Incorrect Recovery of Groups in Molecule Decoding/Encoding Process

[blaschma]

Hi,
I was playing with the nice tutorial provided in the repository. I have adapted the last cell of the notebook:

import numpy as np
import random
random.seed(42)

# to generate new molecules by combining random groups
# start with a grammar of groups, in this case `grammar_fragment` is a grammar of groups

# assign random overload_idxs
for group_name in grammar_fragment.vocab:
    group = grammar_fragment.vocab[group_name]
    group.overload_idx = random.randint(0, 100)
    pass

group_names = list(grammar_fragment.vocab.keys())

# pick n random groups
n = 2

n_pop = 0 # add pop tokens for extra branching
random_groups = [random.choice(group_names) for _ in range(n)] + ['[pop]' for _ in range(n_pop)]
random.shuffle(random_groups)


used_groups = []
# combine them into a new group selfies string
new_gselfies = ''
for i, g in enumerate(random_groups):
    if g == '[pop]':
        new_gselfies += '[pop]'
        continue
    #set high prio
    grammar_fragment.vocab[g].priority=50000
    used_groups.append(grammar_fragment.vocab[g])

    n_attachment_points = len(grammar_fragment.vocab[g].attachment_points)-1
    start = random.randint(0, n_attachment_points)
    
    random_range = np.arange(0, n_attachment_points)    
    random_range = [e for e in random_range if e!=start]
    if not random_range or i == len(random_groups)-1:
        new_block = f"[:{start}{g}]"
    else:        
        end = random.choice(random_range)    
        new_block = f"[:{start}{g}]{INDEX_ALPHABET[end]}"

    mol = grammar_fragment.vocab[g].mol
    for atom in mol.GetAtoms():
        atom.SetProp('atomLabel', str(atom.GetIdx() + 1))
    img = Draw.MolToImage(mol)
    print(g)
    display(img)
    new_gselfies += new_block

print(f"Generated group selfies {new_gselfies}")

#Create new GroupGrammer with used groups
grammar = GroupGrammar(used_groups)   
print(f"grammar vocab {grammar.vocab}")
out = grammar.decoder(new_gselfies)
display(out)

ex = grammar.extract_groups(out)
print(f"Extracted groups", ex)
valid = grammar.full_encoder(out, join=True)
print(f"{valid=}")
out = grammar.decoder(valid)
display(out)

This gives: