XGB.txt

### XGB
!pip install shap
import pandas as pd
import numpy as np
from numpy import *
from sklearn import preprocessing
from sklearn.model_selection import train_test_split
from sklearn.metrics import log_loss, roc_auc_score,f1_score
from sklearn.metrics import classification_report, confusion_matrix
from sklearn.model_selection import GridSearchCV
from sklearn.metrics import accuracy_score
import keras
import math
from sklearn.model_selection import KFold
from keras.wrappers.scikit_learn import KerasClassifier
from xgboost.sklearn import XGBClassifier
import xgboost as xgb
import shap
from sklearn.utils.class_weight import compute_sample_weight
from keras.utils import to_categorical 
import os
import warnings
warnings.filterwarnings("ignore")

### load data
df = pd.read_csv('.../dat.1944.csv')
tdf = pd.read_csv('.../dat.60.csv')
df = df.replace('ASD', 0)
df = df.replace('DLD', 1)
df = df.replace('GDD', 2)
tdf = tdf.replace('ASD', 0)
tdf = tdf.replace('DLD', 1)
tdf = tdf.replace('GDD', 2)
col = df.columns[1:]
col1 = df.columns[2:]
dat = df[col]
grp = dat['diag']
x_test = tdf[col1]
y_test = tdf['diag']

x_train = dat[col[1:]]
y_train = dat['diag']

xlf = XGBClassifier(
max_depth = 4,
learning_rate = 0.0001,
n_estimators = 5000,
silent = True,
eval_metric = 'merror',
objective = 'multi:softprob',
subsample = 0.9,
colsample_bytree = 0.9,
num_class = 3)  

xgc = xlf

xgc.fit(x_train, y_train, sample_weight = compute_sample_weight("balanced", y_train))
pred_test = xgc.predict_proba(x_test)
  ### predict test
pred_res = np.argmax(pred_test, axis=1)
mat_test = confusion_matrix(y_test, pred_res)
acc_all_test = np.diag(mat_test)/mat_test.sum(axis=1)
acc_test = accuracy_score(y_test, pred_res)
acc1 = acc_all_test[0]
acc2 = acc_all_test[1]
acc3 = acc_all_test[2]
explainer = shap.TreeExplainer(xgc)
shap_values = explainer.shap_values(x_test)
print(shap_values)
print('acc:\n', acc_test)
print('r:\n', acc_all_test)
print('mix-matrix:\n', mat_test)