duplicate data? I mean same data point pairs appearing more than 1 times?

[Amit]

Another question is: Is your main purpose is to obtain parameters via regression analysis or plotting?

[Rookshana Trollope]

There is very likely data with the same parameters (x, y values)but they are not the same data point. Yes I want to get multiple regression fits ? That will be my next problem as I might have to observationally choose regression lines. I might have to select clusters from the graphics and I also dont know if matlab can do that.

[Rookshana Trollope]

Herewith a clearer example

[Amit]

#1 2 same (x,yvalue) will be 2 data points but when you plot then those would be overlapping. So you cannot see the difference in scatter plot. It's like plotting (1,1) 30 times or 3000 times but all the points will appear at 1 single point.

#2 Matlab is very capable of doing regression analysis.

Let me rephrase your question and you can tell me if I am right. You have 7500 rows associated with 30 different variables (xi,1; xi,2; ... xi,30) where i is the row. First you want to see which variables have strong influence and then you want to find some correlation. Am I correct?

[Rookshana Trollope]

This data is for brittle failure, yes there will be 30 or 100 failures with the same size and energy but they need to count individually for a regression fit. There are also different types of failures that have a different E to M ratio. Example the big cluster to the left, I am adding one with my interpretation.

[Amit]

Yes, they need to count individually. And Indeed they count individually in regression analysis. The more a data point appears, the more weight it contains towards fit.

Doing this visually might be tricky. However, this fall into statics. You can do test of significance (like null hypothesis) for a variable.

Again my interpretation of your responses. In your case. you are looking for brittle failure. And you have 29 different factors you're testing for? Some of these factor will be insignificant towards your response (brittle failure). You're trying to identify the factors which a significant and then create a model describing the effect of these significant factors towards your response (brittle failure).

[Walter Roberson]

If you are trying to scatter plot data that has multiple points with the same x and y but different value, then you should be using a 3D scatter plot,

scatter3(x, y, value, 'fill')