def behavior(start_date, end_date, account_name, property_name, profile_name, max_results):
# Let pandas fetch the data from Google Analytics, returns a generator object
df_chunks = ga.read_ga(secrets=client_secrets,
account_name=account_name,
property_name=property_name,
profile_name=profile_name,
dimensions=['date', 'hour', 'minute'],
metrics=['pageviews'],
start_date=start_date,
end_date=end_date,
index_col=0,
parse_dates={'datetime': ['date', 'hour', 'minute']},
date_parser=lambda x: datetime.strptime(x, '%Y%m%d %H %M'),
max_results=max_results,
chunksize=10000)
# Concatenate the chunks into a DataFrame and get number of rows
df = pd.concat(df_chunks)
num_rows = df.shape[0]
# Resample into half-hour buckets
df = df.resample('30Min', how='sum')
# Create the behavior table (half-hour x weekday)
grouped = df.groupby([df.index.time, df.index.weekday])
behavior = grouped['pageviews'].aggregate(np.sum).unstack()
# Make sure the table covers all hours and weekdays
behavior = behavior.reindex(index=pd.date_range("00:00", "23:30", freq="30min").time,
columns=range(7))
behavior.columns = ['MO', 'TU', 'WE', 'TH', 'FR', 'SA', 'SU']
return behavior, num_rows
def test_getdata(self):
try:
import httplib2
from pandas.io.ga import GAnalytics, read_ga
from pandas.io.auth import AuthenticationConfigError
except ImportError:
raise nose.SkipTest
try:
end_date = datetime.now()
start_date = end_date - pd.offsets.Day() * 5
end_date = end_date.strftime('%Y-%m-%d')
start_date = start_date.strftime('%Y-%m-%d')
reader = GAnalytics()
df = reader.get_data(metrics=[
'avgTimeOnSite', 'visitors', 'newVisits', 'pageviewsPerVisit'
],
start_date=start_date,
end_date=end_date,
dimensions=['date', 'hour'],
parse_dates={'ts': ['date', 'hour']})
assert isinstance(df, DataFrame)
assert isinstance(df.index, pd.DatetimeIndex)
assert len(df) > 1
assert 'date' not in df
assert 'hour' not in df
assert df.index.name == 'ts'
assert 'avgTimeOnSite' in df
assert 'visitors' in df
assert 'newVisits' in df
assert 'pageviewsPerVisit' in df
df2 = read_ga(metrics=[
'avgTimeOnSite', 'visitors', 'newVisits', 'pageviewsPerVisit'
],
start_date=start_date,
end_date=end_date,
dimensions=['date', 'hour'],
parse_dates={'ts': ['date', 'hour']})
assert_frame_equal(df, df2)
except AuthenticationConfigError:
raise nose.SkipTest
except httplib2.ServerNotFoundError:
try:
h = httplib2.Http()
response, content = h.request("http://www.google.com")
raise
except httplib2.ServerNotFoundError:
raise nose.SkipTest
def test_segment(self):
try:
end_date = datetime.now()
start_date = end_date - pd.offsets.Day() * 5
end_date = end_date.strftime('%Y-%m-%d')
start_date = start_date.strftime('%Y-%m-%d')
reader = GAnalytics()
df = reader.get_data(metrics=[
'avgTimeOnSite', 'visitors', 'newVisits', 'pageviewsPerVisit'
],
start_date=start_date,
end_date=end_date,
segment=-2,
dimensions=['date', 'hour'],
parse_dates={'ts': ['date', 'hour']},
index_col=0)
self.assertIsInstance(df, pd.DataFrame)
self.assertIsInstance(df.index, pd.DatetimeIndex)
self.assertGreater(len(df), 1)
self.assertTrue('date' not in df)
self.assertTrue('hour' not in df)
self.assertEqual(df.index.name, 'ts')
self.assertTrue('avgTimeOnSite' in df)
self.assertTrue('visitors' in df)
self.assertTrue('newVisits' in df)
self.assertTrue('pageviewsPerVisit' in df)
# dynamic
df = read_ga(metrics=[
'avgTimeOnSite', 'visitors', 'newVisits', 'pageviewsPerVisit'
],
start_date=start_date,
end_date=end_date,
segment="source=~twitter",
dimensions=['date', 'hour'],
parse_dates={'ts': ['date', 'hour']},
index_col=0)
assert isinstance(df, pd.DataFrame)
assert isinstance(df.index, pd.DatetimeIndex)
self.assertGreater(len(df), 1)
self.assertTrue('date' not in df)
self.assertTrue('hour' not in df)
self.assertEqual(df.index.name, 'ts')
self.assertTrue('avgTimeOnSite' in df)
self.assertTrue('visitors' in df)
self.assertTrue('newVisits' in df)
self.assertTrue('pageviewsPerVisit' in df)
except AuthenticationConfigError:
raise nose.SkipTest("authentication error")
def test_getdata(self):
try:
import httplib2
from pandas.io.ga import GAnalytics, read_ga
from pandas.io.auth import AuthenticationConfigError
except ImportError:
raise nose.SkipTest
try:
end_date = datetime.now()
start_date = end_date - pd.offsets.Day() * 5
end_date = end_date.strftime('%Y-%m-%d')
start_date = start_date.strftime('%Y-%m-%d')
reader = GAnalytics()
df = reader.get_data(
metrics=['avgTimeOnSite', 'visitors', 'newVisits',
'pageviewsPerVisit'],
start_date=start_date,
end_date=end_date,
dimensions=['date', 'hour'],
parse_dates={'ts': ['date', 'hour']})
assert isinstance(df, DataFrame)
assert isinstance(df.index, pd.DatetimeIndex)
assert len(df) > 1
assert 'date' not in df
assert 'hour' not in df
assert df.index.name == 'ts'
assert 'avgTimeOnSite' in df
assert 'visitors' in df
assert 'newVisits' in df
assert 'pageviewsPerVisit' in df
df2 = read_ga(
metrics=['avgTimeOnSite', 'visitors', 'newVisits',
'pageviewsPerVisit'],
start_date=start_date,
end_date=end_date,
dimensions=['date', 'hour'],
parse_dates={'ts': ['date', 'hour']})
assert_frame_equal(df, df2)
except AuthenticationConfigError:
raise nose.SkipTest
except httplib2.ServerNotFoundError:
try:
h = httplib2.Http()
response, content = h.request("http://www.google.com")
raise
except httplib2.ServerNotFoundError:
raise nose.SkipTest
