Installation Problems

#1

Hello!

I am trying to learn DeepChem by completing the tutorial. I am trying to follow it in a Jupyter notebook.

After executing the following command in a cell:
conda install -c conda-forge rdkit deepchem==2.6.1

My installation appears to be stalling at:

> Collecting package metadata (current_repodata.json): done
> Solving environment: failed with initial frozen solve. Retrying with flexible solve.
> Collecting package metadata (repodata.json): done
> Solving environment:

Any advice on how to proceed? Thanks in advance!

#2

Can you try pip install deepchem? Alternatively, the latest version of deepchem is 2.7.1. Can you try the conda installation command with the latest version? That may avoid the resolver issue

#3

So I originally was trying the pip install deepchem command, and it appeared to work and I get a message saying requirements already satisfied. However when I would try import deepchem as dc then the kernel would die I would be stuck there.

When searching online, I found a github thread suggesting it was a problem with the installation and to try using the conda command instead. I will try using the most up to date version and report back.

Also I left my previous installation using the command in the original post and it has actually proceeded some. It is now at:

conda install -c conda-forge rdkit deepchem==2.6.1
Collecting package metadata (current_repodata.json): done
Solving environment: failed with initial frozen solve. Retrying with flexible solve.
Collecting package metadata (repodata.json): done
Solving environment: failed with initial frozen solve. Retrying with flexible solve.
Solving environment: |
Found conflicts! Looking for incompatible packages.
This can take several minutes. Press CTRL-C to abort.
Examining conflict for nss itemadapter scikit-learn s3transfer filelock nbclas-/

If this helps at all. Thanks again!

#4

Does conda install -c conda-forge rdkit deepchem==2.7.1 work instead?

#5

Ok so I moved from a Jupyter notebook to my terminal, uninstalled deepchem, and ran the command conda install -c conda-forge rdkit deepchem==2.7.1. It is now checking for inconsistencies and attempting to solve the environment.

This is what it looks like so far:

Collecting package metadata (current_repodata.json): done
Solving environment: | 
The environment is inconsistent, please check the package plan carefully
The following packages are causing the inconsistency:

  - defaults/osx-64::scikit-learn==1.0.2=py39hae1ba45_1
  - defaults/osx-64::scikit-image==0.19.2=py39hae1ba45_0
  - defaults/osx-64::scikit-learn-intelex==2021.6.0=py39hecd8cb5_0
  - defaults/osx-64::datashader==0.14.1=py39hecd8cb5_0
  - defaults/osx-64::statsmodels==0.13.2=py39hca72f7f_0
  - defaults/osx-64::gensim==4.1.2=py39he9d5cce_0
  - defaults/noarch::seaborn==0.11.2=pyhd3eb1b0_0
  - defaults/osx-64::anaconda==custom=py39_1
  - defaults/osx-64::_anaconda_depends==2022.10=py39_2
failed with initial frozen solve. Retrying with flexible solve.
Solving environment: failed with repodata from current_repodata.json, will retry with next repodata source.
Collecting package metadata (repodata.json): done
Solving environment: / 
The environment is inconsistent, please check the package plan carefully
The following packages are causing the inconsistency:

  - defaults/osx-64::scikit-learn==1.0.2=py39hae1ba45_1
  - defaults/osx-64::scikit-image==0.19.2=py39hae1ba45_0
  - defaults/osx-64::scikit-learn-intelex==2021.6.0=py39hecd8cb5_0
  - defaults/osx-64::datashader==0.14.1=py39hecd8cb5_0
  - defaults/osx-64::statsmodels==0.13.2=py39hca72f7f_0
  - defaults/osx-64::gensim==4.1.2=py39he9d5cce_0
  - defaults/noarch::seaborn==0.11.2=pyhd3eb1b0_0
  - defaults/osx-64::anaconda==custom=py39_1
  - defaults/osx-64::_anaconda_depends==2022.10=py39_2
failed with initial frozen solve. Retrying with flexible solve.
Solving environment: | 
Found conflicts! Looking for incompatible packages.
This can take several minutes.  Press CTRL-C to abort.

I have been doing all of this in my base environment - do you recommend I create an environment specifically for working with deepchem?

Thanks.

#6

After running the suggested command:

conda install -c conda-forge rdkit deepchem==2.7.1

I am still stuck in a solving environment and examining conflict loop. Strange I was able to successfully install deepchem on my other mac, but on a newer mac I am stuck. Any further suggestions would be appreciated. Thanks.

#7

If you are just looking to play with deepchem, I’d recommend just using Colab and pip install deepchem. If you’d like a serious dev environment on your mac, can you try in a fresh conda environment? It’s possible it’s getting stuck because of existing packages. If you are on a M2, there may possibly be issues. I think there was another forum thread about M2 that may be worth checking out

#8

Just for completeness and in case another poor soul runs into this problem, this is how I solved the above issue.

I created a new environment with no default packages preloaded so I could circumvent the environment conflicts.

conda create --no-default-packages -n chemenv python tensorflow scipy scikit-learn pandas numpy joblib jupyter

Then I installed rdkit and depchem like so:

pip install rdkit deepchem

And now it works if I open the environment within my jupyter notebook! Thanks again for your help @bharath!

-b

1 Like
#9

Which version of python are you using with DeepChem? Per documentation only versions 3.6 or 3.7 are supported.

https://deepchem.readthedocs.io/en/latest/get_started/requirements.html

#10

The documentation is out of date. We support 3.8+ as the primary versions for the latest deepchem. Sorry for the confusion!

#11

Hi @bhumm do you not run into the problem with gilbrat.
Fixed here but not released:

Come to look at it again - that checkin was made by you @bharath…
How come I get blocked by it with a standard conda install like the one above?

#12

We recommend using the pip prerelease for now pip install --pre deepchem. It’s quite stable. We are pushing towards a major version release where we finish porting TensorFlow -> PyTorch so it will still be some time until our next conda release