Update cryocare environment to work on GPU

Programming/anaconda/2019.07/conda-env-defs/cryocare/README.md

@@ -13,8 +13,10 @@ Now configure jupyter
 
 ## Package notes
 
-- cudatoolkit version is fixed because it needs to match merlin6's CUDA module
-  (required by the MotionCor2 package).
-- This environment installs packages for gpu support, but these currently don't
-  seem to be working on merlin.
+- cudatoolkit version should ideally match merlin6's CUDA module
+  (required by the MotionCor2 package). However, tensorflow needs cuda 10.0 while
+  no MotionCor2 binary is available. For now, having a different version in
+  conda and modules is working OK, but it could cause problems with future compilation.
+- You can test GPU support by running this command on a GPU machine:
+  `python -c 'import tensorflow as tf; print(tf.test.is_gpu_available())'`
 

Programming/anaconda/2019.07/conda-env-defs/cryocare/cryocare.yml

@@ -4,30 +4,23 @@ channels:
   - defaults
 dependencies:
   - python=3.7
+  # Explicit cryocare dependencies from singularity
+  - tensorflow-gpu=1.15.0
+  - keras
+  - tifffile
+  - mrcfile
+
+  # Other dev packages
+  - jupyter
   - dask
   - dask-jobqueue
-  - deap
-  - docopt
-  - graphviz
-  - ipython
-  - ipywidgets
-  - keras
-  - matplotlib
-  - mrcfile
-  - nb_conda_kernels
-  - numpy
   - pandas
-  - plotly
-  - pytables
-  - python-graphviz
   - scikit-learn
   - scipy
   - seaborn
-  - tensorflow=1.13.1
-  - tensorflow-gpu
-  - tifffile
-  - cudatoolkit=9.2
-  - cudnn
+  - plotly
+  - pytables
+  - pip
   - pip:
     - csbdeep
     - pixiedust