def test_segment(self):
try:
end_date = datetime.now()
start_date = end_date - pd.offsets.Day() * 5
end_date = end_date.strftime('%Y-%m-%d')
start_date = start_date.strftime('%Y-%m-%d')
reader = GAnalytics()
df = reader.get_data(
metrics=['avgTimeOnSite', 'visitors', 'newVisits',
'pageviewsPerVisit'],
start_date=start_date,
end_date=end_date,
segment=-2,
dimensions=['date', 'hour'],
parse_dates={'ts': ['date', 'hour']},
index_col=0)
self.assertIsInstance(df, pd.DataFrame)
self.assertIsInstance(df.index, pd.DatetimeIndex)
self.assertGreater(len(df), 1)
self.assertTrue('date' not in df)
self.assertTrue('hour' not in df)
self.assertEqual(df.index.name, 'ts')
self.assertTrue('avgTimeOnSite' in df)
self.assertTrue('visitors' in df)
self.assertTrue('newVisits' in df)
self.assertTrue('pageviewsPerVisit' in df)
# dynamic
df = read_ga(
metrics=['avgTimeOnSite', 'visitors', 'newVisits',
'pageviewsPerVisit'],
start_date=start_date,
end_date=end_date,
segment="source=~twitter",
dimensions=['date', 'hour'],
parse_dates={'ts': ['date', 'hour']},
index_col=0)
assert isinstance(df, pd.DataFrame)
assert isinstance(df.index, pd.DatetimeIndex)
self.assertGreater(len(df), 1)
self.assertTrue('date' not in df)
self.assertTrue('hour' not in df)
self.assertEqual(df.index.name, 'ts')
self.assertTrue('avgTimeOnSite' in df)
self.assertTrue('visitors' in df)
self.assertTrue('newVisits' in df)
self.assertTrue('pageviewsPerVisit' in df)
except AuthenticationConfigError:
raise nose.SkipTest("authentication error")
def test_segment(self):
try:
end_date = datetime.now()
start_date = end_date - pd.offsets.Day() * 5
end_date = end_date.strftime('%Y-%m-%d')
start_date = start_date.strftime('%Y-%m-%d')
reader = GAnalytics()
df = reader.get_data(
metrics=['avgTimeOnSite', 'visitors', 'newVisits',
'pageviewsPerVisit'],
start_date=start_date,
end_date=end_date,
segment=-2,
dimensions=['date', 'hour'],
parse_dates={'ts': ['date', 'hour']})
assert isinstance(df, DataFrame)
assert isinstance(df.index, pd.DatetimeIndex)
assert len(df) > 1
assert 'date' not in df
assert 'hour' not in df
assert df.index.name == 'ts'
assert 'avgTimeOnSite' in df
assert 'visitors' in df
assert 'newVisits' in df
assert 'pageviewsPerVisit' in df
#dynamic
df = read_ga(
metrics=['avgTimeOnSite', 'visitors', 'newVisits',
'pageviewsPerVisit'],
start_date=start_date,
end_date=end_date,
segment="source=~twitter",
dimensions=['date', 'hour'],
parse_dates={'ts': ['date', 'hour']})
assert isinstance(df, DataFrame)
assert isinstance(df.index, pd.DatetimeIndex)
assert len(df) > 1
assert 'date' not in df
assert 'hour' not in df
assert df.index.name == 'ts'
assert 'avgTimeOnSite' in df
assert 'visitors' in df
assert 'newVisits' in df
assert 'pageviewsPerVisit' in df
except AuthenticationConfigError:
raise nose.SkipTest
def test_segment(self):
try:
end_date = datetime.now()
start_date = end_date - pd.offsets.Day() * 5
end_date = end_date.strftime('%Y-%m-%d')
start_date = start_date.strftime('%Y-%m-%d')
reader = GAnalytics()
df = reader.get_data(metrics=[
'avgTimeOnSite', 'visitors', 'newVisits', 'pageviewsPerVisit'
],
start_date=start_date,
end_date=end_date,
segment=-2,
dimensions=['date', 'hour'],
parse_dates={'ts': ['date', 'hour']})
assert isinstance(df, DataFrame)
assert isinstance(df.index, pd.DatetimeIndex)
assert len(df) > 1
assert 'date' not in df
assert 'hour' not in df
assert df.index.name == 'ts'
assert 'avgTimeOnSite' in df
assert 'visitors' in df
assert 'newVisits' in df
assert 'pageviewsPerVisit' in df
#dynamic
df = read_ga(metrics=[
'avgTimeOnSite', 'visitors', 'newVisits', 'pageviewsPerVisit'
],
start_date=start_date,
end_date=end_date,
segment="source=~twitter",
dimensions=['date', 'hour'],
parse_dates={'ts': ['date', 'hour']})
assert isinstance(df, DataFrame)
assert isinstance(df.index, pd.DatetimeIndex)
assert len(df) > 1
assert 'date' not in df
assert 'hour' not in df
assert df.index.name == 'ts'
assert 'avgTimeOnSite' in df
assert 'visitors' in df
assert 'newVisits' in df
assert 'pageviewsPerVisit' in df
except AuthenticationConfigError:
raise nose.SkipTest
def behavior(start_date, end_date, account_name, property_name, profile_name,
max_results):
"""
Writes a DataFrame with the number of pageviews per half-hours x weekdays
to the Range "behavior"
"""
# Let pandas fetch the data from Google Analytics, returns a generator object
df_chunks = ga.read_ga(
secrets=client_secrets,
account_name=account_name,
property_name=property_name,
profile_name=profile_name,
dimensions=['date', 'hour', 'minute'],
metrics=['pageviews'],
start_date=start_date,
end_date=end_date,
index_col=0,
parse_dates={'datetime': ['date', 'hour', 'minute']},
date_parser=lambda x: datetime.strptime(x, '%Y%m%d %H %M'),
max_results=max_results,
chunksize=10000)
# Concatenate the chunks into a DataFrame and get number of rows
df = pd.concat(df_chunks)
num_rows = df.shape[0]
# Resample into half-hour buckets
df = df.resample('30Min', how='sum')
# Create the behavior table (half-hour x weekday)
grouped = df.groupby([df.index.time, df.index.weekday])
behavior = grouped['pageviews'].aggregate(np.sum).unstack()
# Make sure the table covers all hours and weekdays
behavior = behavior.reindex(index=pd.date_range("00:00",
"23:30",
freq="30min").time,
columns=range(7))
behavior.columns = ['MO', 'TU', 'WE', 'TH', 'FR', 'SA', 'SU']
