github/smiles-transformer
1
1
Fork 0
mirror of https://github.com/DSPsleeporg/smiles-transformer.git synced 2026-04-03 00:08:50 -06:00
Code Issues 13 Projects Releases Packages Wiki Activity

Discrepancy in Returned Values for the Same SMILES Input #3

New issue
Open
opened 2025-10-14 16:10:45 -06:00 by navan · 0 comments
navan commented 2025-10-14 16:10:45 -06:00
Owner
Copy link

Originally created by @poinco-gogo on 6/30/2023

Dear developers,

Thank you for sharing the source code. As mentioned in the title, upon executing the provided code, I observed that different values are returned for the same SMILES input. Would it be reasonable to consider this behavior as expected?

the code:

import torch
from build_vocab import WordVocab
from pretrain_trfm import TrfmSeq2seq
from utils import split
from transformers import T5EncoderModel, T5Tokenizer
import re
import gc
import numpy as np
import pandas as pd
import pickle
import math


def smiles_to_vec(Smiles):
    pad_index = 0
    unk_index = 1
    eos_index = 2
    sos_index = 3
    mask_index = 4
    vocab = WordVocab.load_vocab('vocab.pkl')
    def get_inputs(sm):
        seq_len = 220
        sm = sm.split()
        if len(sm)>218:
            print('SMILES is too long ({:d})'.format(len(sm)))
            sm = sm[:109]+sm[-109:]
        ids = [vocab.stoi.get(token, unk_index) for token in sm]
        ids = [sos_index] + ids + [eos_index]
        seg = [1]*len(ids)
        padding = [pad_index]*(seq_len - len(ids))
        ids.extend(padding), seg.extend(padding)
        return ids, seg
    def get_array(smiles):
        x_id, x_seg = [], []
        for sm in smiles:
            a,b = get_inputs(sm)
            x_id.append(a)
            x_seg.append(b)
        return torch.tensor(x_id), torch.tensor(x_seg)
    trfm = TrfmSeq2seq(len(vocab), 256, len(vocab), 4)
    trfm.load_state_dict(torch.load('trfm_12_23000.pkl'))
    trfm.eval()
    x_split = [split(sm) for sm in Smiles]
    xid, xseg = get_array(x_split)
    print(f"xid: {xid}") # <-- ここまで同じ値
    X = trfm.encode(torch.t(xid))
    print(f"X: {X}") # <-- 違う値になってる
    return X


Smiles = ['CC']*2
smiles_vec = smiles_to_vec(Smiles)

the output is:

xid: tensor([[3, 6, 6, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
         0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
         0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
         0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
         0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
         0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
         0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
         0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
         0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
         0, 0, 0, 0],
        [3, 6, 6, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
         0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
         0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
         0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
         0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
         0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
         0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
         0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
         0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
         0, 0, 0, 0]])
X: [[-0.08155493 -0.3144304   0.34284085 ...  1.5726329   0.47515997
   0.67139834]
 [-0.07930924 -0.3067594   0.3409212  ...  1.577628    0.4822625
   0.6750506 ]]

Here is the computational environment I used:
python 3.6.5
numpy 1.19.5
torch 1.10.1+cu113
nvidia V100

I apologize if I have overlooked anything as I am not extensively familiar with transformer architecture.

Sincerely,

*Originally created by @poinco-gogo on 6/30/2023* Dear developers, Thank you for sharing the source code. As mentioned in the title, upon executing the provided code, I observed that different values are returned for the same SMILES input. Would it be reasonable to consider this behavior as expected? the code: ``` import torch from build_vocab import WordVocab from pretrain_trfm import TrfmSeq2seq from utils import split from transformers import T5EncoderModel, T5Tokenizer import re import gc import numpy as np import pandas as pd import pickle import math def smiles_to_vec(Smiles): pad_index = 0 unk_index = 1 eos_index = 2 sos_index = 3 mask_index = 4 vocab = WordVocab.load_vocab('vocab.pkl') def get_inputs(sm): seq_len = 220 sm = sm.split() if len(sm)>218: print('SMILES is too long ({:d})'.format(len(sm))) sm = sm[:109]+sm[-109:] ids = [vocab.stoi.get(token, unk_index) for token in sm] ids = [sos_index] + ids + [eos_index] seg = [1]*len(ids) padding = [pad_index]*(seq_len - len(ids)) ids.extend(padding), seg.extend(padding) return ids, seg def get_array(smiles): x_id, x_seg = [], [] for sm in smiles: a,b = get_inputs(sm) x_id.append(a) x_seg.append(b) return torch.tensor(x_id), torch.tensor(x_seg) trfm = TrfmSeq2seq(len(vocab), 256, len(vocab), 4) trfm.load_state_dict(torch.load('trfm_12_23000.pkl')) trfm.eval() x_split = [split(sm) for sm in Smiles] xid, xseg = get_array(x_split) print(f"xid: {xid}") # <-- ここまで同じ値 X = trfm.encode(torch.t(xid)) print(f"X: {X}") # <-- 違う値になってる return X Smiles = ['CC']*2 smiles_vec = smiles_to_vec(Smiles) ``` the output is: ``` xid: tensor([[3, 6, 6, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [3, 6, 6, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]) X: [[-0.08155493 -0.3144304 0.34284085 ... 1.5726329 0.47515997 0.67139834] [-0.07930924 -0.3067594 0.3409212 ... 1.577628 0.4822625 0.6750506 ]] ``` Here is the computational environment I used: python 3.6.5 numpy 1.19.5 torch 1.10.1+cu113 nvidia V100 I apologize if I have overlooked anything as I am not extensively familiar with transformer architecture. Sincerely,
Sign in to join this conversation.
No Branch/Tag specified
Branches Tags
master
No results found.
Labels
Clear labels
No items
No labels
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference: github/smiles-transformer#3
No description provided.
Delete branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?