Separate the analysis into stages

[x] task_0_generate_features.sl -> output is the feature.csv (has all the features necessary)
[x] task_1_find_best_classifier.sl (this will run the gridsearch in parallele) -> output is a pickled best clf
[x] bring in the data generated from task_0 on my local machine to prototype
[x] import ml_tools inside the project
[x] setup the loading and the filtering of the dataset properly depending on parameters
[x] setup the gridsearch properly with the clf mentioned in the abstract
[x] capture the best parameters and the best clf and save it
[x] make the script iterate on each 20 classification to do {2 for type, 2 for epochs, 5 for parameters grouping}. See how to run 20 different tasks on 20 different nodes each having 40 cores over here on stackoverflow

For these checkout : joblib

[x] task_2a_generate_boostrap_interval.sl (this will run bootstrap with 500 cores) -> output bootstraps
[x] take the code from the boostrap_interval.py script from eeg-pain study into this codebase
[x] implement a function to read the pickled output from task_1 and instantiate a classifier with the right parameter
[x] create the slurm file to run one bootstrap computation of n = 1000 on 500 cores (will need to check out joblib for that)
[x] create a bash file similar to what has been done for the previous part to instantiate 20 jobs with 500 cores

Not too bad I don't really need to have that much cores, it doesn't take very long to run 1000 iterations on 40 cores

[x] task_2b_generate_permutation_tests.sl (run permutation with 500 cores) -> output permutations
[x] Reuse the same structure as 2a to get the input output going properly
[x] Bring in the code from eeg-pain study for the permutation
[x] create a bash and a slurm file to run the 20 jobs on 40 cores

BIAPT / AEC-wPLI-comparison