# Write to Excel.
# Time-only values are currently a bit of a pain on Windows, so we set index=False.
Range(sheet_dashboard, 'behavior', index=False).value = behavior
Range(sheet_dashboard, 'effective').value = num_rows
def behavior(start_date, end_date, account_name, property_name, profile_name, max_results):
"""
Writes a DataFrame with the number of pageviews per half-hours x weekdays
to the Range "behavior"
"""
# Let pandas fetch the data from Google Analytics, returns a generator object
df_chunks = ga.read_ga(secrets=client_secrets,
account_name=account_name,
property_name=property_name,
profile_name=profile_name,
dimensions=['date', 'hour', 'minute'],
metrics=['pageviews'],
start_date=start_date,
end_date=end_date,
index_col=0,
parse_dates={'datetime': ['date', 'hour', 'minute']},
date_parser=lambda x: datetime.strptime(x, '%Y%m%d %H %M'),
max_results=max_results,
chunksize=10000)
# Concatenate the chunks into a DataFrame and get number of rows
df = pd.concat(df_chunks)
num_rows = df.shape[0]
# Resample into half-hour buckets
df = df.resample('30Min', how='sum')
# Create the behavior table (half-hour x weekday)
grouped = df.groupby([df.index.time, df.index.weekday])
behavior = grouped['pageviews'].aggregate(np.sum).unstack()
# Make sure the table covers all hours and weekdays
behavior = behavior.reindex(index=pd.date_range("00:00", "23:30", freq="30min").time,
columns=range(7))
behavior.columns = ['MO', 'TU', 'WE', 'TH', 'FR', 'SA', 'SU']
# Write to Excel.
# Time-only values are currently a bit of a pain on Windows, so we set index=False.
Range(sheet_dashboard, 'behavior').options(index=False).value = behavior
Range(sheet_dashboard, 'effective').value = num_rows
#define the dimensions and metrics to use on the analytics api
dimensions = ['pagePath']
metrics = [
'pageviews', 'uniquePageviews', 'entrances', 'bounceRate', 'exitRate'
]
start_date = '2014-08-01'
end_date = '2014-08-31'
df1 = pd.DataFrame(columns=[
'pageviews'
]) #dataframe with one series [column] - nameing the column pageviews
count = 0 #create a counter variable
for i in df[
0]: #loop through the dataframe df column to pass each slug into the analytics filter field separately
hwy = ga.read_ga(
metrics=metrics, #api call
dimensions=dimensions,
start_date=start_date,
end_date=end_date,
filters=['pagePath==' + i],
account_id='26179049')
df1 = df1.append(
hwy
) #each time the api is called a dataframe [hwy] is created - we append this dataframe to df1 on each pass of the loop
count += 1 #counts the number of times the loop is run through
print count #prints out the number of times the loop is run through
print df1.head() #checks the first 5 results in the dataframe
def PageData(page, mod):
'''
args - pagepath for filter
- Model type - takes 'LR' for LASSO and 'RF' for Random Forests
'''
Store = []
t = datetime.today()
t2 = t - timedelta(hours=1) #(2 hours for BST, 1 for UTC which is on Heroku server)
delay = t2.strftime('%Y-%m-%d %H:00')
star = t - timedelta(30)
max_results = 5e7
metrics = ['pageviews']
dimensions = ['pagePath', 'hour', 'date']
dim = ['date', 'hour']
if page != None:
filters = ['pagePath=='+page]
else:
filters = None
df1 = ga.read_ga(metrics,
dimensions = dim,
start_date = star,
end_date = delay,
parse_dates=[['date', 'hour']],
token_file_name = 'static/token/analytics.dat',
secrets = 'static/token/client_secrets.json',
account_id = '26179049',
filters = filters
)
##################### 48 MAX LAG ##############################
ind = []
for i in range(72, len(df1)):
lag = [1,2,3,4,5,10,17,24,48,72]
lagx = list(i - np.array(lag))
Train = df1.ix[lagx]
Target = df1.ix[i]
TT = Train.T
TT.columns = lag
TT['Target'] = Target['pageviews']
ind.append(TT)
rng = date_range(df1.index[lag[-1]], df1.index[len(df1)-1], freq='H')
Set = ind[0].append(ind[1:])
Set.index = rng
SetT = Set.ix[:delay]
print SetT
#############################################################
li = []
if mod == 'LR':
cnt = 1
else:
cnt = 3
feats = len(lag)
SetT = SetT.replace(0,1)
X_Train = SetT[SetT.columns[0:feats]][:-170]
Y_Train = SetT['Target'].ix[:-170]
X_Test = SetT[SetT.columns[0:feats]][-170:]
Y_Test = SetT['Target'][-170:]
Store.append(X_Train)
Store.append(Y_Train)
Store.append(X_Test)
Store.append(Y_Test)
for j in range(0,cnt):
print j
#Train Model
# feats = len(lag)
# SetT = SetT.replace(0,1)
#
# X_Train = SetT[SetT.columns[0:feats]][:-170]
# Y_Train = SetT['Target'].ix[:-170]
#
#
# X_Test = SetT[SetT.columns[0:feats]][-170:]
# Y_Test = SetT['Target'][-170:]
if mod == 'RF':
print 50*'-'
print "Random Forest Regression"
print 50*'-'
rf = RandomForestRegressor(n_estimators=500, max_features=feats)
else:
print 50*'-'
print "LASSO Regression"
print 50*'-'
################################################################################
# Lasso with path and cross-validation using LassoCV path
from sklearn.linear_model import LassoCV
#
lasso_cv = LassoCV()
#
y_ = lasso_cv.fit(X_Train, Y_Train)
rf = y_
# rf = linear_model.LinearRegression()
rf.fit(X_Train, Y_Train)
PredRF = rf.predict(X_Test)
