pip install climetlabSee the
installing instructions <installing>
for more details.
There are two ways to access data using CliMetLab:
Using a Dataset </guide/datasets>
: CliMetLab provides a few demo datasets. In order to access other datasets with :pycml.load_dataset
, the relevant plugin must be installed.Using a data Source </guide/sources>
: A data Source allows loading various kinds of data format and location through :pycml.load_source
. Data sources should be used when there is no dataset plugin for the data you are interested in.
Creating a CliMetLab plugin can be a solution to share some code along with the dataset that you are publishing/using. See the
plugin documentation <contributing-overview>
.
See
/guide/caching
.
It is not recommended to share your cache with others. What you are looking for may be a mirror. This feature is not implemented yet.
from climetlab.utils.dask import start start('local') # or $ climetlab dask start local # or $ climetlab dask local --start
from climetlab.utils.dask import start start('ssh')
from climetlab.utils.dask import start start('slurm')
Note
This is assumes that your HPC admin set up the hpc-name-config-1.yaml file on the appropriate location.
from climetlab.utils.dask import start start('hpc-name-config-1')
todo
todo
The dask cluster and client will usually stop automatically when the python process ends. Nevertheless, it is possible to stop dask if it has been started from climetlab.
from climetlab.utils.dask import stop stop()
Note: In this section a "dask deployement" refers to a client and a cluster. It does not refers to a Cloud deployement using Kubernetes, etc.
Create the yaml file $HOME/.climetlab/dask/hpc-name-config-1.yaml. Then use it with: from climetlab.utils.dask import start start('hpc-name-config-1')
Note
For HPC system admin: Adding yaml files in /opt/climetlab/dask/*.yaml will give global access to all users.
from climetlab.utils.dask import start client = start('local').client
from climetlab.utils.dask import start deploy = start('slurm') deploy.scale(..)