This is the development repository for the search algorithm using semi-unsupervised technique for Summer Research @UC Berkeley SETI. Please note that the repository alone is messy, please follow the menu directory in this readme to browse the code base. This is project used to be Deep SETI but has since matured to a fully usable algorithm.
The original training dataset is described in our paper. Specific links to download the dataset are the following for TRAINING: HIP110750 HIP13402 HIP8497 Note: HIP13402 appears in both the training set AND in the top 8 candidates. However this does NOT mean that we tested or validated on trained data, there are multiple observations of HIP13402 conducted at different times. Thus is not an issue.
For the testing and in general for ANY dataset we use it can be found at the open data archive found here
Preprocessing methods were described in the paper.
Please find the simulation code in scripts here
This requires an installation of SETIGEN which can be found here
pip install setigenPlease find the model training in the following notebook here. To get the trained weights it is located in this repository as well here
We train the random forest in the following script here Weights were not saved in the github repository as the size was far to large to fit. Reproduce this model using the script mentioned above.
We benchmarked the results in the following notebook here
Here is the search list containing all the targets here
Here is the top 8 candidates and their original data here
The visualisation notebook is too large to be place in a github repository. It simply provides the plots visualised in the paper and does not include any scientific novelty.