scoreRF = r2_score(Y_Test,PredRF)
MSE = np.mean(np.square(Y_Test-PredRF))
print 'R2 Score = ', scoreRF
print 'MSE = ', MSE
Res = pd.DataFrame(columns=['res'], index = (range(0,len(Y_Test))))
resid = Y_Test-PredRF
Res['res'] = resid.values
TSDU = Res['res'].mean()+3*Res['res'].std()
TSDL = Res['res'].mean()-3*Res['res'].std()
tsdP = Res[Res['res']>(Res['res'].mean()+3*Res['res'].std())]
tsdN = Res[Res['res']<(Res['res'].mean()-3*Res['res'].std())]
Stats = pd.DataFrame(columns=['yt','pred','resid','TSDU','TSDL'], index=X_Test.index)
Stats['yt'] = Y_Test
Stats['pred'] = PredRF
Stats['resid'] = resid
Stats['TSDU'] = TSDU
Stats['TSDL'] = TSDL
######### Plotting diabled for heroku build ############################
# plt.figure(5)
#
# plt.subplot(2, 1, 1)
# Stats['yt'].plot()
# plt.scatter(Stats['pred'].index, Stats['pred'], s=70, alpha=0.5, c='r')
#
# plt.title("Random Forest Model")
#
# plt.subplot(2,1,2)
#######################################################################
# Stats['resid'].plot()
# Stats['TSDU'].plot(c='r')
# Stats['TSDL'].plot(c='r')
Stats.index.name = 'Time'
Stats['time'] = Stats.index
Stats['pred'].astype('int')
Stats['resid'].astype('int')
Stats['TSDU'].astype('int')
Stats['TSDL'].astype('int')
li.append(Stats)
# plt.title('Residuals and 2 s.d. lines')
cat = pd.concat(([i for i in li]))
Stats = cat.groupby(cat.index).mean()
Stats['time'] = Stats.index
AP = len(tsdP) + len(tsdN)
print 50*'-'
# print "PAGE: " + str(filters[0])
print 50*'-'
print "RANDOM FOREST Number of Anomalous Points is: %i" % AP
print "%% Anomalous points is: %.1f" % (100*AP/len(Y_Test)) +'%'
return Stats, Store, scoreRF
def Trial():
t = datetime.today()
yesterday = t - timedelta(0)
dbyy = t - timedelta(90)
start = dbyy.strftime('%Y-%m-%d')
today = t.strftime('%Y-%m-%d %H:00')
# end = yesterday.strftime('%Y-%m-%d')
start = "2014-06-07"
end = "2014-06-10"
top100 = [u'/jobsearch', u'/search', u'/', u'/tax-disc', u'/renew-adult-passport', u'/student-finance-register-login', u'/visas-immigration', u'/driving-transaction-finished', u'/browse/abroad/passports', u'/apply-uk-visa', u'/browse/driving', u'/check-uk-visa', u'/get-a-passport-urgently', u'/apply-renew-passport', u'/government/organisations/uk-visas-and-immigration', u'/book-practical-driving-test', u'/bank-holidays', u'/change-date-practical-driving-test', u'/government/organisations/driver-and-vehicle-licensing-agency', u'/contact-jobcentre-plus', u'/book-a-driving-theory-test', u'/browse/driving/driving-licences', u'/benefits-calculators', u'/check-uk-visa/y', u'/jobseekers-allowance/how-to-claim', u'/national-minimum-wage-rates', u'/tax-credits-calculator', u'/browse/benefits/tax-credits', u'/browse/benefits', u'/change-address-driving-licence', u'/contact-the-dvla', u'/browse/working/finding-job', u'/calculate-state-pension', u'/passport-fees', u'/browse/working', u'/contact/hm-revenue-customs/tax-credits-enquiries', u'/overseas-passports', u'/track-passport-application', u'/renew-driving-licence', u'/browse/abroad', u'/get-vehicle-information-from-dvla', u'/apply-first-provisional-driving-licence', u'/student-finance/overview', u'/browse/driving/car-tax-discs', u'/general-visit-visa', u'/apply-online-to-replace-a-driving-licence', u'/government/organisations/hm-passport-office', u'/check-mot-status', u'/uk-border-control', u'/get-a-child-passport', u'/practise-your-driving-theory-test', u'/renewing-your-tax-credits-claim', u'/renewtaxcredits', u'/calculate-state-pension/y', u'/student-finance', u'/photos-for-passports', u'/contact-student-finance-england', u'/visa-processing-times', u'/foreign-travel-advice', u'/jobseekers-allowance', u'/contact', u'/browse/education/student-finance', u'/calculate-vehicle-tax-rates', u'/find-a-visa-application-centre', u'/working-tax-credit', u'/renew-driving-licence-at-70', u'/passport-advice-line', u'/call-charges', u'/overseas-passports/y', u'/countersigning-passport-applications', u'/government/topical-events/sexual-violence-in-conflict', u'/how-the-post-office-check-and-send-service-works', u'/visa-fees', u'/government/organisations', u'/browse/driving/learning-to-drive', u'/browse/working/state-pension', u'/vehicle-tax-rate-tables', u'/get-a-child-passport/your-childs-first-passport', u'/calculate-state-pension/y/age', u'/make-a-sorn', u'/jobseekers-allowance/what-youll-get', u'/general-visit-visa/apply', u'/contact/govuk/anonymous-feedback/thankyou', u'/browse/citizenship/citizenship', u'/general-visit-visa/documents-you-must-provide', u'/jobseekers-allowance/overview', u'/uk-border-control/before-you-leave-for-the-uk', u'/government/organisations/foreign-commonwealth-office', u'/government/collections/national-curriculum', u'/government/organisations/ministry-of-defence', u'/ips-regional-passport-office', u'/hand-luggage-restrictions/overview', u'/jobseekers-allowance/eligibility', u'/register-to-vote', u'/disclosure-barring-service-check/overview', u'/browse/benefits/jobseekers-allowance', u'/dvlaforms', u'/tier-4-general-visa', u'/student-finance/loans-and-grants']
