Yakov Pechersky

You're generating noise using uniform sampling between 0 and 1 at every position in the vector. Our true data is some weird subset of the space of all possible one-hot...

Nice! Do you think the simple strings are due to just insufficient training, or its converging to a simple part of the string space? Perhaps including the KL divergence in...

According to https://github.com/fchollet/keras/compare/1.1.0...1.1.1, `ReduceLRonPlateau` is in `1.1.1`.

Version `1.1.1` works for me. If it works for you as well, we'll merge in.

I've also thought a lot about this and better model training. I'd love to hear your thoughts on what's possible and what's worth prioritizing. A SMILES string is approximately a...

What's the downside of increasing the "charset" size? My inclination comes from the fact that we feed the SMILES into a conv layer, so any increase in string size disrupts...

Thanks for coding up the rdkit-networkx code. The first experiment to run, in my opinion, would be to provide an adjacency matrix of (maxlen, maxlen), with 1s for edge between...

Regarding encoding reactions, check out another Duvenaud paper: http://pubs.acs.org/doi/pdf/10.1021/. On Nov 13, 2016 12:50 PM, "dakoner" [email protected] wrote: > Going back to the cellular encoding, one obvious direction for the...

https://dx.doi.org/10.1021/acscentsci.6b00219

Regarding synthetic accessibility, check out these two papers (companions): http://onlinelibrary.wiley.com/doi/10.1002/anie.201202209/full http://onlinelibrary.wiley.com/doi/10.1002/ange.201202210/full They try to enumerate all possible synthetic routes and how to optimize for shortest on cost, time, or availability...