reproducibility results for IAL2023 workshop paper, we are excited to share our findings and discusse more ideas!
A detailed guide on how to run this messy code.
Step 0: make sure you have a Python environment with basic libraries. Check requirements.txt for more details.
Step 1: generate raw data with data_import.py importing from the SurvSet library is needed.
Use the pip install SurvSet==0.2.6 command to install the same version of the library.
Step 2: process the datasets with the process_and_folds.py script to get a 'clean' version of the data.
This also sets-up a 5-fold cross validation and imputes missing values (do we? double check claim).
Imputation is necessary since the underlying RSF model
(as for scikit-survival==0.21) cannot handle missing values.
Step 3: Select relevant datasets for the analysis (see paper for more details), use the filter_datasets.py script for the task.
The current paramters in the script are the ones used in the paper, and represernt a trade-off between having only 'good' datasets and having 'enough' datasets.
Step 4: Run the main_script.py for the whole active learning procedure. This will call functions from other files such as utillities.py.
It will populate the query-info-data and the plot-and-perfomr-info folders.
Step 5: Analysis of the data and plotting was performed with the stat_testing.py and the plotting_script.py.
simul_data_1 and simul_data_2 respectively, but in the project folder they are genearted under the my_simul_data_1 and my_simul_data_3 respectively.