Intro to Deep Learning [September 2022]

Slides and example scripts for running deep neural networks using Keras and TensorFlow v2 in Python and R will be uploaded the day of the class.

Please prepare by installing the necessary packages for Python or R below and use the "hello world" example script in the respective R and Python folders to confirm everything works. Note, this may require help from IT if you do not have admin rights on your machine. If you run into issues with your own machine, you should be able to install these envonments on either the NIAID Locus or NIH Biowulf HPCs by loading the Anaconda or R modules.

Installing the environment on your own machine

Python Users

I highly recommend installing packages using Anaconda or Miniconda on your machine. If you do not have it already, you may need IT support to get this install if you do not have admin rights on your machine. Note - some users appear to be having an issue where some conda repositories are being blocked on the NIH network. If you run into this, please try logging out of the VPN and installing again. If you are not on a laptop, you may need IT support to get around this or to download the packages and install them locally.

To run these you'll need Python (version 3.x) and the following major packages and their dependencies installed:

• tensorflow (version 2 now includes keras)
• scikit-learn
• Pillow
• scipy
• matplotlib
• keras-nlp [only for intermediate course]

I recommend installing the specific package versions listed in the deep_learning_environment.yml file by creating a conda environment. With conda installed, you can easily make an environment called tflow in a terminal window using the command:

conda env create -f deep_learning_environment.yml

You can then activate this environment via:

conda activate tflow

You can install to a specific directory with a custom name for the environment using:

conda env create --prefix ./envs -n myname -f deep_learning_environment.yml

Where, ./envs is the directory you want to install to and myname is the name you want to call the environment. When you are done with your analysis, you can simply deactivate the environment with the command:

conda deactivate

Note For Mac Users! - If you run into problems with the scripts crashing, you might also need to also install the nomkl package to prevent a multithreading bug in numpy. This is not included in the environment YAML file so first activate the tflow environment and install with conda install nomkl)

Feel free to use the editor of your choice, but if you are looking for a free python editor with nice graphical user interface (similar to RStudio), I recommend Spyder. You can install this by first activating the tflow environment, and then typing (note the versions):

pip install spyder==5.2 spyder-kernels==2.1.0

You can then start it inside the tflow environment by typing spyder (on a Unix system try spyder & to have it run in the background). There are also installers available on the Spyder website if you prefer. Note, that Quatro also allows RStudio to seamlessly run Python notebooks as well if you prefer.

R Users

For R users, I highly recommend using RStudio to run your scripts and to install your library packages. You will need to install the following packages:

• reticulate
• tensorflow
• keras
• pROC

A basic install for tensorflow requires you to run the command install_tensorflow() after you install the main tensorflow package. Full installation details are available on the RStudio Tensorflow site.