This repository accompanies the publication ”Synergy Conformal Prediction applied to Large-Scale Bioactivity Datasets and in Federated Learning”
Recommended packages:
python=3 pandas numpy=1.17.2 cloudpickle scikit-learn
pip install nonconformist==2.1.0
Comment:
Some later versions of numpy (>1.17.2) gives an error deprecation message and seems to give different predicted results for some examples.
conformal_prediction2_types.py [-h] [-i INFILE] [-n NMODELS]
[-m {t,p,b}] [-s {t,c}] [-p PREDFILE]
[-a {rf,gb}]
[-c {acp,indicp,icp,indscp,scp,rscp}]
[-f FORCE]
optional arguments:
-h, --help show this help message and exit
-i INFILE, --infile INFILE
input training file
-n NMODELS, --nmodels NMODELS
number of models (default 20 models)
-m {t,p,b}, --mode {t,p,b}
mode: build models, predict new data from models, both
build and predict
-s {t,c}, --sep {t,c}
file separator: tab or comma
-p PREDFILE, --predfile PREDFILE
input prediction file if mode == p
-a {rf,gb}, --algo {rf,gb}
algorithm: RF or GBT
-c {acp,indicp,icp,indscp,scp,rscp}, --cptype {acp,indicp,icp,indscp,scp,rscp}
mode: type of cp: Aggregated, individual ICP, ICP,
individual SCP, SCP, random-SCP
-f FORCE, --force FORCE
force e.g second column name to "class" by -f 2, f < 0
to skipCommand line for predicting external test set using rscp and 10 models:
conformal_prediction2_types.py -n 10 -m b -s t -a rf -c rscp -f -1 -i trainfile -p testfile Using the provided example data:
conformal_prediction2_types.py -n 10 -m b -s t -a rf -c rscp -f -1 -i example_data_train.txt -p example_data_test.txt