keiichishima/gsq

G Square Conditional Independence Test

gsq's Issues

Error handling needed for call to numpy.log

In line 157 of discrete.py, there is a runtime warning that gets raised when numpy.log encounters values equal to or less than zero. Is there anything you can do to fix it?

data issue

Hi, I got a issue when I use discrete.py for my data with discrete data. but my data isn't like 0, 1, 2, 3 ... continues integers. It is like 0, 2, 3, 5. So it got an exception error such as

in line 95
nijk[i, j] += 1
IndexError: index 5 is out of bounds for axis 0 with size 4

data type issue

thank for you rewrite the PC algorithm with python.
But i meet some error with the G square conditional test part.
My array class was numpy.float64 for it's continuous features , and i'm try to conduct the conditional test with ci_test_dis as my conditional test function. But it raise error with my data type:
File "pcalg.py", line 299, in <module> alpha=0.01) File "pcalg.py", line 91, in estimate_skeleton **kwargs) File "E:\Users\jiaxin\Miniconda3\lib\site-packages\gsq\ci_tests.py", line 27, in ci_test_dis return g_square_dis(data_matrix, x, y, s, levels) File "E:\Users\jiaxin\Miniconda3\lib\site-packages\gsq\discrete.py", line 91, in g_square_dis nijk = np.zeros((levels[x], levels[y])) TypeError: 'numpy.float64' object cannot be interpreted as an integer
Is it because the G square test can only applied on the discrete data?
Can you give me some advices which conditional test is suite for my data type?
Thank you!

binary.py question

I don't clearly comprehend the code in the binary.py how it judge two nodes are D-separation.
Would you please add some comments on that file?
Thank you.