schwallergroup / ai4chem_course Goto Github PK

When I run through notebook 7 "Reaction prediction" upon training the model I see these warnings appear:

[2024-02-14 12:29:19,397 WARNING] The batch will be filled until we reach 1,its size may exceed 32 tokens

Any guidance on what to do? Related: in the hyperparams I see you use batch_size: 6144, which is a bit of an odd number and looks like you may have tuned it to get rid of the warning?

Thank you

This warning is also addressed on the ONMT forum (link) and they suggest to set batch_type to sents, but I believe this is for version 3 upwards and the notebooks explicitly installs OpenNMT-py==2.2.0. Had a quick test with version 3 but that quickly led me to other issues so I'll stick with 2.2.0 for now.

Hi, I notice in notebook "07 - Reaction Prediction" you use these hyperparam rnn_size. This may be a leftover from some copy pasted code? Otherwise I am confused because we're using transformer architecture, not RNN?

!onmt_train -config example_run/run_config.yaml \
        -seed 42 -gpu_ranks 0  \
        -param_init 0 \
        -param_init_glorot -max_generator_batches 32 \
        -batch_type tokens -batch_size 6144\
        -normalization tokens -max_grad_norm 0  -accum_count 4 \
        -optim adam -adam_beta1 0.9 -adam_beta2 0.998 -decay_method noam  \
        -warmup_steps 8000 -learning_rate 2 -label_smoothing 0.0 \
        -layers 4 -rnn_size  384 -word_vec_size 384 \
        -encoder_type transformer -decoder_type transformer \
        -dropout 0.1 -position_encoding -share_embeddings  \
        -global_attention general -global_attention_function softmax \
        -self_attn_type scaled-dot -heads 8 -transformer_ff 2048 \
        -tensorboard True -tensorboard_log_dir log_dir

Thanks!

Hi,

I try to use the retrosynthetic notebook and I managed to get everything except the "apply_template" where I got the following error:

NameError Traceback (most recent call last)
in <cell line: 2>()
1 prod_1 = rxn_example.split('>>')[1]
----> 2 pred_reactants = apply_template(tplt_example, prod_1)
3
4 # This is the result of applying the template.
5 visualize_mols(pred_reactants[0])

/content/utils.py in apply_template(tmplt, reacts)
47
48 def apply_template(tmplt, reacts):
---> 49 rxn = rdChemReactions.ReactionFromSmarts(tmplt)
50 reactants = [Chem.MolFromSmiles(x) for x in reacts.split('.')]
51 products = rxn.RunReactants(reactants)

NameError: name 'rdChemReactions' is not defined

in the utils.py the function is the following:

def apply_template(tmplt, reacts):
rxn = rdChemReactions.ReactionFromSmarts(tmplt)
reactants = [Chem.MolFromSmiles(x) for x in reacts.split('.')]
products = rxn.RunReactants(reactants)
return products

the rdChemReactions is properly imported I think (I tried manually/separated as well) but what I am able to run it properly only if the input is something like this "'C:1O.[N:3]>>C:1[N:3]'" not the product of the previous function.

https://www.rdkit.org/docs/source/rdkit.Chem.rdChemReactions.html

The notebook is amazing and I am looking forward to use it.
Thanks

Took me a bit to realize there was something under the .qmd file (I looked at the commit history 6e4ea6c)