Python Environment

Supercomputer

In general, you do not need to install your own copy of python/tensorflow/keras. Instead, you can reference mine.

Execute:
```
. /home/fagg/tf_setup.sh
```
Alternatively, you can add the following line to the end of your .bashrc file:
```
. /home/fagg/tf_setup.sh
```
This will automatically set your conda paths up every time you login to the supercomputer.
Activate the environment:
```
conda activate dnn
```
Thereafter, starting python and loading packages will be done in the context of this environment.
Of course, you will not be able to install additional packages into my space (if you want something, please ask).

Personal Installation

These instructions are for those wanting an up-to-date version of the python environment on their own machine. You want to be using Tensorflow 2.16 or better for this class.

Installation through Miniconda by Hand

Download and install Miniconda
- It will prompt you as to whether you should init the environment. Answer yes
- I prefer not to automatically activate the base environment when you first create a shell:
```
conda config --set auto_activate_base false
```

Then do the following (the choice of "tf" is arbitrary):

conda update -n base -c defaults conda
conda create --name tf
conda activate tf
conda install mamba -c conda-forge
mamba install python=3.12 -c conda-forge

# There will be some package downgrades for now to support tensorflow-probability       
mamba install tensorflow tensorflow-probability -c conda-forge
       OR
mamba install tensorflow-gpu tensorflow-probability -c conda-forge

mamba install scikit-learn pypng matplotlib pandas bitarray jupyterlab pydot scikit-image keras-tuner xarray seaborn -c conda-forge
mamba install scikit-learn-intelex -c conda-forge
mamba install netcdf4 geopy opencv wandb jupyter -c conda-forge
mamba install plotly openpyxl -c conda-forge


       EXTRAS:
       
mamba install pytorch-gpu torchvision -c conda-forge

Once completed, when you start a new shell, you can activate your environment using:
```
conda activate tf       
```

Jupyter Lab

Jupyter Lab is an interactive development environment that you should only be using on your local machine (never on the supercomputer!). It is useful for quickly testing out ideas and code before you execute experiments.

If you followed the instructions above, then you already have Jupyter Lab installed. Otherwise ... Jupyter Lab installation

Useful software to support jupyter:

pandoc (need executable and python package)
xelatex (executable)

andrewhfagg AT gmail.com

Last modified: Mon Jan 13 17:08:56 2025