Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Activity
    Michael He
    @m-whe
    I'm still getting the same error with the nightly build.
    alat-rights
    @alat-rights
    @m-whe are you sure you imported Deepchem? Have you been thru the DC tutorials?
    also, whats ur environment?
    Bharath Ramsundar
    @rbharath
    @m-whe Hmm, it sounds like the DeepChem install is failing on your local environment. Could you try the DeepChem colab tutorials? It might be easier to run on Colab if you're just doing some experimentation
    Sumudu Leelananda
    @sumuduleelananda
    Hi all, I built a deepchem classification model. However, I can't save the model I built. Can somebody please help? model = dc.models.GraphConvTensorGraph(len(tasks), batch_size=50, mode='classification')
    model.save() (doesn't work)
    Bharath Ramsundar
    @rbharath
    @sumuduleelananda This model should autosave so no need to call model.save(). Just make sure you set model_dir in the constructor to be the directory you want the model to be saved
    Sumudu Leelananda
    @sumuduleelananda
    model = dc.models.GraphConvTensorGraph(len(tasks), batch_size=50, mode='classification', model_dir="/Users/deepchem_predictions/")
    like this?
    Bharath Ramsundar
    @rbharath
    Yes, that should do it if /Users/deepchem_predictions is where you'd like to save. Models will autowrite checkpoints when you call model.fit()
    Sumudu Leelananda
    @sumuduleelananda
    ah i see, after i specified the directory and did the model.fit it saved some models. I'll look into these models. Thank you very much!
    Bharath Ramsundar
    @rbharath
    Glad to help!
    Michael He
    @m-whe
    Tutorials look great! Any plans to make them into an online course?
    alat-rights
    @alat-rights
    So, actually, kinda! There has been talks in one of the meetings about partnering with O’Reily to make a course sequence as a companion to the Deepchem book. It’s still far off though. @m-whe
    alat-rights
    @alat-rights
    image.png
    Hey guys wondering if anyone here has experience with the Deepchem AWS infrastructure. I’m getting a 403 Forbidden when trying to fetch some files that I uploaded, and I don’t think I have permissions to allow public downloads on the files I uploaded even though the deepchemdata bucket is public? Is there a fix?
    Sahar Rohanizeidanlou
    @SaharRohaniZ

    Hi everyone - I am trying to load PDBbind dataset using this code:

    pdbbind_tasks, pdbbind_datasets, transformers = load_pdbbind(featurizer=RdkitGridFeaturizer(flatten=False), splitter="random", subset="refined")
    train_dataset, valid_dataset, test_dataset = pdbbind_datasets

    But I am getting below error:

    Failed to featurize datapoint 4777. Appending empty array

    I am new to Deepchem and I can't figure out the problem. I appreciate if you could help me on this. thanks so much in advance.

    Bharath Ramsundar
    @rbharath
    @alat-rights You should be able to set the permissions on the files. Try logging into the AWS console and checking each file specifically to set permissions
    @SaharRohaniZ Hmm, my apologies for the trouble. We're in the process of overhauling our structural featurization support and it looks like there might be issues. Which deepchem version are you on?
    Sahar Rohanizeidanlou
    @SaharRohaniZ
    Hi @rbharath THank you for your reply. I am using deepchem version 2.4.0
    Sahar Rohanizeidanlou
    @SaharRohaniZ
    @rbharath I am trying to replicate the work done on "Atomic Convolutional Networks for Predicting
    Protein-Ligand Binding Affinity". Is there a stable version that I can use to replicate the work ? thanks
    Bharath Ramsundar
    @rbharath
    @SaharRohaniZ If you check the contrib/ folder there is an old version of the code, but this depends on old versions of DeepChem / TF. We're working to create an updated version of the Atomic convs but this is a work in progress
    Sahar Rohanizeidanlou
    @SaharRohaniZ
    Thanks @rbharath . I will check that.
    alat-rights
    @alat-rights
    I feel so silly!!! Thanks for pointing that out @rbharath
    Bharath Ramsundar
    @rbharath
    @alat-rights Glad it seems to have worked :)
    alat-rights
    @alat-rights
    :smile:
    Zhaolihui
    @Blockbastermaker
    I have problem with import,eg:
    from deepchem.models.tensorgraph.layers import GraphPool, GraphGather
    from deepchem.models.tensorgraph.layers import Dense, L2Loss, WeightedError, Stack
    from deepchem.models.tensorgraph.layers import Label, Weights
    thanks for checking!
    myjenniferr22
    @myjenniferr22
    Hi guys, I met some difficulties when I try to use deepchem to build a convolutional mode.
    When I try to use
    model = dc.models.TensorGraph(batch_size=1000), it says "module 'deepchem.models' has no attribute 'TensorGraph'". Does anyone know why it happened? thank you very much!!!!
    myjenniferr22
    @myjenniferr22

    I have problem with import,eg:

    I got the same issue here. Plz let me know if you find the solution :D Thank you!!!

