marlin-codes / hie Goto Github PK

It appears that the existing code does not include provisions for link prediction (LP). Therefore, could you kindly provide guidance on how to undertake training for the LP task?

Hello,

I'm attempting to create the HDO distribution graph as depicted in Figure 4, specifically for the Cora dataset. I have acquired the embedding results best_emb from the raw HGCN model and am in the process of calculating the hyperbolic distances from the origin. However, the graph I'm generating does not align well with the one presented in the article. Below is the code I'm using; could you please assist me in pinpointing any possible issues? I would greatly appreciate your help.

`manifold = geoopt.PoincareBall()
origin = torch.zeros(2708, 8)
hyperbolic_distance = manifold.dist(best_emb, origin)
print("Hyperbolic distance to the origin:", hyperbolic_distance)

import networkx as nx
import matplotlib.pyplot as plt

hdo_values = hyperbolic_distance.cpu().detach().numpy()
hdo_values.shape #(2708,)
hdo_values.mean() #5.1924605
hdo_values.min() #2.5806413
hdo_values.max() #6.1599727, especially the max value is very big!

plt.figure(figsize=(10, 6))
plt.hist(hdo_values, color='blue', alpha=0.5)
plt.title('HDO Distribution')
plt.xlabel('HDO Value')
plt.ylabel('Ratio')
plt.show()

`

Best regards.