def train(lowerBidLimit=200, upperBidLimit=400, bidIncrement=1):
print('Training linear bidding strategy model:')
advertisers = list(set(trainingDF['advertiser']))
n = len(advertisers) * len(range(lowerBidLimit, upperBidLimit, bidIncrement))
i = 0
j = 1
optimalBaseBidPrices = []
optimalBaseBids = {}
for advertiser in advertisers:
highest = (advertiser, 0, 0) # tuple containing highest ctr and corresponding bid prices
for bid in range(lowerBidLimit, upperBidLimit, bidIncrement):
i += 1
progress(i, n, '('+str(j)+'/'+str(len(advertisers))+')')
rows = trainingDF[(trainingDF['advertiser']==advertiser) & (trainingDF['bidprice']<=bid)]
imps = len(rows) # imps = number of impressions
clicks = len(rows[rows['click']==1])
if imps > 0:
ctr = clicks / imps # ctr = click through rate
else:
ctr = 0
if ctr > highest[1]:
highest = (advertiser, ctr, bid)
optimalBaseBidPrices.append(highest)
optimalBaseBids[advertiser] = highest[2]
j += 1
for optimalBaseBidPrice in optimalBaseBidPrices:
print(optimalBaseBidPrice)
return optimalBaseBids
def testFunc(trainResults, csv, silent):
if silent == 1:
print 'Testing:'
sums = {}
bidprices = {}
data = {}
data['bidid'] = []
data['bidprice'] = []
n = len(csv)
for i in range(0, n):
if silent == 1:
progress(i + 1, n)
slotprice = csv.slotprice.values[i]
advertiser = csv.advertiser.values[i]
bidid = csv.bidid.values[i]
bidprice = trainResults[advertiser][slotprice]
data['bidid'].append(bidid)
data['bidprice'].append(bidprice)
bidprices[bidid] = bidprice
return data, bidprices
def predict(constant_bidprices, validation_set):
print('Predicting bid prices using constant bidding strategy model:')
bids = {}
validationDF = pd.read_csv(validation_set)
n = len(validationDF)
for i in range(0, n):
progress(i + 1, n)
bidid = validationDF.bidid.values[i]
advertiser = validationDF.advertiser.values[i]
constBidPrice = constant_bidprices[advertiser]
bids[bidid] = constBidPrice
return bids
def predict(advertiser_bidrange, validation_set):
print('Predicting bid prices using random bidding strategy model:')
bids = {}
validationDF = pd.read_csv(validation_set)
n = len(validationDF)
for i in range(0, n):
progress(i + 1, n)
bidid = validationDF.bidid.values[i]
advertiser = validationDF.advertiser.values[i]
minBidPrice = advertiser_bidrange[advertiser][0]
maxBidPrice = advertiser_bidrange[advertiser][1]
diff = maxBidPrice - minBidPrice
constBidPrice = random.randint(minBidPrice, maxBidPrice + diff)
bids[bidid] = constBidPrice
return bids
def train(training_set, lowerBidLimit=200, upperBidLimit=400, bidIncrement=1):
print('Training constant bidding strategy model:')
trainingDF = pd.read_csv(training_set)
advertisers = list(set(trainingDF['advertiser']))
n = len(advertisers) * len(
range(lowerBidLimit, upperBidLimit, bidIncrement))
i = 0
j = 1
optimalConstBidPrices = []
optimalConstBids = {}
for advertiser in advertisers:
highest = (
advertiser, 0, 0, 1
) # tuple containing highest ctr and corresponding bid prices
for bid in range(lowerBidLimit, upperBidLimit, bidIncrement):
i += 1
progress(i, n, '(' + str(j) + '/' + str(len(advertisers)) + ')')
rows = trainingDF[(trainingDF['advertiser'] == advertiser)
& (trainingDF['bidprice'] <= bid)]
imps = len(rows) # imps = number of impressions
clicks = len(rows[rows['click'] == 1])
price = sum(rows.payprice.values)
