Issue opening a CPython *.npy file

[A622266]

Description In ulab, I cannot open a *.npy file that was created in CPython.

Background I want to use the ulab function 'sosfilt' on a data stream. I use CPython filter design tools found in Scipy to create an ndarray describing second order sections. I use CPython Numpy function 'save' to save the file in *.npy format. However, when I try to open this file in ulab using ulab function 'load', I get an error indicating that ulab cannot open the file. I have verified that the data saved using CPython 'save' is in ndarray format by using the 'type' function. Of course I can make progress by simply print the output of the CPython filter design tool and creating an ndarray in ulab. But I am hoping to have a more automated process for rapid experimenting. Is it currently possible?

To reproduce Run following code with attached sos.npy (after unzipping)

import ulab
from ulab import numpy as np
from ulab import scipy as spy
print('version string: ', ulab.__version__)

np.load('sos.npy')

Output version string: 6.0.0-2D-c Traceback (most recent call last): File "", line 6, in OSError: [Errno 2] ENOENT

Expected behavior Open 'sos.npy' and recognize it as ndarray in ulab sos.zip

[v923z]

Before I can fix the issue, there are two ways for you to proceed: first is to simply save your data in a text file with savetxt, and then load the contents text file with loadtxt. In fact, you can even do something very similar in python, you don't even need the savetxt/loadtxt functions. Writing your own data reader in python is probably safer, because you can manipulate the exact data format in a much more flexible way.

However, if you know what type of numbers you except, you could even save save them in a binary format, and then fill up your ndarray from a binary string.

But I have the feeling that the issue is that your numbers are of type float64, and ulab most probably doesn't support that on the ESP32. You could find this out by looking at the output of .itemsize in CPython, and micropython, or simply

from ulab import numpy as np

a = np.array([1, 2, 3])
print(a)

[A622266]

Yes you are right the data type is float64 from numpy, and so that probably explains why it doesn't work on esp32. I'll try to come up with a python solution, and avoid use of savetxt/loadtxt as you suggest.

[v923z]

You can explicitely cast your results to float32 in numpy before calling the save function.

But this still needs to be fixed, because the load function should simply bail out, of the format is not supported, instead of crashing.