# the above should not be hardcoded. Calculate each time so as to avoid page name changes etc.
max_results = 5e7
metrics = ['pageviews']
dimensions = ['pagePath', 'hour', 'date']
dim = ['date', 'hour']
filters = ['pagePath=='+top100[97]]
###############
#Find Top 100 pages by pageviews - (pv fine in this case rather than upv)
# df = ga.read_ga(metrics,
# dim,
# start_date = start,
# end_date = end,
# token_file_name = 'static/token/analytics.dat',
# secrets = 'static/token/client_secrets.json',
# account_id = '26179049',
# max_results=max_results,
# chunksize=5000
# )
# df1b = pd.concat([x for x in df])
# df1c = df1b.sort(columns=['pageviews'], ascending=0)
df1 = ga.read_ga(metrics,
dimensions = dim,
start_date = dbyy,
end_date = yesterday,
parse_dates=[['date', 'hour']],
token_file_name = 'static/token/analytics.dat',
secrets = 'static/token/client_secrets.json',
account_id = '26179049'
# filters = filters
)
##################### 48 MAX LAG ##############################
ind = []
for i in range(48, len(df1)):
lag = [1,2,3,24,48]
lagx = list(i - np.array(lag))
Train = df1.ix[lagx]
Target = df1.ix[i]
TT = Train.T
TT.columns = [1,2,3,24,48]
TT['Target'] = Target['pageviews']
ind.append(TT)
rng = date_range(df1.index[48], df1.index[len(df1)-1], freq='H')
Set = ind[0].append(ind[1:])
Set.index = rng
SetT = Set.ix[:today][:-1]
print SetT
##################### 7 day trial ##############################
# ind = []
#
# for i in range(168, len(df1)):
#
# lag = [1,2,3,24,48, 168]
# lagx = list(i - np.array(lag))
#
# Train = df1.ix[lagx]
# Target = df1.ix[i]
#
# TT = Train.T
# TT.columns = [1,2,3,24,48, 168]
# TT['Target'] = Target['pageviews']
# ind.append(TT)
#
# rng = date_range(df1.index[168], df1.index[len(df1)-1], freq='H')
# Set = ind[0].append(ind[1:])
# Set.index = rng
# SetT = Set.ix[:today][:-1]
# print SetT
#################################################
TrainSamp = 0.8
feats = 5
TS = int(np.round(TrainSamp*len(SetT)))
X_Train = SetT[SetT.columns[0:feats]].head(TS)
Y_Train = SetT['Target'].head(TS)
X_Test = SetT[SetT.columns[0:feats]].ix[TS:]
Y_Test = SetT['Target'].ix[TS:]
X_Train = X_Train.replace(0,1)
X_Test = X_Test.replace(0,1)
Y_Train = Y_Train.replace(0,1)
Y_Test = Y_Test.replace(0,1)
print 50*'-'
print "Random Forest Regression"
print 50*'-'
rf = RandomForestRegressor(n_estimators=500, max_features=feats)
rf.fit(X_Train, Y_Train)
PredRF = rf.predict(X_Test)
scoreRF = r2_score(Y_Test,PredRF)
MSE = np.mean(np.square(Y_Test-PredRF))
print 'R2 Score = ', scoreRF
print 'MSE = ', MSE
Res = pd.DataFrame(columns=['res'], index = (range(0,len(Y_Test))))
resid = Y_Test-PredRF
Res['res'] = resid.values
TSDU = Res['res'].mean()+3*Res['res'].std()
TSDL = Res['res'].mean()-3*Res['res'].std()
tsdP = Res[Res['res']>(Res['res'].mean()+3*Res['res'].std())]
tsdN = Res[Res['res']<(Res['res'].mean()-3*Res['res'].std())]
plt.figure(2)
plt.plot(Y_Test, Y_Test)
plt.scatter(Y_Test,PredRF, s=40, alpha=0.5, c='r')
##############################
plt.figure(1)
plt.subplot(2, 2, 1)
plt.plot(range(0,len(Y_Test)), Y_Test)
plt.scatter(range(0,len(Y_Test)), PredRF, s=70, alpha=0.5, c='r')
plt.xlim([0,len(Y_Test)])
plt.title("Random Forest Model")
plt.subplot(2,2,3)
plt.plot(range(0,len(Y_Test)),resid)
plt.plot(range(0,len(Y_Test)), [TSDU]*len(Y_Test), c='r')
plt.plot(range(0,len(Y_Test)), [TSDL]*len(Y_Test), c='r')
plt.xlim([0,len(Y_Test)])
###############################
AP = len(tsdP) + len(tsdN)
print 50*'-'
print "PAGE: " + str(filters[0])
print 50*'-'
print "RANDOM FOREST Number of Anomalous Points is: %i" % AP
print "%% Anomalous points is: %.1f" % (100*AP/len(Y_Test)) +'%'
#################################################################
print 50*'-'
print "Linear Model"
print 50*'-'
clf = linear_model.LinearRegression()
clf.fit(np.log(X_Train), np.log(Y_Train))
LMPred = clf.predict(np.log(X_Test))
scoreLM = r2_score(np.log(Y_Test),LMPred)
MSELM = np.mean(np.square(Y_Test-np.exp(LMPred)))
print "R2 Score = ", scoreLM
print "MSE = ", MSELM
###########################################################
print 50*'-'
print "LASSO Regression"
print 50*'-'
################################################################################
# Lasso with path and cross-validation using LassoCV path
from sklearn.linear_model import LassoCV
lasso_cv = LassoCV()
y_ = lasso_cv.fit(X_Train, Y_Train).predict(X_Test)
# .predict(X_Test)
print "Optimal regularization parameter = %s" % lasso_cv.alpha_
# Compute explained variance on test data
print "r^2 on test data : %f" % (1 - np.linalg.norm(Y_Test - y_)**2
/ np.linalg.norm(Y_Test)**2)
print "LASSO MSE: ", np.mean(np.square((Y_Test-(y_))))
Res2 = pd.DataFrame(columns=['res'], index = (range(0,len(Y_Test))))
resid2 = Y_Test-y_
Res2['res'] = resid2.values
TSDU2 = Res2['res'].mean()+3*Res2['res'].std()
TSDL2 = Res2['res'].mean()-3*Res2['res'].std()
