Possible speed up

[vinayak-mehta]

@majd1239 wrote in atlanhq/camelot#427:

While using camelot to extract tables from pdfs. I noticed it's really slow. I profiled the code and turns out that %60 of the bottleneck is from np.isclose here and here as well as multiple other places in core.py:

camelot/camelot/core.py

Line 103 in cd8ac79
 if np.isclose(te.x, x_coord, atol=0.5): 

camelot/camelot/core.py

Line 67 in cd8ac79
 if np.isclose(self.y0, y0, atol=edge_tol): 

The slowdown makes sense since there is a very big overhead with np.isclose if we are dealing with native python floats instead of numpy types.

I switched the method to math.isclose instead and the processing time was reduced to more than half!

I can submit an Pull Request with the changes if the devs agree this is a safe change to make.

Thanks

[FrancoisHuet]

See my comments in the original thread. Once the hybrid and network parsers are merged in, this should no longer be an issue.

[ivoytov]

I made the same local changes in my setup