Open rbharath opened 7 years ago
Agreed, it doesn't look like any of their code is open source though. Would a Keras implementation be the desired way to go? Ref: https://github.com/fchollet/keras/blob/master/examples/variational_autoencoder.py
We dropped Keras support a while ago due the rapidly changing API which broke the build a few times. Has the Keras API stabilized now?
It might also be straightforward to do this in TensorGraph (cc @lilleswing)
The main network topology is described in this paper -- https://papers.nips.cc/paper/6581-improved-variational-inference-with-inverse-autoregressive-flow.pdf
I had trouble following their semi-supervised learning scheme so I don't know if I can comment on how easy it would be to port.
@LRParser the code you linked is already ported into deepchem here Gene Expressions are expressed in a bitvector really easily -- but small molecules are not. This is the main challenge for creating a molecular autoencoder.
https://github.com/deepchem/deepchem/tree/master/deepchem/models/autoencoder_models
Their semi-supervised learning scheme works because they can train both x_1 and x_2 over the encoder without training the gene expression changes as shown here.
This should be doable to port to Tensorgraph. Getting our hands on the datasets to prove that it worked might be the trickiest part.
I'd love to get some gene expression data in DeepChem. There are some very impressive results here, notably:
http://stm.sciencemag.org/content/3/96/96ra77
If we could get that implemented in DeepChem, would be amazing!
Ok, this one is over my head for the moment, but I'll do my best to follow along and learn Tensorgraph.
This would be a good intermediate level contribution, especially for those with prior deep learning experience.
https://arxiv.org/abs/1706.08203
Looks interesting. We should add an implementation to DeepChem