tsdP2 = Res2[Res2['res']>(Res2['res'].mean()+3*Res2['res'].std())]
tsdN2 = Res2[Res2['res']<(Res2['res'].mean()-3*Res2['res'].std())]
AP = len(tsdP2) + len(tsdN2)
print 50*'-'
print "PAGE: " + str(filters[0])
print 50*'-'
print "LASSP Number of Anomalous Points is: %i" % AP
print "%% Anomalous points is: %.1f" % (100*AP/len(Y_Test)) +'%'
# plt.figure(3) #################################################
plt.subplot(2, 2, 2)
plt.plot(range(0,len(Y_Test)), Y_Test)
plt.scatter(range(0,len(Y_Test)), y_, s=70, alpha=0.5, c='r')
plt.xlim([0,len(Y_Test)])
plt.title('LASSO Model')
plt.subplot(2,2,4)
plt.plot(range(0,len(Y_Test)),resid2)
plt.plot(range(0,len(Y_Test)), [TSDU2]*len(Y_Test), c='r')
plt.plot(range(0,len(Y_Test)), [TSDL2]*len(Y_Test), c='r')
plt.xlim([0,len(Y_Test)])
import pandas.io.ga as ga
# prints top 30 landing pages by pageviews in descending order
df = ga.read_ga(['pageviews','entrances'],
dimensions=['date', 'landingPagePath'],
start_date='2013-04-01')
reset_df = df.reset_index()
sorted_df = reset_df.sort(columns=['pageviews','entrances'], ascending = False)
print (sorted_df.head(30))
start_date = '2016-01-01'
end_date = '2017-01-01'
dimensions = ['date', 'cityId', 'city']
# Metrics for Request
metrics = ['sessions', 'users']
tempStartTime = start_date
count = 0
while str(tempStartTime) <= end_date:
temp_df = ga.read_ga(account_id=account_id,
profile_id=view_id,
property_id=property_id,
metrics=metrics,
dimensions=dimensions,
start_date=tempStartTime,
end_date=today,
index_col=0)
tempStartTime = temp_df.index.max()
if count == 0:
df = temp_df
else:
df = pd.concat([df, temp_df])
count += 1
time.sleep(.2)
# In[22]:
# -*- coding: utf-8 -*-
import pandas.io.ga as ga
import pandas as pd
# goes through profiles,accounts and provides aggregate of visits/pageviews
ids = {'123':'456'}
filters = "source==Facebook";"medium==Social"
all_data = pd.DataFrame()
for profile, account in ids.iteritems():
df = ga.read_ga(['visits', 'pageviews'],
dimensions=['date', 'landingPagePath', 'medium', 'campaign', 'source'],
start_date='2013-03-01', end_date='2013-04-24',
account_id=account, profile_id=profile, filters=filters,
chunksize = 1000)
for d in df:
d['profile'] = profile
d = d.reset_index()
all_data = all_data.append(d)
all_data.to_csv('C:\\tmp\\' + '322.csv', ',', line_terminator = '\n')
def PageData(page, mod):
'''
args - pagepath for filter
- Model type - takes 'LR' for LASSO and 'RF' for Random Forests
'''
Store = []
t = datetime.today()
t2 = t - timedelta(
hours=1) #(2 hours for BST, 1 for UTC which is on Heroku server)
delay = t2.strftime('%Y-%m-%d %H:00')
star = t - timedelta(30)
max_results = 5e7
metrics = ['pageviews']
dimensions = ['pagePath', 'hour', 'date']
dim = ['date', 'hour']
if page != None:
filters = ['pagePath==' + page]
else:
filters = None
df1 = ga.read_ga(metrics,
dimensions=dim,
start_date=star,
end_date=delay,
parse_dates=[['date', 'hour']],
token_file_name='static/token/analytics.dat',
secrets='static/token/client_secrets.json',
account_id='26179049',
filters=filters)
##################### 48 MAX LAG ##############################
ind = []
for i in range(72, len(df1)):
lag = [1, 2, 3, 4, 5, 10, 17, 24, 48, 72]
lagx = list(i - np.array(lag))
Train = df1.ix[lagx]
Target = df1.ix[i]
TT = Train.T
TT.columns = lag
TT['Target'] = Target['pageviews']
ind.append(TT)
rng = date_range(df1.index[lag[-1]], df1.index[len(df1) - 1], freq='H')
Set = ind[0].append(ind[1:])
Set.index = rng
SetT = Set.ix[:delay]
print SetT
#############################################################
li = []
if mod == 'LR':
cnt = 1
else:
cnt = 3
feats = len(lag)
SetT = SetT.replace(0, 1)
X_Train = SetT[SetT.columns[0:feats]][:-170]
Y_Train = SetT['Target'].ix[:-170]
X_Test = SetT[SetT.columns[0:feats]][-170:]
Y_Test = SetT['Target'][-170:]
Store.append(X_Train)
Store.append(Y_Train)
Store.append(X_Test)
Store.append(Y_Test)
for j in range(0, cnt):
print j
#Train Model
# feats = len(lag)
# SetT = SetT.replace(0,1)
#
# X_Train = SetT[SetT.columns[0:feats]][:-170]
# Y_Train = SetT['Target'].ix[:-170]
#
#
# X_Test = SetT[SetT.columns[0:feats]][-170:]
# Y_Test = SetT['Target'][-170:]
if mod == 'RF':
print 50 * '-'
print "Random Forest Regression"
print 50 * '-'
rf = RandomForestRegressor(n_estimators=500, max_features=feats)
else:
print 50 * '-'
print "LASSO Regression"
print 50 * '-'
################################################################################
# Lasso with path and cross-validation using LassoCV path
from sklearn.linear_model import LassoCV
#
lasso_cv = LassoCV()
#
y_ = lasso_cv.fit(X_Train, Y_Train)
rf = y_
# rf = linear_model.LinearRegression()
rf.fit(X_Train, Y_Train)
PredRF = rf.predict(X_Test)
scoreRF = r2_score(Y_Test, PredRF)
MSE = np.mean(np.square(Y_Test - PredRF))
print 'R2 Score = ', scoreRF
print 'MSE = ', MSE
Res = pd.DataFrame(columns=['res'], index=(range(0, len(Y_Test))))
resid = Y_Test - PredRF
