def dump_now(cls):
""" Flush hits to Whale and increment """
# Get the incoming hits from Hail
r=cls.hail_driver()
set_number_name = 'hail_number'
r.setnx(set_number_name, 0)
set_number = r.incr(set_number_name) - 1
set_name = 'hail_%s' % set_number
try: keys_from_hail = r.smembers(set_name)
except: return
if not len(keys_from_hail):
r.delete(set_name)
return
def get_keys_from_json(k):
try:
class_name, pk, dimensions, metrics, at = json.loads(r[k])
#at = datetime.datetime.fromtimestamp(float(t))
return (pk, dimensions, metrics, at)
except Exception as e:
print e
return False
keys_to_update = map(get_keys_from_json, keys_from_hail)
for packed in keys_to_update:
if packed:
pk, dimensions, metrics, at = packed
Whale.count_now(pk, dimensions, metrics, at=at)
# Delete the hits
map(r.delete, keys_from_hail)
r.delete(set_name)
def dump_now(cls):
""" Flush hits to Whale and increment """
# Get the incoming hits from Hail
r = cls.hail_driver()
set_number_name = 'hail_number'
r.setnx(set_number_name, 0)
set_number = r.incr(set_number_name) - 1
set_name = 'hail_%s' % set_number
try:
keys_from_hail = r.smembers(set_name)
except:
return
if not len(keys_from_hail):
r.delete(set_name)
return
def get_keys_from_json(k):
try:
class_name, pk, dimensions, metrics, at = json.loads(r[k])
#at = datetime.datetime.fromtimestamp(float(t))
return (pk, dimensions, metrics, at)
except Exception as e:
print e
return False
keys_to_update = map(get_keys_from_json, keys_from_hail)
for packed in keys_to_update:
if packed:
pk, dimensions, metrics, at = packed
Whale.count_now(pk, dimensions, metrics, at=at)
# Delete the hits
map(r.delete, keys_from_hail)
r.delete(set_name)
def dump_now(cls):
""" Flush hits to Whale and increment """
# Get the incoming hits from Hail
from whale import Whale
whale = Whale()
r=cls.driver()
_s_n_n = 'hail_number'
r.setnx(_s_n_n, 0)
set_number = r.incr(_s_n_n) - 1
set_name = 'hail_%s'%set_number
try: keys_from_hail = r.smembers(set_name)
except: return
if len(keys_from_hail) is 0:
r.delete(set_name)
return
def get_keys_from_json(k):
try:
class_name, categories, dimensions, metrics, t = json.loads(r[k])
at = datetime.datetime.fromtimestamp(float(t))
return (categories, dimensions, metrics, at)
except Exception as e:
print e
return False
keys_to_update = map(get_keys_from_json, keys_from_hail)
for packed in keys_to_update:
if not packed: continue
categories, dimensions, metrics, at = packed
whale.count_now(categories, dimensions, metrics, at=at)
# Delete the hits
map(r.delete, keys_to_update)
r.delete(set_name)
class TestHailWHale(unittest.TestCase):
def setUp(self):
from hail import Hail
from whale import Whale
self.hail = Hail()
self.whale = Whale()
def testGetSubdimensions(self):
self.whale.count_now("test", {"a": 1, "b": 2})
subs = self.whale.get_subdimensions("test")
assert ["a"] in subs
assert ["b"] in subs
def testGetAllSubdimensions(self):
self.whale.count_now("test", {"a": 1, "b": 2})
subs = self.whale.all_subdimensions("test")
assert ["a"] in subs
assert ["a", "1"] in subs
assert ["b"] in subs
assert ["b", "2"] in subs
def testCrunch(self):
# Unique key for every test
t = str(time.time())
self.whale.count_now("test_crunch", [t, "a"], {"value": 5})
self.whale.count_now("test_crunch", [t, "b"], {"value": 1})
self.whale.count_now("test_crunch", [t, "c"], {"value": 15})
data = self.whale.crunch("test_crunch", [t], "value")
def count_now():
whale = Whale()
at = g('at', False)
tzoffset = None