    Bharath Ramsundar
    @rbharath
    @Blockbastermaker Are you on DeepChem 2.4.0? We removed the tensorgraph layers since we moved everything over to keras. You should be able to do all the imports as before but just removing the tensorgraph:
    from deepchem.models.layers import GraphPool, GraphGather
    @myjenniferr22 Sounds like you're on DeepChem 2.4.0. You'll want to just use dc.models.GraphConvModel for example. Check out our tutorial series which has been updated to use the new style https://github.com/deepchem/deepchem/tree/master/examples/tutorials
    In general, everything will work the same as before (same API and everything), but just with the tensorgraph removed from he imports
    Zhaolihui
    @Blockbastermaker
    @rbharath I was on DeepChem=2.4, Thank you very much!!!
    myjenniferr22
    @myjenniferr22
    @rbharath Yes! I am following the code wrote on a book called "Deep Learning for the life sciences". Since DeepChem had been updated, I am always getting confused because they are using the old version..... Thank you very much for your replay!!!
    myjenniferr22
    @myjenniferr22
    Does anyone also follow along with this book??? Plz let me know!!
    myjenniferr22
    @myjenniferr22
    And also if anyone is reading chapter 6, could you please let me know how you run the code? Everything seems not working because of the version updates. Thank you everyone :D
    myjenniferr22
    @myjenniferr22
    In a example that shows the convolutional modeling of TF binding, the code from the book uses "layers.Feature", "layers.Label", "layers.Weights", "layers.Conv1D"........ There are lots of stuff that have already been changed....
    Sahar Rohanizeidanlou
    @SaharRohaniZ
    Hi @rbharath . I have looked at the Deepchem/Contrib content and right now I am trying to replicate the work done on "Atomic Convolutional Networks for Predicting
    Protein-Ligand Binding Affinity" paper. I have tried running the code but I am getting lots of errors that are mostly versioning errors. Is there any documentation on the deepchem versioning, what has been changed or any information on version compatibility for Deepchem and Tensorflow? thanks
    Bharath Ramsundar
    @rbharath
    @myjenniferr22 Thanks for reporting! I've started an issue to track the needed changes deepchem/DeepLearningLifeSciences#26. Could you report the issues you're seeing there?
    We will try to update the online version of the code to run in the latest DeepChem shortly
    @SaharRohaniZ Unfortunately this paper was written several years ago (in something like DeepChem 0.5 or 1.0) and our versioning/documentation system wasn't that great back then. This is definitely on my radar to clean up and get in a better documented state, but it's a big project. We're making some progress but it might be a while
    Bharath Ramsundar
    @rbharath
    @myjenniferr22 As an add on, most of the change should be pretty superficial. You should be able to mostly get by just deleting "tensorgraph" everywhere and everything else should be the same
    Sahar Rohanizeidanlou
    @SaharRohaniZ
    @rbharath I understand. Do you know when the featurization model going to work in the current version of deepchem?
    Bharath Ramsundar
    @rbharath
    @SaharRohaniZ Do you mean the AtomicConvModel? I will try to get that fixed by DeepChem 2.5.0 which releases at the end of February
    Sahar Rohanizeidanlou
    @SaharRohaniZ
    @rbharath I mean the featurizer.

    I tried to run below code and got an error and you mentioned that you are in the process of overhauling the structural featurization support. I just wanna know when do you think the featurization model for the current version of deepchem going to be available?

    pdbbind_tasks, pdbbind_datasets, transformers = load_pdbbind(featurizer=RdkitGridFeaturizer(flatten=False), splitter="random", subset="refined")
    train_dataset, valid_dataset, test_dataset = pdbbind_datasets

    I am getting below error:

    Failed to featurize datapoint 4777. Appending empty array