#python

Deep Learning Resources Here are two popular and powerful machine learning libraries used for deep learning (refer to Machine Learning Resources to understand the concepts): Activate your Python virtual environment before running the following commands in the terminal: TensorFlow Install the latest TensorFlow version: conda install tensorflow-gpu -y PyTorch Install the latest PyTorch version: conda install pytorch torchvision -y DioscoriDESS DioscoriDESS is a segmentation framework developed by UCSF Radiology. [Read More]

DioscoriDESS - Make sure to replace `/MY_PATH/dioscoridess/` with a location of your choice. Make a directory where your input files (.yaml) and output files (.h5) will be stored: mkdir -p /MY_PATH/dioscoridess/cfgs mkdir -p /MY_PATH/dioscoridess/logs mkdir -p /MY_PATH/dioscoridess/ckpt Create your yaml file and save in /MY_PATH/dioscoridess/cfgs/my_input.yaml: # Sample ./yaml config filecommon:seed:1234vis_GPU:'1'#single GPU compatibilitylog_path:'./experiments/training/logs/'save_path:'./experiments/training/ckpt/'print_freq:25data_train:data_root:'/data/bigbone5/JacobOeding/DL_code/splits/axial_train.csv'batch_size:4im_dims:!!python/tuple[512,512]crop :!!python/tuple[0,0,0,0]num_classes:3idx_classes :[0,1,2]num_channels:1normalization_file:'/data/bigbone5/JacobOeding/DL_code/dataInfo/Axial_MinMaxDataset.csv'data_val:data_root :'/data/bigbone5/JacobOeding/DL_code/splits/axial_val_onlySegmentedSlices.csv'batch_size:4im_dims:!!python/tuple[512,512]crop :!!python/tuple[0,0,0,0]num_classes:3idx_classes :[0,1,2]num_channels:1normalization_file:'/data/bigbone5/JacobOeding/DL_code/dataInfo/Axial_MinMaxDataset.csv'evaluate_mode:TruedataLoader_folder:'dataLoader'learn:max_steps:500000# instead of epochslr :0.00005optimizer:'adam'loss:'dice_loss_softmax'metrics:'dice_score'weights:[1,1.2,1]num_classes:3dataloader:'data_loader'comp:['femur','tibia','background']key_slice:!!python/tuple[0,0,0]patience:15# in number of validation runskeep_prob :0. [Read More]

Jupyter Resources One of the simplest and most versatile ways to use python is through Jupyter Lab. In order to get it working within your virtual environment, follow these steps: Installation Follow the earlier steps to SSH into one of the login node servers. Activate your virtual environment Run the following commands: If you created an empty environment, use the following command first: conda install jupyterlab -y jupyter lab --no-browser --port 9XXX --notebook-dir=/ Replace 9XXX with the number used for the SSH port forwarding e. [Read More]

Machine Learning Resources

Helpful Stanford introductory course to Machine Learning.

---

• Stanford CS229: Machine Learning Overview

---

Deep Learning Resources

Helpful Stanford introductory course to Deep Learning and Convolutional Neural Networks (CNN).

---

• Stanford CS231n: Convolutional Neural Networks for Visual Recognition

---

Useful 3blue1brown educational video on CNNs

---

Python Resources The servers have several versions of python, with the default version being python 2.7. The best practice when coding in python is to set up a virtual environment with the libraries you will be using. To do this, follow these steps: Set up a virtual environment Use the following command to initialize the latest conda in the servers: /netopt/rhel7/versions/python/Anaconda3-edge/bin/conda init bash tcsh Close the current terminal and open a new one for the changes to take place Use the following commands to set up the directory where the python environments and packages will be stored: [Read More]