oxcsml / kerpy Goto Github PK

Hi,

Thanks a lot for making this HSIC independence test package available!

I have python 3.6.1 on a windows machine. I tried "pip install kerpy" but ended with an error message:
Could not find a version that satisfies the requirement kerpy (from versions:)
no matching distribution found for kerpy

Then, I downloaded the package and noticed that there is not a setup.py file.

What is the best way to install kerpy?

Many thanks,

Ivan

Good morning,

First it was a pleasure to find these package to compute HSIC index. Congratulations, it is an impressive work.

I use python 3.6 on ubuntu16.04 and i had some troubles to install/use kerpy. I solved all the issues.
So here my guide line:

1. Update all python files from python2.x to python3.x bu using 2to3-3.6 helper:

• 2to3-3.6 -w independence_testing/*.py
• 2to3-3.6 -w kerpy/*.py
• 2to3-3.6 -w tools/*.py
• 2to3-3.6 -w weak_conditional_independence_testing/*.py

2. Update file "independence_testing/HSICBlockTestObject.py"by correcting this line: "num_blocks = ( self.num_samples ) / self.blocksize" by the new one "num_blocks = int(( self.num_samples ) // self.blocksize)". Indeed, without that, you have this error: "TypeError: 'float' object cannot be interpreted as an integer".

Thank you for this package. Again, it is a very good work.

Best regards.

I tried to reproduce your reported type I error being asymptotically alpha with the simple linear dependencies and using the spectral approach for HSIC.

I changed line 19 to run python ExperimentsHSICSpectral.py 10 --dimX 10 --hypothesis null over 200 trials.
Shouldn't I expect a type I error of 0.05 for these results? I get a type I error of 0.

The same issue with test power / type II error. Test power becomes 0 when I expected 1 for sample size 100 and dimensions 10.

https://github.com/oxmlcs/kerpy/blob/50b175961d13e0e1f625aa987ae41cb98bfe4d84/independence_testing/ExampleHSIC.py#L15

that should be from independence_testing.HSICTestObject import HSICTestObject