An advance algorithm for depression and anxiety detection from extracted facial and audio features. The implementation is based of several AVEC challenge participants with a number of personal improvements and additions.
To create the environment run:
conda env create -f darwin_env.yaml
Then activate it with
conda activate darwin
If you install any new libraries, make sure you do it with conda unless there is no other choice. Then, update the environment with
conda env export --from-history > darwin_env.yaml
To run you own experiment you have to first edit the config.ini
file
paying careful attention to naming conventions and ensuring you are not
duplicating already existing functionality with a different name.
model.py
)+
signmodel.py
and add a
call to it in the switcher
statement. Make sure to use try
except
clause inside
you function to avoid errors.data.py
functionality. Make sure the final data is formatted correctly.mlflow ui
in same directory
as mlruns folderYou can optimize your code by running pipelie.optimize
and selecting parameter appropriately. Use a tuple
to determine min and max for each of the desired parameters. For example (100, 1000)
will search
for the best parameter in range between 100 and 1000. The optimization uses a Bayesian Optimization technique
and more information can be found here.