ENH: hierarchical clustering

[kzsigmond]

Speeding up hierarchical clustering

The python version of hierarchical clustering could potentially be faster. The code calculates the similarity in max_indices with gen_sim_dict. We are calculating all the similarity indices in the dictionary with the default k=1 value. If I am not mistaken, calculate_isim calculates the same for RR, JT, SM indices as gen_sim_dict in this case.

Proposed change

replace line 67 in iSIM/iSIM/clustering.py with:
s = calculate_isim(data=fp1+fp2, n_objects=n, n_ary=n_ary)

Initial timing results

This is for 50 fingerprints (size 2048):

• gen_sim_dict took 5.3 seconds
• calculate_isim took 1.2 seconds

Potential issues

We would not be able to do hierarchical clustering with the other similarities in gen_sim_dict and/or try different k values.