Installation.txt

"""
Authors : Zafiirah Hosenie
Email : zafiirah.hosenie@gmail.com or zafiirah.hosenie@postgrad.manchester.ac.uk
Affiliation : The University of Manchester, UK.
License : MIT
Status : Under Development
Description :
Python implementation for FRBID: Fast Radio Burst Intelligent Distinguisher.
This code is tested in Python 3 version 3.5.3  
"""

'------------------------------------------------------------------------------------------------------------'
				     SETTING UP PYTHON ENVIRONMENT
'------------------------------------------------------------------------------------------------------------'
# Create new environment
conda create -n name_env python=3.6 for e.g conda create -n frbrfi python==3.6

# Activate the environment
conda activate frbrfi

# Install the required packages
conda install numpy pandas jupyter notebook matplotlib scipy h5py
conda install tensorflow-gpu==1.9.0
pip install -U imbalanced-learn
pip install scikit-learn==0.23.1
pip install Keras==2.0.9

# Note that atleast 1 GPU should be available to run the codes
If only one GPU available, on terminal run
1. export CUDA_VISIBLE_DEVICES=1 # This choose the first GPU
2. If you are running jupyter notebook, then type jupyter notebook else, just run the .py script normally. For e.g python train.py or python predict.py


'------------------------------------------------------------------------------------------------------------'
					TRAIN MODEL FROM SCRATCH
'------------------------------------------------------------------------------------------------------------'
The model can be trained either by using 'train.py' or 'FRBID - DEMO.ipynb'

With Notebook:
On terminal: Jupyter notebook and then open 'FRBID - DEMO.ipynb' in a browser and run all cells. Note that there are some parameters that needs to be changed : For e.g directory where the training and test set are located. Also other parameters for e.g Model Name can be changed accordingly.

With Python Script:
On terminal, type: python train.py. Some parameters need to be changed - training and test set directory, model_name...


'------------------------------------------------------------------------------------------------------------'
				    PREDICTION USING ALREADY TRAINED MODEL
'------------------------------------------------------------------------------------------------------------'
New hdf5 candidates can be predicted directly without training the model again. Here we use weights from FRBID_model to perform prediction. For this step, one can use either
1. predict.py, or
2. FRBID - prediction-phase.ipynb

With Python Script:
Note that, predict.py takes some arguments that can be specified during prediction
1. --data_dir (directory where hdf5 files located), --result_dir(directory to save the csv file), --model_cnn_name (NET2. NET3,...), --n_images ('dm_fq_time), --probability (0.5) --> Open script to have more info about those parameters.

One example how to run it: python predict.py -d ./data/test_set/ -r ./data/results_csv/ -n dm_fq_time -m NET3 -p 0.5 or run directly with default settings: python predict.py

With Jupyter notebook:
Change the parameters in the second cells and then run all cells