Res['res'] = resid.values
TSDU = Res['res'].mean() + 3 * Res['res'].std()
TSDL = Res['res'].mean() - 3 * Res['res'].std()
tsdP = Res[Res['res'] > (Res['res'].mean() + 3 * Res['res'].std())]
tsdN = Res[Res['res'] < (Res['res'].mean() - 3 * Res['res'].std())]
Stats = pd.DataFrame(columns=['yt', 'pred', 'resid', 'TSDU', 'TSDL'],
index=X_Test.index)
Stats['yt'] = Y_Test
Stats['pred'] = PredRF
Stats['resid'] = resid
Stats['TSDU'] = TSDU
Stats['TSDL'] = TSDL
######### Plotting diabled for heroku build ############################
# plt.figure(5)
#
# plt.subplot(2, 1, 1)
# Stats['yt'].plot()
# plt.scatter(Stats['pred'].index, Stats['pred'], s=70, alpha=0.5, c='r')
#
# plt.title("Random Forest Model")
#
# plt.subplot(2,1,2)
#######################################################################
# Stats['resid'].plot()
# Stats['TSDU'].plot(c='r')
# Stats['TSDL'].plot(c='r')
Stats.index.name = 'Time'
Stats['time'] = Stats.index
Stats['pred'].astype('int')
Stats['resid'].astype('int')
Stats['TSDU'].astype('int')
Stats['TSDL'].astype('int')
li.append(Stats)
# plt.title('Residuals and 2 s.d. lines')
cat = pd.concat(([i for i in li]))
Stats = cat.groupby(cat.index).mean()
Stats['time'] = Stats.index
AP = len(tsdP) + len(tsdN)
print 50 * '-'
# print "PAGE: " + str(filters[0])
print 50 * '-'
print "RANDOM FOREST Number of Anomalous Points is: %i" % AP
print "%% Anomalous points is: %.1f" % (100 * AP / len(Y_Test)) + '%'
return Stats, Store, scoreRF
def Trial():
t = datetime.today()
yesterday = t - timedelta(0)
dbyy = t - timedelta(90)
start = dbyy.strftime('%Y-%m-%d')
today = t.strftime('%Y-%m-%d %H:00')
# end = yesterday.strftime('%Y-%m-%d')
start = "2014-06-07"
end = "2014-06-10"
top100 = [
u'/jobsearch', u'/search', u'/', u'/tax-disc',
u'/renew-adult-passport', u'/student-finance-register-login',
u'/visas-immigration', u'/driving-transaction-finished',
u'/browse/abroad/passports', u'/apply-uk-visa', u'/browse/driving',
u'/check-uk-visa', u'/get-a-passport-urgently',
u'/apply-renew-passport',
u'/government/organisations/uk-visas-and-immigration',
u'/book-practical-driving-test', u'/bank-holidays',
u'/change-date-practical-driving-test',
u'/government/organisations/driver-and-vehicle-licensing-agency',
u'/contact-jobcentre-plus', u'/book-a-driving-theory-test',
u'/browse/driving/driving-licences', u'/benefits-calculators',
u'/check-uk-visa/y', u'/jobseekers-allowance/how-to-claim',
u'/national-minimum-wage-rates', u'/tax-credits-calculator',
u'/browse/benefits/tax-credits', u'/browse/benefits',
u'/change-address-driving-licence', u'/contact-the-dvla',
u'/browse/working/finding-job', u'/calculate-state-pension',
u'/passport-fees', u'/browse/working',
u'/contact/hm-revenue-customs/tax-credits-enquiries',
u'/overseas-passports', u'/track-passport-application',
u'/renew-driving-licence', u'/browse/abroad',
u'/get-vehicle-information-from-dvla',
u'/apply-first-provisional-driving-licence',
u'/student-finance/overview', u'/browse/driving/car-tax-discs',
u'/general-visit-visa', u'/apply-online-to-replace-a-driving-licence',
u'/government/organisations/hm-passport-office', u'/check-mot-status',
u'/uk-border-control', u'/get-a-child-passport',
u'/practise-your-driving-theory-test',
u'/renewing-your-tax-credits-claim', u'/renewtaxcredits',
u'/calculate-state-pension/y', u'/student-finance',
u'/photos-for-passports', u'/contact-student-finance-england',
u'/visa-processing-times', u'/foreign-travel-advice',
u'/jobseekers-allowance', u'/contact',
u'/browse/education/student-finance', u'/calculate-vehicle-tax-rates',
u'/find-a-visa-application-centre', u'/working-tax-credit',
u'/renew-driving-licence-at-70', u'/passport-advice-line',
u'/call-charges', u'/overseas-passports/y',
u'/countersigning-passport-applications',
u'/government/topical-events/sexual-violence-in-conflict',
u'/how-the-post-office-check-and-send-service-works', u'/visa-fees',
u'/government/organisations', u'/browse/driving/learning-to-drive',
u'/browse/working/state-pension', u'/vehicle-tax-rate-tables',
u'/get-a-child-passport/your-childs-first-passport',
u'/calculate-state-pension/y/age', u'/make-a-sorn',
u'/jobseekers-allowance/what-youll-get', u'/general-visit-visa/apply',
u'/contact/govuk/anonymous-feedback/thankyou',
u'/browse/citizenship/citizenship',
u'/general-visit-visa/documents-you-must-provide',
u'/jobseekers-allowance/overview',
u'/uk-border-control/before-you-leave-for-the-uk',
u'/government/organisations/foreign-commonwealth-office',
u'/government/collections/national-curriculum',
u'/government/organisations/ministry-of-defence',
u'/ips-regional-passport-office',
u'/hand-luggage-restrictions/overview',
u'/jobseekers-allowance/eligibility', u'/register-to-vote',
u'/disclosure-barring-service-check/overview',
u'/browse/benefits/jobseekers-allowance', u'/dvlaforms',
u'/tier-4-general-visa', u'/student-finance/loans-and-grants'
]
