Python wrapper seems to go into an infinite loop on small filtrations

[ctralie]

Hi,
First of all, thanks for the great software and for making this open source.

I'm writing here because I'm having some trouble with the Python wrapper, and I wanted to see if you could replicate the issue or tell me if I'm doing something wrong. I'm finding that the python wrapper seems to stall and go into an infinite loop for most of the boundary matrices that I feed it, while the C++ code runs fine on those same boundary matrices. I'm attaching an example boundary matrix where this happens, in the file "crashfiltration.txt." This boundary matrix is not large; there are only 9 vertices and 129 simplices total
crashfiltration.txt

When I run the following command with the phat binary

./phat --ascii crashfiltration.txt out.txt

It gives me an answer very quickly. When, on the other hand, I try to wrap it into python with the following code

import phat

fin = open("crashfiltration.txt")
columns = []
for line in fin.readlines():
    fields = [int(s) for s in line.split()]
    columns.append((fields[0], fields[1::]))
fin.close()
print(columns)

boundary_matrix = phat.boundary_matrix(columns = columns)
pairs = boundary_matrix.compute_persistence_pairs()
pairs.sort()

The code seems to hang. Any idea what's going on?
Many thanks,
Chris Tralie

[xoltar]

Hi @ctralie, thanks for the issue report! I can duplicate the problem on my system. While I'm figuring out what's wrong, you can proceed using this workaround I found for you: If you load the file with the load() method on boundary_matrix instead of creating the matrix by hand, it works with your example.

Here's an example:

import phat

boundary_matrix = phat.boundary_matrix()
boundary_matrix.load('crashfiltration.txt', 't') # 't' for 'text', default is 'b' for binary

pairs = boundary_matrix.compute_persistence_pairs()
pairs.sort()
print("\nThere are %d persistence pairs: " % len(pairs))
for pair in pairs:
    print("Birth: %d, Death: %d" % pair)

[xoltar]

Fixed in 90ed83d (fix-2 branch if you need it immediately, otherwise it will come to master once it's been merged upstream)

[xoltar]

BTW the issue was that the load method (which is what the phat executable uses) was sorting the boundaries for you, and the python wrapper was not. But now it will. :-)

[ctralie]

Ah okay awesome, thank you for the quick reply! And that makes perfect sense. That was actually one line fix for me client side, so I can fix it before you even push to pypi.

Also, you might be interested in a little interface I've been working on around your phat wrapper (which is how I found this bug):
https://github.com/ctralie/phatcech

I even added alpha filtrations there, with a similar runtime to the GUDHI Python bindings

[xoltar]

That's very cool, thanks for sharing!