setup.md

title
Setup

Setup Keras kernel in Palmetto's JupyterLab

We will use Palmetto cluster for this workshop with Jupyter Lab.

Please follow this guideline to create a new conda environment and install scikit-learn package. Log into Clemson's OpenOnDemand

Under Clusters select Palmetto Shell Access

Log into the Palmetto Command Line Shell

Run the following qsub command

$ qsub -I -l select=1:ncpus=8:mem=15gb:chip_type=e5-2680v4,walltime=2:00:00

• Depending on cluster availability, you can set chip_type to be e5-2680v4, 6148g, and 6248g.
• You can use whatsfree and freeres to identify an appropriate chip_type setting.

Next, run the following commands.

$ module load anaconda3/2021.05-gcc/8.3.1 cuda/11.1.0-gcc/8.4.1 cudnn/8.1.0.77-11.2-linux-x64-gcc/8.4.1
$ conda create -n tf_2.5 python=3.8 -y
$ source activate tf_2.5
$ export PYTHONNOUSERSITE=1
$ pip install tensorflow==2.5 seaborn scikit-learn matplotlib jupyterlab

=> Note: while using skln conda environment, if we are missing anything, we can always come back and update using pip install or conda install method.

Go back to OpenOnDemand Dashboard, under Interactive Apps select Jupyter Notebook

Make the selection on the Jupyter Notebook App as follows:

• Anaconda Version: anaconda3/2021.05-gcc/8.3.1
• List of modules to be loaded, separate by an empty space: cuda/11.1.0-gcc/8.4.1 cudnn/8.1.0.77-11.2-linux-x64-gcc/8.4.1
• Path to Python virtual/conda environment: source activate tf_2.5
• Notebook Workflow: Standard Jupyter Notebook
• Number of resource chunks (select): 1
• CPU cores per chunk (ncpus): 8
• Amount of memory per chunk (mem): 15gb
• Interconnect: any
• Extra PBS resource allocation request: :chip_type=e5-2680v4
• Walltime: 04:00:00

Click Launch.

Click Connect to Jupyter once the job is ready.

Open a new notebook using the default Python 3 kernel. Test for the valid installation of tensorflow.

{% include links.md %}