PPI Model underperforms

[tauseef09]

Hi there @SwiftieH,

Your impressive and intriguing work led me to try running the code myself. However, I found that the performance of the PPI model does not surpass the 50% mark even after 1000 iterations whereas the paper reports the result to be around 97%. I see this issue has been reported before and I apologize for raising it again. I tried training the model using default settings and changing the value of kappa as it could be one possible reason for the underperformance as mentioned by @SwiftieH in one of the other issues. But doing these brings no improvement. The performance remains within the 40-50% ballpark throughout the training phase which may indicate that the model is not learning. What may be the cause of this and what is the potential solution?

Would appreciate it if anyone could help out.