retrotemp's Introduction

retrotemp

Neural network for predicting template relevance a la Segler and Waller's Neural Symbolic paper.

Dependencies if you want to use the final model

RDKit (most versions should be fine)
numpy

Dpendencies if you want to retrain on your own data

RDKit (most versions should be fine)
tensorflow (r0.12.0)
h5py
numpy

About

Learn to predict template relevance.

Grab reaction precedents from templates stored in MongoDB python scripts/get_reaxys_data.py
Calculate fingerprints and store in .h5 file python scripts/make_data_file.py data/reaxys_limit1000000000_reaxys_v2_transforms_retro_v9_10_5.txt 2048
Train model python retrotemp/nntrain_fingerprint.py -t data/reaxys_limit1000000000_reaxys_v2_transforms_retro_v9_10_5.txt -o 163723 -m models/6d3M_Reaxys_10_5 --fp_len 2048

Find best validation performance

regex="model\.(.*)\.meta"
for f in `ls -tr models/6d3M_Reaxys_10_5/*.meta`
do
    if [[ $f =~ $regex ]]
    then
        ckpt="${BASH_REMATCH[1]}"
        echo $ckpt
        python retrotemp/nntrain_fingerprint.py  -o 163723 -m models/6d3M_Reaxys_10_5 --fp_len 2048 -c "$ckpt" -t data/reaxys_limit1000000000_reaxys_v2_transforms_retro_v9_10_5.txt --test valid
    fi
done

Retrain on whole dataset (?) for same number of epochs. Note: this is because we want a high-performing deployed model and no longer need to hold out any data. python retrotemp/nntrain_fingerprint.py -t data/reaxys_limit1000000000_reaxys_v2_transforms_retro_v9_10_5.txt -o 163723 -m models/6d3M_Reaxys_10_5 --fp_len 2048 --fixed_epochs_train_all 15
Run standalone tensorflow version to dump to numpy arrays

retrotemp's People

Contributors

Stargazers

Watchers

retrotemp's Issues

No module named 'rdkit'

python scripts/get_reaxys_data.py
Traceback (most recent call last):
File "scripts/get_reaxys_data.py", line 3, in
import rdkit.Chem as Chem
ModuleNotFoundError: No module named 'rdkit'

'Tensor' object has no attribute 'sparse_read'

While training the model, by running 'nntrain_fingerprint' script (following the instructions in README), I get the following error -

Traceback (most recent call last):
File "/opt/conda/lib/python3.6/site-packages/tensorflow/python/ops/array_ops.py", line 2734, in gather
return params.sparse_read(indices, name=name)
AttributeError: 'Tensor' object has no attribute 'sparse_read'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "/opt/conda/lib/python3.6/site-packages/tensorflow/python/framework/op_def_library.py", line 510, in _apply_op_helper
preferred_dtype=default_dtype)
File "/opt/conda/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 1104, in internal_convert_to_tensor
ret = conversion_func(value, dtype=dtype, name=name, as_ref=as_ref)
File "/opt/conda/lib/python3.6/site-packages/tensorflow/python/framework/constant_op.py", line 235, in _constant_tensor_conversion_function
return constant(v, dtype=dtype, name=name)
File "/opt/conda/lib/python3.6/site-packages/tensorflow/python/framework/constant_op.py", line 214, in constant
value, dtype=dtype, shape=shape, verify_shape=verify_shape))
File "/opt/conda/lib/python3.6/site-packages/tensorflow/python/framework/tensor_util.py", line 441, in make_tensor_proto
_GetDenseDimensions(values)))
ValueError: Argument must be a dense tensor: range(0, 1) - got shape [1], but wanted [].

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "/opt/conda/lib/python3.6/site-packages/tensorflow/python/framework/op_def_library.py", line 524, in _apply_op_helper
values, as_ref=input_arg.is_ref).dtype.name
File "/opt/conda/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 1104, in internal_convert_to_tensor
ret = conversion_func(value, dtype=dtype, name=name, as_ref=as_ref)
File "/opt/conda/lib/python3.6/site-packages/tensorflow/python/framework/constant_op.py", line 235, in _constant_tensor_conversion_function
return constant(v, dtype=dtype, name=name)
File "/opt/conda/lib/python3.6/site-packages/tensorflow/python/framework/constant_op.py", line 214, in constant
value, dtype=dtype, shape=shape, verify_shape=verify_shape))
File "/opt/conda/lib/python3.6/site-packages/tensorflow/python/framework/tensor_util.py", line 441, in make_tensor_proto
_GetDenseDimensions(values)))
ValueError: Argument must be a dense tensor: range(0, 1) - got shape [1], but wanted [].

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "retrotemp/nntrain_fingerprint.py", line 90, in
mol_hiddens = tf.nn.relu(linearND(input_mol, hidden_size, scope="encoder0"))
File "/home/jupyter/notebooks/utils/nn.py", line 22, in linearND
X_shape = tf.gather(tf.shape(input_), range(ndim-1))
File "/opt/conda/lib/python3.6/site-packages/tensorflow/python/ops/array_ops.py", line 2736, in gather
return gen_array_ops.gather_v2(params, indices, axis, name=name)
File "/opt/conda/lib/python3.6/site-packages/tensorflow/python/ops/gen_array_ops.py", line 3065, in gather_v2
"GatherV2", params=params, indices=indices, axis=axis, name=name)
File "/opt/conda/lib/python3.6/site-packages/tensorflow/python/framework/op_def_library.py", line 528, in _apply_op_helper
(input_name, err))
ValueError: Tried to convert 'indices' to a tensor and failed. Error: Argument must be a dense tensor: range(0, 1) - got shape [1], but wanted []