import numpy as np
from sklearn.linear_model import LinearRegression
from math import floor
import matplotlib.pyplot as plt

#daten laden
path = "C:/temp/cardata.csv"

data =  np.loadtxt(fname = path, delimiter = ',')
rows, cols = data.shape

allindices = np.asarray([x for x in range(rows)])
# training / evaluation 
training = np.random.choice(rows,floor(0.25*rows))
evaluation = np.setdiff1d(allindices,training)

# trainingsdaten
# data[training,:]



def getSquaredError(xtrain,ytrain,xeval,yeval):
    reg = LinearRegression(fit_intercept=True).fit(xtrain, ytrain)
    ypred = reg.predict(xeval) # äquivalent zu a*x+b
    # von Hand
    # ypred2 = reg.coef_*xeval+reg.intercept_
    # ypred und ypred2 sind identisch; a*x+b; a=reg.coef_, b=reg.intercept_
    RSS = sum((ypred-yeval)**2)
    return (RSS/ypred.size)

## example
print(getSquaredError(data[training,8:9],data[training,7:8],data[evaluation,8:9],data[evaluation,7:8]))

results = [0]*cols

for i in range(cols):
    if i==7:
       results[i]  = float('inf')
       continue
    results[i] = float(getSquaredError(data[training,i:(i+1)],data[training,7:8],data[evaluation,i:(i+1)],data[evaluation,7:8]))

# Vergleiche Resultate
print(results)
# Vergleiche Resultate as log
print(np.log(results))

minval = results[0]
minindex = 0
for i in range(1, len(results)):
    if(results[i] < minval):
        minindex = i
        minval = results[i]

print(minval)
print(minindex) 
#automaisch      
print(np.amin(np.array(results,dtype=object)))