brainage/Enigma/predict_age_stacking.py

#%%
import os.path as op
import pandas as pd
import numpy as np
import joblib
import sys
from glob import glob
from sklearn.gaussian_process import GaussianProcessRegressor
from sklearn.base import BaseEstimator, RegressorMixin
from sklearn.gaussian_process.kernels import WhiteKernel, DotProduct

# model_path = "/data/project/BrainAge4AD/features_new"

dir_path = sys.argv[1]

model_path = "/model"

class EnigmaGaussianProcessRegressor(RegressorMixin, BaseEstimator):
    def __init__(self, *, noise_level=-1, optimizer="fmin_l_bfgs_b"):
        # default values
        self.constant_mean = 100
        self.noise_level = noise_level
        self.kernel = self.create_kernel()
        self.optimizer = optimizer

    def create_kernel(self):
        # creates a kernel using these default parameters.
        # ConstantKernel for the constant mean function,
        # WhiteKernel for the noise level function with the specified default parameters.
        likhyp = np.log10(0.1)  # Log of noise variance hyperparameter
        # Mean hyperparameter
        kernel = DotProduct() + WhiteKernel(noise_level=10 ** (2 * likhyp))

        return kernel

    def fit(self, X, y):
        self.gpr = GaussianProcessRegressor(
            kernel=self.kernel, optimizer=self.optimizer
        )
        self.gpr.fit(X, y - self.constant_mean)

    def predict(self, X):
        return self.gpr.predict(X) + self.constant_mean
#%%
stacking_model = joblib.load(op.join(model_path, "stacking_enigmaModel.pkl"))
subjs = glob(dir_path + "/**/*features.csv", recursive=True)

#%%
for sub in subjs:
    subj_name = op.basename(sub).removesuffix('_features.csv')
    subj_path = op.dirname(sub)
    sub_feats = pd.read_csv(sub, index_col=0)
    col_feats = sub_feats.columns[sub_feats.columns.str.contains('p[1,2]_r[4,8]s[4,8]_*')]
    pred = stacking_model.predict(sub_feats[col_feats])
    sub_report = [{'subj-path': sub, 'name': op.splitext(subj_name)[0], 'pred-age': pred[0]}]
    df_res = pd.DataFrame(sub_report)
    out_path = op.join(subj_path, op.splitext(subj_name)[0] + '_prediction.csv')
    df_res.to_csv(out_path)

#%%