def run_perm_test(row): row_array = row.split(';') perm_ind = row_array[0] perm_y = row_array[1] perm_y = np.asarray(perm_y.split(',')).astype(int) analysis_name = 'schaefer100_noTiv_perm_' + str(perm_ind) data_folder = '/scratch/tmp/wintern/iq_frankfurt/' project_folder = '/scratch/tmp/wintern/iq_frankfurt/results/noTivRescaling/perm/schaefer100' cache_dir = '/scratch/tmp/wintern/cache' # get data data = IQData(data_folder=data_folder, tiv_rescaled=False) covariates = np.asarray([data.age, data.gender, data.handedness]).T data.load_schaefer100(use_cached=True) # run analysis pipe = construct_hyperpipe_schaefer(analysis_name, project_folder, cache_dir) pipe.groups = data.fsiq pipe.fit(data.schaefer100, perm_y, **{'covariates': covariates}) os.remove(pipe.output_settings.pretrained_model_filename)
Last updated: 18-02-2019 Author ------ Nils R. Winter [email protected] Translationale Psychiatrie Universitaetsklinikum Muenster """ import sys sys.path.append('/scratch/tmp/wintern/iq_frankfurt/photonai') sys.path.append('/scratch/tmp/wintern/iq_frankfurt/') from analyses.analysis_base import construct_hyperpipe_schaefer from data.data import IQData import numpy as np import os analysis_name = 'schaefer100_no_tiv' data_folder = '/scratch/tmp/wintern/iq_frankfurt/' project_folder = '/scratch/tmp/wintern/iq_frankfurt/results/noTivRescaling/schaefer100' cache_dir = '/scratch/tmp/wintern/cache' # get data data = IQData(data_folder=data_folder, tiv_rescaled=False) covariates = np.asarray([data.age, data.gender, data.handedness]).T data.load_schaefer100(use_cached=False) # run analysis pipe = construct_hyperpipe_schaefer(analysis_name, project_folder, cache_dir) pipe.fit(data.schaefer100, data.fsiq, **{'covariates': covariates}) os.remove(pipe.output_settings.pretrained_model_filename)