prime_english.py

# -*- coding: utf-8 -*-
# Made available at http://www.mariofilho.com
# For educational purposes only, Author offers NO GUARANTEES
from sklearn.linear_model import LinearRegression
from sklearn.metrics import mean_absolute_error
import numpy as np
import pandas as pd


#MAPE calculation function
def mape(y_pred,y_true):

    return np.mean(np.abs((y_true - y_pred) / y_true)) * 100


#Loading the data

data = pd.read_csv('prime.csv',header=0,index_col=0).sort_index()

x_data = []
y_data = []

# Creates a feature matrix with values from previous 6 months
for d in xrange(6,data.shape[0]):
    
    x = data.iloc[d-6:d].values.ravel()
    y = data.iloc[d].values[0]

    x_data.append(x)
    y_data.append(y)


x_data = np.array(x_data)
y_data = np.array(y_data)




#Lists to store the predictions of the models
y_pred = []
y_pred_last = []
y_pred_ma = []
y_true = []



#Iterate over the time series creating a new model each month
end = y_data.shape[0]
for i in range(30,end):

    x_train = x_data[:i,:]
    y_train = y_data[:i]
    
    x_test = x_data[i,:]
    y_test = y_data[i]


    model = LinearRegression(normalize=True)
    model.fit(x_train,y_train)

    y_pred.append(model.predict(x_test.reshape(1, -1))[0])
    y_pred_last.append(x_test[-1])
    y_pred_ma.append(x_test.mean())
    y_true.append(y_test)


#Transforms the lists into numpy arrays
y_pred = np.array(y_pred)
y_pred_last = np.array(y_pred_last)
y_pred_ma = np.array(y_pred_ma)
y_true = np.array(y_true)


#Print errors
print '\nMean Absolute Percentage Error'
print 'MAPE Linear Regression', mape(y_pred,y_true)
print 'MAPE Last Value Benchmark', mape(y_pred_last,y_true)
print 'MAPE Moving Average Benchmark', mape(y_pred_ma,y_true)


print '\nMean Absolute Error'
print 'MAE Linear Regression', mean_absolute_error(y_pred,y_true)
print 'MAE Last Value Benchmark', mean_absolute_error(y_pred_last,y_true)
print 'MAE Moving Average Benchmark', mean_absolute_error(y_pred_ma,y_true)


#Cria um gráfico dos valores reais, previsões da regressão linear e do modelo utilizando o último valor
# OPCIONAL - REQUER MATPLOTLIB
#from matplotlib import pyplot as plt
#plt.title('Prime Rate Brazil - Monthly - 2005 to 2014')
#plt.ylabel('Prime Rate')
#plt.xlabel(u'Periods (Months)')
#reg_val, = plt.plot(y_pred,color='b',label=u'Linear Regression')
#true_val, = plt.plot(y_true,color='g', label='True Values')
#plt.xlim([0,85])
#plt.legend(handles=[true_val,reg_val])
#plt.show()