Missing dependency?

[agitter]

When I worked through the tutorial on my machine, I received an error because the shapely package was not available. Should this be added as a required package in setup.py?

Everything else was great!

[akundaje]

Thanks for the feedback Anthony. Will look into it.

Anshul On Jul 1, 2016 4:41 PM, "Anthony Gitter" notifications@github.com wrote:

When I worked through the tutorial on my machine, I received an error because the shapely package was not available. Should this be added as a required package in setup.py?

Everything else was great!

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/kundajelab/dragonn/issues/2, or mute the thread https://github.com/notifications/unsubscribe/AAI7EUJE7iVZ2UUUuqauxNIHqArB7Kuuks5qRXtbgaJpZM4JDd0A .

[jisraeli]

@agitter: Thanks for the feedback. shapely has been added to setup.py, can you try the tutorial again?

[agitter]

Thanks. That fixed the shapely dependency during installation.

Does the Keras version need to be exactly 0.3.2? I was testing with 1.0.4 and could run the tutorial without problems. When I was trying the installer now, I accidentally downgraded the Keras version from 1.0.4 to 0.3.2 because a specific version is specified.

I also saw the following if I didn't install DeepLIFT before DragoNN:

Collecting deeplift (from dragonn==0.1)
  Could not find a version that satisfies the requirement deeplift (from dragonn==0.1) (from versions: )
No matching distribution found for deeplift (from dragonn==0.1)

This is all when installing directly from the source so anyone who uses conda install -c kundajelab dragonn as suggested may not encounter these problems anyway.

[jisraeli]

@agitter: DeepLIFT is set up to work with the keras versions 0.3.x, we haven't tested extensively for sequential models in versions >= 1.0, we could test those soon. Regarding the DeepLIFT installation issue, the installation from source hasn't been tested as extensively as the conda package installation. I will look into this tomorrow, it seems as if the dependency link isn't working.

[agitter]

Thanks for the additional guidance. Because you already provide the conda package, there's no need to spend time working on the installation from source on my behalf. You can close the issue if you would like.

[jisraeli]

@agitter: Let's try to sort this out, on behalf of the Kundaje lab. I just tried installing on a Linux machine without the DragoNN dependencies and wasn't able to replicate this error. Are you installing on a Linux machine?

[agitter]

This was a Linux machine. I'll try a fresh install from source again to see whether I can replicate the error myself.

[agitter]

I was able to troubleshoot my problems. I tested installing DragoNN from source using Travis CI to avoid any problems that are specific to my operating system or Anaconda installation. I first found that I needed to add --process-dependency-links to pip install to process the DeepLIFT dependency correctly. This was the error I was seeing originally: https://travis-ci.org/agitter/dragonn/builds/142800973

Then I had to update setup.py to include two other required packages. That was sufficient to install DragoNN with Travis CI.

If you want to include this automated build testing, I created a pull request #3. If you don't want it, you can just close the pull request and copy the minor setup.py change.

[jisraeli]

@agitter: Great, I'll merge the pull request.