Question

I have 2D data that I am trying to display with imshow. Regions of the data are nan and marked as masked.

Because of things like anisotropy, I use bilinear interpolation and would like to keep doing so. Using is_bad I am pretty sure I am getting the masked values are being rendered the way I want them, but the nan color seems to bleed into the good part of the data causing a blur.

Is there a good way to sharpen up the interface between nan and non-nan data? Do I have to identify the border and draw a line over it or can I do this with imshow parameters?

Thanks!

Eli

Answer 1

Well, this is a bit of a convoluted solution, but I don’t know of a better one.

Just to demonstrate the problem you’re talking about to people who may not be familiar with it:

import numpy as np
import matplotlib.pyplot as plt
data = np.random.random((10,10))
data[[2, 5, 7], [5, 8, 1]] = np.nan

plt.imshow(data)
plt.show()

The blurring is because matplotlib interpolates alpha (transparency) values. Changing this would require low-level tweaking.

However, as long as you don’t need to show the layer beneath the image in the regions of no data, you can do something like this:

import numpy as np
import matplotlib.pyplot as plt
import matplotlib as mpl

# Generate some data...
data = np.random.random((10,10))
data[[2, 5, 7], [5, 8, 1]] = np.nan

# Convert it to a masked array instead of just using nan's..
data = np.ma.masked_invalid(data)

# Plot a version with the invalid data replaced with the mean...
plt.imshow(data.filled(data.mean()))

# Now plot a white (or whatever color you'd like), nearest-interpolated array
# over the invalid values...
bad_data = np.ma.masked_where(~data.mask, data.mask)
plt.imshow(bad_data, interpolation='nearest', cmap=mpl.cm.gray_r)

plt.show()

If we don’t plot the original data using the “filled” version (data.filled(data.mean())), we’ll get “halos” around the blocked-out invalid values: