Skip to content

Latest commit

 

History

History
executable file
·
89 lines (68 loc) · 2.28 KB

README.md

File metadata and controls

executable file
·
89 lines (68 loc) · 2.28 KB

Semantic segmentation (pixel-wise classification) network to perform cosmic ray and beam particle separation in prototype DUNE detector.

alt text

Instructions

Git clone the code

git clone https://github.com/ArbinTimilsina/DeepLearningWithProtoDUNE.git
cd DeepLearningWithProtoDUNE

Download the dataset

# 320x320 inputs; min 2000 beam hits 
wget -O input_files.zip https://www.dropbox.com/sh/a00fbuye3i1c0sj/AACeI2l-iEpIoeDbDBtogjJKa?dl=1
unzip input_files.zip -d input_files
rm -rf input_files.zip

Train the model

python train_model.py --help

# Example
python train_model.py -o Development -e 5

Details can be found in the configuration file.

Analyze the model

python analyze_model.py --help

# Example
python analyze_model.py -p 5 -s Development

Additional information

To create a conda environment (Python 3)

conda create --name envDeepLearningWithProtoDUNE python=3.5
conda activate envDeepLearningWithProtoDUNE
pip install --upgrade pip
pip install -r requirements/cpu_requirements.txt
conda install pydot graphviz

To run with singularity container

singularity pull --name DeepLearningWithProtoDUNE.img shub://ArbinTimilsina/Base-Singularity:deeplearningwithprotodune

# If using GPUs, don't forget --nv option
singularity exec --nv DeepLearningWithProtoDUNE.img python train_model.py -o Development -e 5

To switch Keras backend to TensorFlow

KERAS_BACKEND=tensorflow python -c "from keras import backend"

To calculate the weights

python calculate_weights.py

It will run over the default traning files in the configuration. Median for each class will be displayed in plots/weights_median.pdf.

To make plots of events

# For 10 events
python plot_events.py --events 10

To open jupyter notebook

Create an IPython kernel for the environment

# Create an IPython kernel for the environment
python -m ipykernel install --user --name envDeepLearningWithProtoDUNE --display-name "envDeepLearningWithProtoDUNE"

# Open the notebook
jupyter notebook miscellaneous/model_creation_playground.ipynb

# Note: Make sure to change the kernel to envDeepLearningWithProtoDUNE using the drop-down menu (Kernel > Change kernel > envDeepLearningWithProtoDUNE)