Make ray-core an optional dependency in meta.yaml and provide details…

… on this in docs

conda-recipe/meta.yaml

@@ -24,20 +24,21 @@ requirements:
     - deeptools >=3.5.0
     - lmfit >=1.0.1
     - matplotlib-base ==3.1.1
-    - modin[ray] # [linux]
     - modin  ==0.8.2
     - numpy >=1.19.5
     - pandas >=1.1.4
     - joblib >=1.0.0
     - pybedtools >=0.8.1
     - pybigwig >=0.3.17
-    - ray-core ==1.1.0 # [linux]
     - scikit-learn >=0.24.0
     - scipy >=1.6.0
     - swifter >=1.0.6
     - parallel
     - seaborn ==0.11.2
     - pysam ==0.16.0.1
+  run_constrained:
+    - ray-core ==1.1.0
+    - ray ==1.1.0
 
 test:
   commands:

docs/source/intro.rst

@@ -121,15 +121,22 @@ LIQUORICE in its own conda environment to keep dependencies clean:
 
 .. code-block:: bash
 
-    # to install
-    conda create -e LIQUORICE -c bioconda LIQUORICE
+    # to install on Linux
+    conda create -e LIQUORICE -c bioconda -c conda-forge LIQUORICE ray-core
+
+    # to install on macOS
+    # conda create -e LIQUORICE -c bioconda -c conda-forge LIQUORICE ray-core
 
     # to activate the environment
     conda activate LIQUORICE # or: 'source activate LIQUORICE' for older conda versions
 
     # to run
     LIQUORICE <flags and arguments>
 
+'ray-core' is an optional dependency of LIQUORICE that helps it run faster on multiple cores. It is not possible to
+install it via conda on macOS, so installing it can either be skipped (find details about an alternative parallelization
+approach :ref:`here <parallelization>`) or it can be installed as described `here <https://docs.ray.io/en/latest/installation.html>`
+(e.g. `pip install -U ray==1.1.0`). LIQUORICE has been tested for ray version 1.1.0.
 
 Citation
 ========

docs/source/liquorice_commandline_tool.rst

@@ -165,6 +165,7 @@ If `\-\-all_bins_same_size` is specified, `LIQUORICE` does the following:
 Also here, we have decided to set the former option as default because it yielded somewhat better results for our own samples.
 We do note, however, that differences in bin-size might introduce some slight biases in the coverage profile. While we have nevertheless found that this option works well for us, we do encourage you to try both options for your own cohort.
 
+.. _parallelization:
 
 Parallelization
 ***************

liquorice/LIQUORICE.py

@@ -302,17 +302,17 @@ def main():
                                 f" used for parallel processing ({len_path_to_tmpdir_in_bytes} bytes observed, "
                                 f"{maxlen_tmp_dir_path_in_bytes} bytes allowed). "
                                 f"Will attempt to use ray's default, '/tmp', instead.")
-                ray.init(num_cpus=args.n_cpus)
+                ray.init(num_cpus=args.n_cpus,include_dashboard=False)
             else:
-                ray.init(num_cpus=args.n_cpus,_temp_dir=args.tmpdir)
+                ray.init(num_cpus=args.n_cpus,_temp_dir=args.tmpdir,include_dashboard=False)
         else:
             logging.warning(f"Failed to initialize ray with the specified tmpdir '{args.tmpdir}' and "
                             f"num_cpus {args.n_cpus}."
                             f" This happens if the ray version is lower than 1.1.0. Setup will be done by modin "
                             f"instead, but this can mean the tmpdir will not be respected.")
         n_cpus_workflows=args.n_cpus
     except ModuleNotFoundError:
-        os.environ["MODIN_ENGINE"] = "dask"
+        #os.environ["MODIN_ENGINE"] = "dask"
         if args.n_cpus>1:
             logging.warning("Could not import ray library - this happens if running on macOS. Parallelization is "
                             "therefore only "