import xgboost as xgb

import numpy as np

from scipy.sparse import csr_matrix

from sklearn import metrics

import csv

from SendEmail import sendEmail

# XGBoost 101 found at http://xgboost.readthedocs.io/en/latest/python/python_intro.html

def get_data(data):

return d

def format_data(data):

return x, y

errorCounter = 0

if __name__ == "__main__":

with open('/home/rmendoza/Desktop/XGBoost/XGB-Grid-Results7-hourly.csv', 'w') as file:

try:

# Inputting training and testing set

wr = csv.writer(file, quoting = csv.QUOTE_MINIMAL)

wr.writerow(['J-score','AUC','Recall','Filter','Cut','Net_Savings', 'eta', 'num_round', 'day_trained', 'day_predicted','hour_trainedAndTested'])

for i in range(4,23): #i is the day, goes to 24 to test on 25 and end. :P

for j in range(0,24): # j is the hour

num_round = 500

eta = 0.1

alpha = 0

ph0 = str(j).rjust(2,'0') #the hour on which to train and test

p0 = str(i).rjust(2,'0') #the day to train

p1 = str(i+1).rjust(2,'0') #the day to test

#train_data, train_label = format_data("/home/kbhalla/Desktop/Data/day_samp-06-"+p0+".npy")

train_data, train_label = format_data('/media/54D42AE2D42AC658/DataHourly/output_new_06'+p0+ph0+'.npy')

dtrain = xgb.DMatrix(train_data, label=train_label)

#test_data, test_label = format_data("/home/kbhalla/Desktop/Data/day_samp-06-"+p1+".npy")

test_data, test_label = format_data('/media/54D42AE2D42AC658/DataHourly/output_new_06'+p1+ph0+'.npy')

dtest = xgb.DMatrix(test_data, label=test_label)

p = np.count_nonzero(train_label)

n = len(train_label) - p

# Setting parameters

evallist = [(dtrain,'train'), (dtest,'eval')]

# Train Model 1...

try:

print 'Loading the model... '

bst.load_model('/home/rmendoza/Desktop/XGBoost/testHourly/testHourly' + p0 + '_to_' + p1 + ph0 + '_v2.txt') # load model

except:

print "Some mistake in loading the model, so now we train again..." #pass #to skip

param = {'booster':'gbtree', # Tree, not linear regression

'objective':'binary:logistic', # Output probabilities

'eval_metric':['auc'],

'bst:max_depth':5, # Max depth of tree

'bst:eta':eta, # Learning rate (usually 0.01-0.2)

'bst:gamma':8.5, # Larger value --> more conservative

'bst:min_child_weight':1,

'scale_pos_weight':n/float(p), # Often num_neg/num_pos

'subsample':.8,

'silent':1, # 0 outputs messages, 1 does not

'save_period':0, # Only saves last model

'nthread':6, # Number of cores used; otherwise, auto-detect

'seed':25}

#num_round = int(100*0.2/float(eta)) # Number of rounds of training, increasing this increases the range of output values

bst = xgb.train(param,

dtrain,

num_round,

evallist) # If error doesn't decrease in n rounds, stop early

bst.dump_model('/home/rmendoza/Desktop/XGBoost/testHourly/testHourly' + p0 + '_to_' + p1 + ph0 + '_v2.txt')

######### Predict/test the model on next day

y_true = test_label

y_pred = bst.predict(dtest)

# J score, AUC score, best recall, best filter rate, best cutoff

results = [0, 0, 0, 0, 0, 0, eta, num_round, p0, p1,ph0]

for cutoff in range(0, 31):

cut = cutoff/float(100) # Cutoff in decimal form

y = y_pred > cut # If y values are greater than the cutoff

recall = metrics.recall_score(test_label, y)

# true_negative_rate = sum(np.logical_not(np.logical_or(test_label, y)))/float(len(y_pred))

filter_rate = sum(np.logical_not(y))/float(len(y_pred))

if recall*6.7+filter_rate > results[0]:

results[0] = recall*6.7+filter_rate

results[1] = metrics.roc_auc_score(test_label, y)

results[2] = recall

results[3] = filter_rate

results[4] = cut

results[5] = 127000*filter_rate -5200 -850000*(1-recall)

wr.writerow(results)

print 'done for the hour', j

print '--------------------------'

print 'done for the DAY', i

print '-------------------------------------'

print '-------------------------------------'

print '_______________________________________________________________________'

print '_______________________________________________________________________'

except:

pass

errorCounter += 1

print 'There was an error'

subjeto = 'Error on code... countOfError' + str(errorCounter)

sendEmail('moralesmendozar@gmail.com',subjeto,"XGBoost-trainHoursDaily.py encountered an error. :P")

#time.sleep(20) #sleep

sendEmail('moralesmendozar@gmail.com','Code Done2',"XGBoost-trainHoursDaily.py ended running in the local RIPS computer. :P")