MMI estimation

Dependencies

• pytorch
• joblib
• sklearn
• numpy
• scipy
• matplotlib
• tqdm

Dependencies for if-estimators

• oct2py

Dependencies for AIC (better install following packages with brew if your machine is Mac)

• cmake
• boost
• python3
• boost-python3

build AIC module

cd MMI/IC
cmake src/
# set paths of python3, boost and boost-python3 in CMakeCache.txt
# and then cmake again
# until you see "-- Generating done"
make
# if successful, you will see
# Scanning dependencies of target AIC
# [ 50%] Building CXX object CMakeFiles/AIC.dir/pylib.cpp.o
# [100%] Linking CXX shared library AIC.so
# [100%] Built target AIC
# now AIC is built

How to run

cd ..
# check the configuration in settings.py
python -m MMI.main

How to add new mi estimators

• go to model/
• create a new class (see mine.py/linear_regression.py for example)
• must contain the predict function (given a N by 2 matrix, return the MI estimator (float))
• go to settings.py, add the corresponding configuration in the variable model

How to add new synthetic data

• go to data/
• create a new class (see bimodal.py for example)
• must contain the two property: data and ground_truth (see bimodal.py for example)
• go to settings.py, add the corresponding configuration in the variable data