How to sample points from a data set using a grid?

Question

So I have some data with around a million (r, phi) coordinates, along with their intensities. I want to sample this data in a grid pattern so I can reduce memory used, and plot faster. However I want to sample the data in X,Y as I will be converting the coordinates to (X,Y) coordinates to plot them.

I was thinking I could use a meshgrid to come up with a template I'd like to sample, but I'm stuck on the next step.

I can't seem to find anything useful searching on google or here, but apologies if this is too simple a question!

I'm using numpy and my data is stored as three seperate arrays right now. I was planning to use np.meshgrid and later scipy.interpolate.griddata for interpolation.

r, phi and intensity are all np.arrays with shape (million,)

e.g.

r = array([1560.8, 1560.8003119, 1560.8006238, ..., 3556.831746,
           3558.815873 , 3560.8      ])

I started with this;

r = data[:, 0]  # radius
phi = data[:, 1]  # altitude angle
h2o = data[:, 2]  # intensity

x = r * np.sin(phi)  # It's a left handed coordinate system
z = r * np.cos(phi)

And for the sampling grid I have got this;

Xscale = np.linspace(min(x), max(x), 1000)
Zscale = np.linspace(min(z), max(z), 1000)

[X, Z] = np.meshgrid(Xscale, Zscale)

user_4611565 · Answer

It would have been nice if you have provided some data to work on. It doesn't matter, we will create some.

Lets create x,y values from r,theta arbitrary values :

import numpy as np
import matplotlib.pyplot as plt

theta=np.linspace(0.,50.,1000)
r=np.linspace(5.,10,1000)

x=r*np.sin(theta)
y=r*np.cos(theta)

plt.plot(x,y,linestyle='',marker='.')

The plot gives :

Now add arbitrary intensity values :

intensity=np.sqrt(x**2+y**2)

fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
ax.scatter(x, y, intensity)

The scatter plot gives :

If I understand well we should not be far from your starting point. We have now 3 arrays with 1000 values. We are going to reduce it to a 20x20 mesgrid. We have to first create the x and y bins, then call the binned_statistic_2d method from scipy and that's it.

import scipy.stats as stats

binx=np.linspace(-10.,10.,20)
biny=np.linspace(-10.,10.,20)

ret = stats.binned_statistic_2d(x, y, intensity, 'mean', bins=[binx,biny])

Z=ret.statistic
Z = np.ma.masked_invalid(Z) # allow to mask Nan values got in bins where there is no value
X, Y = np.meshgrid(binx,biny)

plt.pcolor(X,Y,Z)
plt.show()

The pcolor plot gives :

As requested in your comment, we can now go back to the original x,y,z arrays structure.

First, we have to calculate the center coordinates of the bins

binx_centers=(binx[1:] + binx[:-1])/2
biny_centers=(biny[1:] + biny[:-1])/2
Xcenters, Ycenters = np.meshgrid(binx_centers,biny_centers)

Then we can get the not masked values (see explanation above)

xnew=np.ma.masked_array(Xcenters, Z.mask).compressed()
ynew=np.ma.masked_array(Ycenters, Z.mask).compressed()
znew=Z.compressed()

We can check the new size :

print(znew.shape)

Gives only 235 values (instead of 1000.):

(235L,)

And the new scatter plot with the compressed values :

fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
ax.scatter(xnew, ynew, znew)

We obtain :

How to sample points from a data set using a grid?

1 Answers