Identification of Fast Radio Burst/Single Pulses (FRB/SP) and Radio Frequency Interference (RFI) using Deep Convolutional Neural Network for MeerKAT facility. The code uses two inputs: the DM-Time and Frequency-Time images/arrays. Each image acts as an input to a CNN. At the end both DM-T CNN and Freq-T CNN are fused and pass through dense layers as shown in the figure below.
Follow the instructions in Installation.txt to install all dependencies.
To train the model from scratch, either use FRBID - DEMO - MULTIINPUT.ipynb or train.py. Note that there are several parameters that need to be changed if one want different configuration, else run the code as follows:
python train.py
or run all cells in FRBID - DEMO - MULTIINPUT.ipynb
To make prediction on new candidates that do not have a label, use either FRBID - prediction-phase.ipynb or predict.py. Note that a directory containing all h5 candidate files should be available and some parameters need to be specified, for e.g the model_name, the directory to save the csv file containing the prediction, the directory of the h5 files and the threshold probability.
Run prediction on new candidate files as follows:
python predict.py -d ./data/test_set/ -r ./data/results_csv/ -p 0.5
or run prediction on default settings as follows:
python predict.py
or run all cells in FRBID - prediction-phase.ipynb