Maria Korshunova

@ylp812 @gmseabra I downloaded JAK2 dataset from CHEMBL and uploaded it into data folder. Now everything should work

Hi @Donald954732 Looks like you are using pytorch 0.4. The error you got was due to updates in pytorch between versions 0.3 and 0.4. I updated the code, now everything...

Hi @lorybaby This is an ongoing project, so I'm constantly working on it and adding new features. You can find the demo for training the LSTM model in RecurrentQSAR-example-jak2.ipynb and...

Hi @toushi68, Thanks for pointing this out. I've made some update to modules recently and will upload an updated version of demos in the nearest future (today/tomorrow).

Hi @vmarar I'd say this is expected behavior. Further in the code `rl_loss` is backpropagated by applying `rl_loss.backward()`. But your changes detach `rl_loss` from the computational graph and there is...

> would this be the case using rl_loss -= (log_probs[0, top_i].item()*discounted_reward) as well? Would there still be a detachment? Yes, this is exactly what `.item()` does. It returns just the...

> I tried converting from numpy to tensor and then pointing it towards cuda:0, no error but model weights are still not changing Yes, that wouldn't help, because you are...

I uploaded file with jak2 dataset from CHEMBL. And I will also update RecurrentQSAR-example.ipynb to a newer and better version soon.

Hi @gmseabra I uploaded updated examples for QSAR with RNNs. Those are `RecurrentQSAR-example-jak2.ipynb` and `RecurrentQSAR-example-logP.ipynb` notebooks. Models are built with our new toolkit https://github.com/Mariewelt/OpenChem . It's the same architectures as...

Hi @hello1910 This file is created in the ipython-notebook automatically.