# the above should not be hardcoded. Calculate each time so as to avoid page name changes etc.
max_results = 5e7
metrics = ['pageviews']
dimensions = ['pagePath', 'hour', 'date']
dim = ['date', 'hour']
filters = ['pagePath==' + top100[97]]
###############
#Find Top 100 pages by pageviews - (pv fine in this case rather than upv)
# df = ga.read_ga(metrics,
# dim,
# start_date = start,
# end_date = end,
# token_file_name = 'static/token/analytics.dat',
# secrets = 'static/token/client_secrets.json',
# account_id = '26179049',
# max_results=max_results,
# chunksize=5000
# )
# df1b = pd.concat([x for x in df])
# df1c = df1b.sort(columns=['pageviews'], ascending=0)
df1 = ga.read_ga(metrics,
dimensions=dim,
start_date=dbyy,
end_date=yesterday,
parse_dates=[['date', 'hour']],
token_file_name='static/token/analytics.dat',
secrets='static/token/client_secrets.json',
account_id='26179049'
# filters = filters
)
##################### 48 MAX LAG ##############################
ind = []
for i in range(48, len(df1)):
lag = [1, 2, 3, 24, 48]
lagx = list(i - np.array(lag))
Train = df1.ix[lagx]
Target = df1.ix[i]
TT = Train.T
TT.columns = [1, 2, 3, 24, 48]
TT['Target'] = Target['pageviews']
ind.append(TT)
rng = date_range(df1.index[48], df1.index[len(df1) - 1], freq='H')
Set = ind[0].append(ind[1:])
Set.index = rng
SetT = Set.ix[:today][:-1]
print SetT
##################### 7 day trial ##############################
# ind = []
#
# for i in range(168, len(df1)):
#
# lag = [1,2,3,24,48, 168]
# lagx = list(i - np.array(lag))
#
# Train = df1.ix[lagx]
# Target = df1.ix[i]
#
# TT = Train.T
# TT.columns = [1,2,3,24,48, 168]
# TT['Target'] = Target['pageviews']
# ind.append(TT)
#
# rng = date_range(df1.index[168], df1.index[len(df1)-1], freq='H')
# Set = ind[0].append(ind[1:])
# Set.index = rng
# SetT = Set.ix[:today][:-1]
# print SetT
#################################################
TrainSamp = 0.8
feats = 5
TS = int(np.round(TrainSamp * len(SetT)))
X_Train = SetT[SetT.columns[0:feats]].head(TS)
Y_Train = SetT['Target'].head(TS)
X_Test = SetT[SetT.columns[0:feats]].ix[TS:]
Y_Test = SetT['Target'].ix[TS:]
X_Train = X_Train.replace(0, 1)
X_Test = X_Test.replace(0, 1)
Y_Train = Y_Train.replace(0, 1)
Y_Test = Y_Test.replace(0, 1)
print 50 * '-'
print "Random Forest Regression"
print 50 * '-'
rf = RandomForestRegressor(n_estimators=500, max_features=feats)
rf.fit(X_Train, Y_Train)
PredRF = rf.predict(X_Test)
scoreRF = r2_score(Y_Test, PredRF)
MSE = np.mean(np.square(Y_Test - PredRF))
print 'R2 Score = ', scoreRF
print 'MSE = ', MSE
Res = pd.DataFrame(columns=['res'], index=(range(0, len(Y_Test))))
resid = Y_Test - PredRF
Res['res'] = resid.values
TSDU = Res['res'].mean() + 3 * Res['res'].std()
TSDL = Res['res'].mean() - 3 * Res['res'].std()
tsdP = Res[Res['res'] > (Res['res'].mean() + 3 * Res['res'].std())]
tsdN = Res[Res['res'] < (Res['res'].mean() - 3 * Res['res'].std())]
plt.figure(2)
plt.plot(Y_Test, Y_Test)
plt.scatter(Y_Test, PredRF, s=40, alpha=0.5, c='r')
##############################
plt.figure(1)
plt.subplot(2, 2, 1)
plt.plot(range(0, len(Y_Test)), Y_Test)
plt.scatter(range(0, len(Y_Test)), PredRF, s=70, alpha=0.5, c='r')
plt.xlim([0, len(Y_Test)])
plt.title("Random Forest Model")
plt.subplot(2, 2, 3)
plt.plot(range(0, len(Y_Test)), resid)
plt.plot(range(0, len(Y_Test)), [TSDU] * len(Y_Test), c='r')
plt.plot(range(0, len(Y_Test)), [TSDL] * len(Y_Test), c='r')
plt.xlim([0, len(Y_Test)])
###############################
AP = len(tsdP) + len(tsdN)
print 50 * '-'
print "PAGE: " + str(filters[0])
print 50 * '-'
print "RANDOM FOREST Number of Anomalous Points is: %i" % AP
print "%% Anomalous points is: %.1f" % (100 * AP / len(Y_Test)) + '%'
#################################################################
print 50 * '-'
print "Linear Model"
print 50 * '-'
clf = linear_model.LinearRegression()
clf.fit(np.log(X_Train), np.log(Y_Train))
LMPred = clf.predict(np.log(X_Test))
scoreLM = r2_score(np.log(Y_Test), LMPred)
MSELM = np.mean(np.square(Y_Test - np.exp(LMPred)))
print "R2 Score = ", scoreLM
print "MSE = ", MSELM
###########################################################
print 50 * '-'
print "LASSO Regression"
print 50 * '-'
################################################################################
# Lasso with path and cross-validation using LassoCV path
from sklearn.linear_model import LassoCV
lasso_cv = LassoCV()
y_ = lasso_cv.fit(X_Train, Y_Train).predict(X_Test)
# .predict(X_Test)
print "Optimal regularization parameter = %s" % lasso_cv.alpha_
# Compute explained variance on test data
print "r^2 on test data : %f" % (
1 - np.linalg.norm(Y_Test - y_)**2 / np.linalg.norm(Y_Test)**2)
print "LASSO MSE: ", np.mean(np.square((Y_Test - (y_))))
Res2 = pd.DataFrame(columns=['res'], index=(range(0, len(Y_Test))))
resid2 = Y_Test - y_
Res2['res'] = resid2.values
TSDU2 = Res2['res'].mean() + 3 * Res2['res'].std()
TSDL2 = Res2['res'].mean() - 3 * Res2['res'].std()
tsdP2 = Res2[Res2['res'] > (Res2['res'].mean() + 3 * Res2['res'].std())]
tsdN2 = Res2[Res2['res'] < (Res2['res'].mean() - 3 * Res2['res'].std())]
AP = len(tsdP2) + len(tsdN2)
print 50 * '-'
print "PAGE: " + str(filters[0])
print 50 * '-'
print "LASSP Number of Anomalous Points is: %i" % AP
print "%% Anomalous points is: %.1f" % (100 * AP / len(Y_Test)) + '%'
# plt.figure(3) #################################################
plt.subplot(2, 2, 2)
plt.plot(range(0, len(Y_Test)), Y_Test)
plt.scatter(range(0, len(Y_Test)), y_, s=70, alpha=0.5, c='r')
plt.xlim([0, len(Y_Test)])
plt.title('LASSO Model')
plt.subplot(2, 2, 4)
plt.plot(range(0, len(Y_Test)), resid2)
plt.plot(range(0, len(Y_Test)), [TSDU2] * len(Y_Test), c='r')
plt.plot(range(0, len(Y_Test)), [TSDL2] * len(Y_Test), c='r')
plt.xlim([0, len(Y_Test)])