To set up environment the following libraries are needed :
scikit-image=0.18.* scikit-learn=1.0.* numpy=1.19.* pytorch=1.7.1 python=3.8.* matplotlib=3.4.3 networkx=2.6.3
Note snorkel requires that you use networkx 0.2.6 while these modules require 2.6.*. Will need to make two seperate environments to run the code, one for snorkel, one for the visualization. A bummer, but a consequence of dependencies I guess.
Because of the difficulty switching between environments due to conflicting libraries, we have included a .ipynb file containing all of the code needed to run the Snorkel stuff separately. Snorkel can be installed using the first cell and the subsequent cells do different operations like plotting histograms or getting accuracies of labeling functions.
If you need help running the code please contact bcjohnson7@wisc.edu for help in running the visualizations and cbailey9@wisc.edu for help running the labeling function accuracies and histogram function plotting. SnorkelEnvironment.ipynb
To run the output of a labeling function go to pipeline.py and specify the input via a choice of a labeling function in the input dictionary.
Then look at output in the corresponding folder ./data/<labeling_func>
To look at the plots made for the nested sampling figures, look at paper_plots.py
and just run this function.
Description of major modules:
ns.py- contains the code for the nested sampling rundg.py- contains the code to return the x and y points for the disconnectivity graph visualization.pipeline.py- entry point to do runs over different labeling functions.