# Conda environment for GATK Python Tools
#
# Only update this environment if there is a *VERY* good reason to do so!
# If the build is broken but could be fixed by doing something else, then do that thing instead.
# Ensuring the correct environment for canonical (or otherwise reasonable) usage of our standard Docker takes precedence over edge cases.
# If you break the environment, you are responsible for fixing it and also owe the last developer who left this in a reasonable state a beverage of their choice.
# (This may be yourself, and you'll appreciate that beverage while you tinker with dependencies!)
#
# When changing dependencies or versions in this file, check to see if the "supportedPythonPackages" DataProvider
# used by the testGATKPythonEnvironmentPackagePresent test in PythonEnvironmentIntegrationTest needs to be updated
# to reflect the changes.
#
name: gatk
channels:
# if channels other than conda-forge are added and the channel order is changed (note that conda channel_priority is currently set to flexible),
# verify that key dependencies are installed from the correct channel and compiled against MKL
- conda-forge
- anaconda
- defaults
dependencies:

# core python dependencies
- conda-forge::python=3.6.10        # do not update
- pip=20.0.2                        # specifying channel may cause a warning to be emitted by conda
- conda-forge::mkl=2019.5           # MKL typically provides dramatic performance increases for theano, tensorflow, and other key dependencies
- conda-forge::mkl-service=2.3.0
- conda-forge::joblib=1.1.1         # must pin joblib - versions after 1.1.1 no longer support python 3.6
- conda-forge::numpy=1.17.5         # do not update, this will break scipy=1.0.0
                                    #   verify that numpy is compiled against MKL (e.g., by checking *_mkl_info using numpy.show_config())
                                    #   and that it is used in tensorflow, theano, and other key dependencies
- conda-forge::theano=1.0.4         # it is unlikely that new versions of theano will be released
                                    #   verify that this is using numpy compiled against MKL (e.g., by the presence of -lmkl_rt in theano.config.blas.ldflags)
- anaconda::tensorflow=1.15.0       # update only if absolutely necessary, as this may cause conflicts with other core dependencies
                                    #   verify that this is using numpy compiled against MKL (e.g., by checking tensorflow.pywrap_tensorflow.IsMklEnabled())
- conda-forge::scipy=1.0.0          # do not update, this will break a scipy.misc.logsumexp import (deprecated in scipy=1.0.0) in pymc3=3.1
- conda-forge::pymc3=3.1            # do not update, this will break gcnvkernel
- conda-forge::h5py=2.10.0          # required by keras 2.2.4
- conda-forge::keras=2.2.4          # updated from pip-installed 2.2.0, which caused various conflicts/clobbers of conda-installed packages
                                    #   conda-installed 2.2.4 appears to be the most recent version with a consistent API and without conflicts/clobbers
                                    #   if you wish to update, note that versions of conda-forge::keras after 2.2.5
                                    #   undesirably set the environment variable KERAS_BACKEND = theano by default
- anaconda::intel-openmp=2019.0
- conda-forge::scikit-learn=0.23.1
- conda-forge::matplotlib=3.2.1
- conda-forge::pandas=1.0.3
- conda-forge::typing_extensions=4.1.1   # see https://github.com/broadinstitute/gatk/issues/7800 and linked PRs
- conda-forge::dill=0.3.4                # used for pickling lambdas in TrainVariantAnnotationsModel

# core R dependencies; these should only be used for plotting and do not take precedence over core python dependencies!
- r-base=3.6.2
- r-data.table=1.12.8
- r-dplyr=0.8.5
- r-getopt=1.20.3
- r-ggplot2=3.3.0
- r-gplots=3.0.3
- r-gsalib=2.1
- r-optparse=1.6.4
- r-backports=1.1.10

# other python dependencies; these should be removed after functionality is moved into Java code
- biopython=1.76
- pyvcf=0.6.8
- bioconda::pysam=0.15.3            # using older conda-installed versions may result in libcrypto / openssl bugs

# pip installs should be avoided, as pip may not respect the dependencies found by the conda solver
- pip:
  - gatkPythonPackageArchive.zip