search

daniel-muthukrishna / astrorapid

Real-time Automated Photometric IDentification (RAPID) of astronomical transients using deep learning
https://astrorapid.readthedocs.io
MIT License
15 stars 14 forks source link

Installation instructions incomplete #12

Open klukosiute opened 3 years ago

klukosiute commented 3 years ago

I ran into several issues when installing astrorapid from source, none of which were accounted for in the installation instructions.

There was a dependency that I needed to install (just for package import) that was not included in the documentation, namely keras-tcn.

I was also getting version matching errors (from setup.py of astrorapid, but not from conda when building the environment) between numpy and tensorflow. I'm not sure what's happening there but I won't investigate further since I got it to work for myself.

I suggest adding a requirements.txt. Since I use conda, the best I can offer is the following for environment.yml:

name: rapid
channels:
  - conda-forge
  - defaults
dependencies:
  - python==3.7
  - numpy==1.19
  - tensorflow
  - keras
  - astropy
  - pandas
  - extinction
  - scikit-learn
  - matplotlib
  - scipy
  - pip:
    - keras-tcn==3.3.1
LydiaMak commented 3 years ago

I am also getting the "error" with keras-tcn. I am also having issues with matching versions. @klukosiute could you give a bit more details on how you solved it?