if imps > 0:
ctr = clicks / imps # ctr = click through rate
cpc = price / clicks # cost per click
else:
ctr = 0
cpc = 1
if (ctr > 0) and (ctr / cpc) > (highest[1] / highest[3]):
highest = (advertiser, ctr, bid, cpc)
optimalConstBidPrices.append(highest)
optimalConstBids[advertiser] = highest[2]
j += 1
for optimalConstBidPrice in optimalConstBidPrices:
print(optimalConstBidPrice)
return optimalConstBids
def predict(base_bidprices):
bids = {}
n = len(validationDF)
print('Calculating pCTRs using logistic regression:')
pctrs = getPCTRS(trainingDF, validationDF)
print('Calculating avgCTR:')
imps = len(validationDF) # imps = number of impressions
clicks = len(validationDF[validationDF['click']==1])
avgCTR = clicks / imps # ctr = click through rate
print('Predicting bid prices using linear bidding strategy model:')
for i in range(0, n):
progress(i+1, n)
bidid = validationDF.bidid.values[i]
advertiser = validationDF.advertiser.values[i]
baseBidPrice = int(base_bidprices[advertiser] * pctrs[i] / avgCTR)
bids[bidid] = baseBidPrice
return bids
def train(training_set):
print('Training random bidding strategy model:')
trainingDF = pd.read_csv(training_set)
advertisers = list(set(trainingDF['advertiser']))
n = len(advertisers)
i = 0
advertiserBidRange = {}
for advertiser in advertisers:
i += 1
progress(i, n, '(' + str(i) + '/' + str(len(advertisers)) + ')')
rows = trainingDF[(trainingDF.advertiser == advertiser)]
bids = rows.bidprice.values
minBidPrice = min(bids)
maxBidPrice = max(bids)
advertiserBidRange[advertiser] = (minBidPrice, maxBidPrice)
for advertiser in advertiserBidRange:
print(advertiser, advertiserBidRange[advertiser])
return advertiserBidRange
def training(csv, silent):
if silent == 1:
print 'Training:'
advertisers = list(set(csv['advertiser']))
data = {}
for advertiser in advertisers:
#if silent == 1:
print advertiser
data[advertiser] = {}
advertiserRows = csv[(csv['advertiser'] == advertiser)]
for slotprice in range(0, 400):
if silent == 1:
progress (slotprice + 1, 400)
rows = advertiserRows[(advertiserRows['slotprice'] == slotprice)]
clicks = rows[(rows['click'] == 1)]
if len(rows) > 0:
difference = float(len(rows)) - float(len(clicks))
avgbid = float(sum(rows['bidprice'].values)) / float(len(rows))
bidprice = avgbid - difference * float(0.01001)
# bidprice = avgbid - difference * float(0.01001) + float(len(clicks))
if bidprice <= float(0):
# if difference < float(100000):
# bidprice = avgbid - difference * float(0.0000425)
# # bidprice = avgbid - difference * float(0.000043)
# else:
# bidprice = avgbid - difference * float(0.00000425)
# # bidprice = avgbid - difference * float(0.0000043)
# data[advertiser][slotprice] = round(bidprice)
data[advertiser][slotprice] = 0
else:
won = rows[(rows['bidprice'] <= round(bidprice))]
wonClicks = won[(won['click'] == 1)]
if len(won) > 0:
ctr = float(len(wonClicks)) / float(len(won))
else:
ctr = 0
newbidprice = round(bidprice) + 1
new = rows[(rows['bidprice'] <= newbidprice)]
newClicks = new[(new['click'] == 1)]
if len(new) > 0:
ctrMore = float(len(newClicks)) / float(len(new))
else:
ctrMore = 0
# if len(new) > len(won):
# bidprice = bidprice + 1
if ctr < ctrMore:
bidprice = bidprice + 1
data[advertiser][slotprice] = round(bidprice)
else:
data[advertiser][slotprice] = 0
return data