Data values in the initial state are rounded

[BMP-TUD]

Dear Dr Zwicker,

Thank you for the great package, but I have a question concerning the data formats: When I load an initial state with the Scalar_Field.from_image function and I call .data than the results are roughly correct, but everything that is behind a comma is rounded. Is this intentional or a bug? Thank you for your help in advance,

Best wishes,
Bartosz

[david-zwicker]

I think this might simply be an issue with the display style. Does calling numpy.set_printoptions(precision=15) before you print the data attribute help?

[BMP-TUD]

Thanks for the quick response, it actually does not help, output is the following:

np.set_printoptions(precision=15)

state.data
array([[ 1.,  1.,  1., ...,  1.,  1.,  1.],
       [ 1.,  1.,  1., ...,  1.,  1.,  1.],
       [ 1.,  1.,  1., ...,  1.,  1.,  1.],
       ...,
       [ 1.,  1.,  1., ..., 11., 10.,  9.],
       [ 1.,  1.,  1., ...,  9.,  9.,  9.],
       [ 1.,  1.,  1., ...,  8.,  8.,  8.]])

but it should be:

u
array([[ 1.415278196334839,  1.410358190536499,  1.426051259040833, ...,
         9.334187507629395,  9.264062881469727,  8.686315536499023],
       [ 1.648768186569214,  1.66365647315979 ,  1.695865750312805, ...,
        10.378127098083496,  9.108780860900879,  8.364289283752441],
       [ 1.62944757938385 ,  1.662785649299622,  1.704971432685852, ...,
        11.031474113464355,  9.501060485839844,  8.784432411193848],
       ...,
       [ 1.27515983581543 ,  1.260051012039185,  1.252041459083557, ...,
         1.130288124084473,  1.174511432647705,  1.178279638290405],
       [ 1.275601506233215,  1.277632474899292,  1.279035210609436, ...,
         1.167059898376465,  1.179918646812439,  1.176928877830505],
       [ 1.271483778953552,  1.27246105670929 ,  1.279974102973938, ...,
         1.171810865402222,  1.184486389160156,  1.18511164188385 ]])

Could there be a problem if I load the data from a .tif?

[david-zwicker]

Interesting. The function to read images loads the file using matplotlib.pyplot.imread. Could you test whether this function works on your image?

[BMP-TUD]

This is probably causing the problem, as the results are rounded and the image is loaded as a uint8 and gives this:

test

array([[[  1,   1,   1, 255],
        [  1,   1,   1, 255],
        [  1,   1,   1, 255],
        ...,
        [  9,   9,   9, 255],
        [  9,   9,   9, 255],
        [  8,   8,   8, 255]],
        ...,
        [  1,   1,   1, 255],
        [  1,   1,   1, 255],
        [  1,   1,   1, 255]]], dtype=uint8)

But my original image is of type float32 if read in with tifffile.imread. I checked that pyplot is using either .png format or the format is detected by PIL.Image.open. If I run this than the format is still correct: <PIL.TiffImagePlugin.TiffImageFile image mode=F size=186x138 at 0x1E3296EA0C8> but matplotlib.pyplot.imread is not recognizing it correctly. I changed the imread function in scalar.py to that from tifffile and the data is read in correctly. Maybe there must be an additional check for the format, if it is a .tif and you could implement tifffile.imread instead?

Anyway thanks for your help :)

[david-zwicker]

This sounds like a good idea! Could you create a branch/fork with both options for reading images (e.g., using an argument to distinguish different variants or detecting the right reader based on file extension)? I'm happy to accept pull requests and integrate this into the main branch, but I won't have time to work on it myself during the next weeks.