from sklearn.datasets import make_regression

import numpy as np

from sklearn.model_selection import KFold,train_test_split

from sklearn.metrics import mean_absolute_error,mean_squared_error

from sklearn.neural_network import MLPRegressor
from sklearn.ensemble import AdaBoostRegressor

from sklearn.decomposition import PCA


from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
import time

trainx, trainy = make_regression(n_samples=2000, n_features=100, random_state=0, noise=4.0,
                       bias=2.0)

print trainx[0][0:5]
print trainy[0:5]
def evalTrainData(trainDatax,trainV,random_state=2016,eid=''):
    fold=10

    scores=[]
    pccs=[]
    cvcount=0
    assert len(trainDatax)==len(trainV)

    for roundindex in range(0,3):
        skf=KFold(fold,shuffle=True,random_state=random_state+roundindex)
        for trainIndex,evalIndex in skf.split(trainDatax):
            t1=time.time()
            cvTrainx,cvTrainy=trainDatax[trainIndex],trainV[trainIndex]
            cvEvalx,cvEvaly=trainDatax[evalIndex],trainV[evalIndex]

            scaler=StandardScaler()
            cvTrainy=scaler.fit_transform(cvTrainy.reshape(-1, 1)).ravel()
            lsvr=getxlf(random_state=random_state)
            lsvr.fit(cvTrainx,cvTrainy)
            predict=lsvr.predict(cvEvalx)
            predict=scaler.inverse_transform(predict.reshape(-1,1)).ravel()
            score=mean_absolute_error(cvEvaly,predict)
            pcc=np.corrcoef(cvEvaly,predict)[0, 1]
            print (cvcount,'MAE',score,'PCC',pcc,time.time()-t1,time.asctime( time.localtime(time.time()) ) ,'Train sahpe:',cvTrainx.shape,'eval sahpe:', cvEvalx.shape)
            scores.append(score)
            pccs.append(pcc)
            cvcount+=1

    print ('###',eid,'MAE',np.mean(scores),'PCC',np.mean(pccs))


pca_n_components=100
def getxlf(random_state=2016):
    xlf1= Pipeline([
                          ('svd',PCA(n_components=pca_n_components)),
                          ('regressor',AdaBoostRegressor(#random_state=int(time.time()),

                           base_estimator=MLPRegressor(random_state=random_state,early_stopping=True,max_iter=2000)
                                                         ,n_estimators=30,learning_rate=0.01)),
                          ])

    return xlf1

evalTrainData(trainx,trainy)