from OLSgradientDescent import gradientDescent
import pandas as pd
from sklearn.decomposition import PCA
from sklearn import linear_model
nInstances = 500
nParams = 100
noiseLevel = 500
trueWeights = np.asarray(range(nParams))
# trueWeights[5:] = 0
# trueWeights[:4] = 0
X = np.random.rand(nInstances,nParams)
Y = np.dot(X,trueWeights) + noiseLevel*np.random.rand(np.size(nInstances))
test = gradientDescent(Y, X, 0, 1000, 0.01)
input = pd.read_csv('OnlineNewsPopularity/OnlineNewsPopularity.csv')
inp = input.as_matrix()
X = inp[:,1:60]
Y = inp[:,60]
Y = Y.reshape(Y.shape[0],1)
X, Y = shuffle(X,Y)
pca = PCA(n_components = 3)
pca.fit(X)
print(pca.explained_variance_ratio_)
X_new = pca.transform(X)
clf = linear_model.Lasso(alpha=lambArray[j], max_iter = 10000)
clf.fit(Xtrain,Ytrain)
Ypred = clf.predict(Xvalid)
error = mse(Ypred,Yvalid)
lassoErrorArray[i,j] = error
#Estimation and prediction using closed form solution
wEst = OLSClosed(Xtrain,Ytrain,L2 = lambArray[j])
Ypred = np.dot(Xvalid,wEst)
error = mse(Ypred,Yvalid)
closedErrorArray[i,j]= error
#Estimation and prediction using gradient descent
wEst = gradientDescent(Ytrain,Xtrain,0.0001, 10000, 0.01, 'ridge', lambArray[j])
Ypred = np.dot(Xvalid,wEst['Weights'])
error = mse(Ypred,Yvalid)
gradientErrorArray[i,j]= error
#Estimation and prediction using closed form solution AND PCA
wEst = OLSClosed(np.hstack((np.ones((np.size(Xtrain,axis=0) ,1)),pca.transform(Xtrain[:,1:]))),Ytrain,L2 = lambArray[j])
Ypred = np.dot(np.hstack((np.ones((np.size(Xvalid,axis=0) ,1)),pca.transform(Xvalid[:,1:]))),wEst)
error = mse(Ypred,Yvalid)
pcaErrorArray[i,j]= error
aveLassoError = np.average(lassoErrorArray,axis = 0)
aveClosedError = np.average(closedErrorArray,axis = 0)
aveGradientError = np.average(gradientErrorArray,axis = 0)
avePCAError = np.average(pcaErrorArray,axis = 0)