Wengong Jin
Wengong Jin
This is very surprising. Which tensorflow version are you using? I know there's discrepancy between tf.0.12.0 and tf 1.0+. We only tested our code on v0.12.0.
Hi, Suppose you downloaded our code at $BASEDIR/graph_knn, please run the following command to add all modules in the pythonpath. ``` export PYTHONPATH=$BASEDIR/graph_knn ``` Then python will not complain about...
Hi, I apologize for my late reply. I tested my code on ubuntu 14.04 / python 2.7 / cuda 8.0 / pytorch 0.3.1 / rdkit 2017_09 with NVIDIA TITAN X...
Oh I see. Uh... There is no particular ordering I think... The order doesn't change the result and I don't remember why there are in this order...
Hi, You can try three things: 1. reduce batch size 2. reduce hidden size from 450 -> 300, for instance. 2. In chemutils.py, change `MAX_NCAND` to smaller value. Please also...
Hi, I didn't find the line aenzynes in fast_jtnn/chemutils.py. Could you pull the latest version from github? Wengong
Currently not. In terms of conditional optimization, we developed a new method called graph-to-graph translation that performs much better than running gradient ascent over VAE latent space. The paper and...
Hi, Thanks for pointing out this issue. It is NOT on purpose (or used as a trick). For MolTreeNode class, I really meant to set default value of `clique=[]`.
Hi, I did some quick check: For `MolTreeNode()` constructor, I tested the line `super_root = MolTreeNode("")` by printing `super_root.clique`, it always prints `[]` as desired. For `enum_assemble()` function, I tested...
I don't think this problem will cause memory issues.