search

maxhodak / keras-molecules

Autoencoder network for learning a continuous representation of molecular structures.
MIT License
519 stars 146 forks source link

Encoding representations #64

Open hkmztrk opened 7 years ago

hkmztrk commented 7 years ago

Hello,

I'm pretty new to autoencoders and I know we can use utilize them for unsupervised learning. Is it possible to use this model to create representations (with encoding) for a set of SMILES?

If so, I guess first I had to preprocess my data set, then usesample.py ?

Thanks!

pechersky commented 7 years ago

That's exactly right. Have you taken a look at the Gomez-Bombarelli paper linked in the README?

On Wed, May 17, 2017 at 6:54 AM, hkmztrk notifications@github.com wrote:

Hello,

I'm pretty new to autoencoders and I know we can use utilize them for unsupervised learning. Is it possible to use this model to create representations (with encoding) for a set of SMILES?

If so, I guess first I had to preprocess my data set, then usesample.py ?

Thanks!

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/maxhodak/keras-molecules/issues/64, or mute the thread https://github.com/notifications/unsubscribe-auth/AFGDhslq3v_TLfGwwvU9PtNIjZdSQbkZks5r6tHTgaJpZM4NdshN .

hkmztrk commented 7 years ago

Yes, it was that paper led me to here. Thanks!

liamnaka commented 7 years ago

Last time I checked the VAE encodings were pretty substantial in size, so if you're trying to learn from a large set of SMILES it might be more feasible to generate them on the fly.

hkmztrk commented 7 years ago

Hello, thanks for your suggestion. Sorry for my asking but, how do we generate them on the fly? Aren't we supposed to learn the model first? How do we do that?

pechersky commented 7 years ago

Those reasons are why we developed sample_gen.py and train_gen.py, which generate the one-hot representations on the fly and can train the model using the fit_generator functionality of Keras.

To train the model in such a way, run python train_gen.py structures.h5 model.h5, where structures.h5 is an h5 file containing structures under the 'structure' key. Check out the source for train_gen.py for more details.

On Thu, May 25, 2017 at 3:01 AM, hkmztrk notifications@github.com wrote:

Hello, thanks for your suggestion. Sorry for my asking but, how do we generate them on the fly? Aren't we supposed to learn the model first? How do we do that?

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/maxhodak/keras-molecules/issues/64#issuecomment-303939205, or mute the thread https://github.com/notifications/unsubscribe-auth/AFGDhp7WKoES35uWL3zu54DUfSzIf7Upks5r9SdagaJpZM4NdshN .