question with RLTL_demo

[jackhu5]

something was wrong with TL when using the notebook, here is the error message:

Traceback(most recent call last):
File "/home/labxu/miniconda3/envs/reinvent4-4.4/bin/reinvent", line 8, in
sys.exit(main())
File "/home/labxu/miniconda3/envs/reinvent4-4.4/lib/python3.10/site-packages/reinvent/Reinvent.py", line 334, in main
runner(
File "/home/labxu/miniconda3/envs/reinvent4-4.4/lib/python3.10/site-packages/reinvent/runmodes/TL/run_transfer_learning.py", line 155, in run_transfer_learning
optimize()
File "/home/labxu/miniconda3/envs/reinvent4-4.4/lib/python3.10/site-packages/reinvent/runmodes/TL/learning.py", line 177, in optimize
mean_nll = self.train_epoch()
File "/home/labxu/miniconda3/envs/reinvent4-4.4/lib/python3.10/site-packages/reinvent/runmodes/TL/reinvent.py", line 45, in train_epoch
return self._train_epoch_common()
File "/home/labxu/miniconda3/envs/reinvent4-4.4/lib/python3.10/site-packages/reinvent/runmodes/TL/learning.py", line 226, in _train_epoch_common
for step, batch in enumerate(self.dataloader):
File "/home/labxu/miniconda3/envs/reinvent4-4.4/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 631, in next
data = self._next_data()
File "/home/labxu/miniconda3/envs/reinvent4-4.4/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 675, in _next_data
data = self._dataset_fetcher.fetch(index) # may raise StopIteration
File "/home/labxu/miniconda3/envs/reinvent4-4.4/lib/python3.10/site-packages/torch/utils/data/_utils/fetch.py", line 51, in fetch
data = [self.dataset[idx] for idx in possibly_batched_index]
File "/home/labxu/miniconda3/envs/reinvent4-4.4/lib/python3.10/site-packages/torch/utils/data/_utils/fetch.py", line 51, in
data = [self.dataset[idx] for idx in possibly_batched_index]
File "/home/labxu/miniconda3/envs/reinvent4-4.4/lib/python3.10/site-packages/reinvent/models/reinvent/models/dataset.py", line 29, in _getitem
encoded = self._vocabulary.encode(tokens)
File "/home/labxu/miniconda3/envs/reinvent4-4.4/lib/python3.10/site-packages/reinvent/models/reinvent/models/vocabulary.py", line 65, in encode
raise RuntimeError(msg)
RuntimeError: unknown token |: ['^', 'C', '1', 'C', 'C', '(', 'c', '2', 'c', 'c', 'c', '(', '-', 'c', '3', '[nH]', 'c', '(', '=', 'O', ')', 'c', '4', 'c', 'c', '(', 'F', ')', 'c', 'c', '(', 'F', ')', 'c', '4', 'n', '3', ')', 'c', 'c', '2', ')', 'C', 'C', 'N', '1', 'C', '|', 'C', '(', 'F', ')', '(', 'F', ')', '(', 'C', '(', 'O', ')', '=', 'O', ')', 'F', '$']

and here is the toml and log
toml:

run_type = "transfer_learning"
device = "cuda:0"
tb_logdir = "tb_TL"
[parameters]
num_epochs = 50
save_every_n_epochs = 2
batch_size = 100
sample_batch_size = 2000
input_model_file = "stage1.chkpt"
output_model_file = "TL_reinvent.model"
smiles_file = "tnks2_train.smi"
validation_smiles_file = "tnks2_validation.smi"
standardize_smiles = true
randomize_smiles = true
randomize_all_smiles = false
internal_diversity = true

log:

21:45:54 Started REINVENT 4.4.22 (C) AstraZeneca 2017, 2023 on 2024-07-14
21:45:54 Command line: /home/labxu/miniconda3/envs/reinvent4-4.4/bin/reinvent -l transfer_learning.log transfer_learning.toml
21:45:54 User labxu on host labxu
21:45:54 Python version 3.10.14
21:45:54 PyTorch version 2.2.1+cu121, git 6c8c5ad5eaf47a62fafbb4a2747198cbffbf1ff0
21:45:54 PyTorch compiled with CUDA version 12.1
21:45:54 RDKit version 2023.09.5
21:45:54 Platform Linux-6.8.0-38-generic-x86_64-with-glibc2.39
21:45:54 CUDA driver version 550.90.07
21:45:54 Number of PyTorch CUDA devices 1
21:45:54 Using CUDA device:0 NVIDIA GeForce RTX 4070 Ti SUPER
21:45:54 GPU memory: 15417 MiB free, 16071 MiB total
21:45:54 Writing TensorBoard summary to /home/labxu/R4_notebooks_output/tb_TL
21:45:54 Starting Transfer Learning
21:45:54 /home/labxu/R4_notebooks_output/stage1.chkpt has valid hash:
{ 'comments': ['RL'],
'creation_date': 0,
'date_format': 'UNIX epoch',
'hash_id': '8d6e6710fe8f1af8c6feeb97c556d1a6',
'hash_id_format': 'xxhash.xxh3_128_hex 3.4.1',
'model_id': '55d68f8a81c04f5a86304ebe1723a0ea',
'model_id_format': 'uuid.uuid4 3.10.12',
'origina_data_source': 'ChEMBL 25',
'updates': [1720964643.1013598]}
21:45:54 Number of network parameters: 5,805,602
21:45:54 Network architecture:
RNN(
(_embedding): Embedding(34, 256)
(_rnn): LSTM(256, 512, num_layers=3, batch_first=True)
(_linear): Linear(in_features=512, out_features=34, bias=True)
)
21:45:54 Using generator Reinvent
21:45:54 Applying filter default to input SMILES
21:45:54 Read 600 input SMILES from /home/labxu/R4_notebooks_output/tnks2_train.smi
21:45:54 Read 150 validation SMILES from /home/labxu/R4_notebooks_output/tnks2_validation.smi
21:45:55 unknown token |: ['^', 'C', '1', 'C', 'C', '(', 'c', '2', 'c', 'c', 'c', '(', '-', 'c', '3', '[nH]', 'c', '(', '=', 'O', ')', 'c', '4', 'c', 'c', '(', 'F', ')', 'c', 'c', '(', 'F', ')', 'c', '4', 'n', '3', ')', 'c', 'c', '2', ')', 'C', 'C', 'N', '1', 'C', '|', 'C', '(', 'F', ')', '(', 'F', ')', '(', 'C', '(', 'O', ')', '=', 'O', ')', 'F', '$']

hope to get any help, thanks.

[jackhu5]

but it's fine with reinvent4-4.3.5

[halx]

Hi,

many thanks for your interest in REINVENT and welcome to the community!

The issue is that some of the SMILES contain a dot . which is a regular fragment operator (the dot then gets translated to the pipe symbol, which, for hostoricaly reasons, has been used for the same reason). You would need to either deleted those SMILES or extract the relevant fragment form it.

I will update the notebook some time later when I am back to work.

Many thanks,
Hannes.

[jackhu5]

thank you