Negative weights or values in matrix A during STATIS execution

[saga-one]

Hi,
First of all, thank you for sharing this useful package.

I’m currently working with pySTATIS using my own data (symmetric contact matrices), and I encountered an issue while running the STATIS pipeline. Specifically, the matrix A returned by get_A_STATIS() contains negative values, which then causes a failure when taking the square root of A in the gsvd() function (due to NaNs or Infs).

I traced this back and found that the issue might come from the weights generated by rv_pca()—in my case, some of them are negative. However, when I tested the code with the provided wine dataset, the weights were all positive and the pipeline worked fine.

I’m wondering:

• Is it expected that weights from rv_pca() can be negative depending on the data?

• Or does this suggest I might be doing something wrong with the input (e.g., affinity calculation or normalization)?

• Should negative weights be clipped or adjusted before building matrix A?

I’d appreciate any clarification or advice on how to handle this properly. Thank you again for your work!