if not at:
at = times.now()
else:
from dateutil.parser import parse
at = parse(g('at'))
at = at.replace(tzinfo=None)
val = whale.count_now(at=at, **default_params())
return 'OK'
def tracker():
from periods import Period
import random
params = default_params()
# LOLOL THIS SHOULD REALLY CHANGE
key = hashlib.sha256('hailwhale_weak_key').digest()
if 'pk' not in req.GET and 'pixel' in req.GET:
from Crypto.Cipher import AES
from base64 import b64encode, b64decode
from urllib import quote_plus
mode = AES.MODE_CBC
encryptor = AES.new(key, mode)
text = g('pixel')
INTERRUPT = u'\u0001'
PAD = u'\u0000'
# Since you need to pad your data before encryption,
# create a padding function as well
# Similarly, create a function to strip off the padding after decryption
def AddPadding(data, interrupt, pad, block_size):
new_data = ''.join([data, interrupt])
new_data_len = len(new_data)
remaining_len = block_size - new_data_len
to_pad_len = remaining_len % block_size
pad_string = pad * to_pad_len
return ''.join([new_data, pad_string])
def StripPadding(data, interrupt, pad):
return data.rstrip(pad).rstrip(interrupt)
def hw_encoded(t):
return quote_plus(
b64encode(encryptor.encrypt(AddPadding(t, INTERRUPT, PAD,
32))))
def hw_decoded(t):
return StripPadding(encryptor.decrypt(b64decode(t)), INTERRUPT,
PAD)
params['pk'] = hw_decoded(text)
pk = params['pk']
whale = Whale()
hail = Hail()
val = whale.count_now(at=times.now(), **params)
#val = whale.count_now(**params)
uid = g('uid')
if not uid or uid == '_new':
default = random.randrange(10**6, 10**9)
uid = str(req.get_cookie('uid', str(default), key))
hail.spy_log(uid, params)
response.set_cookie('uid', uid, key)
return str(uid)
def count_now():
whale = Whale()
vals = default_params()
at = vals.get("at")#g('at', False)
tzoffset = None
if not at:
at = times.now()
else:
from dateutil.parser import parse
at = parse(at)
val = whale.count_now(at= at, pk=vals.get("pk"), metrics=vals.get("metrics"), dimensions=vals.get("dimensions"))
return 'OK'
def count_now():
whale = Whale()
vals = default_params()
at = vals.get("at") #g('at', False)
tzoffset = None
if not at:
at = times.now()
else:
from dateutil.parser import parse
at = parse(at)
val = whale.count_now(at=at,
pk=vals.get("pk"),
metrics=vals.get("metrics"),
dimensions=vals.get("dimensions"))
return 'OK'
def tracker():
from periods import Period
import random
params = default_params()
# LOLOL THIS SHOULD REALLY CHANGE
key = hashlib.sha256('hailwhale_weak_key').digest()
if 'pk' not in req.GET and 'pixel' in req.GET:
from Crypto.Cipher import AES
from base64 import b64encode, b64decode
from urllib import quote_plus
mode = AES.MODE_CBC
encryptor = AES.new(key, mode)
text = g('pixel')
INTERRUPT = u'\u0001'
PAD = u'\u0000'
# Since you need to pad your data before encryption,
# create a padding function as well
# Similarly, create a function to strip off the padding after decryption
def AddPadding(data, interrupt, pad, block_size):
new_data = ''.join([data, interrupt])
new_data_len = len(new_data)
remaining_len = block_size - new_data_len
to_pad_len = remaining_len % block_size
pad_string = pad * to_pad_len
return ''.join([new_data, pad_string])
def StripPadding(data, interrupt, pad):
return data.rstrip(pad).rstrip(interrupt)
def hw_encoded(t):
return quote_plus(b64encode(encryptor.encrypt(AddPadding(t, INTERRUPT, PAD, 32))))
def hw_decoded(t):
return StripPadding(encryptor.decrypt(b64decode(t)), INTERRUPT, PAD)
params['pk'] = hw_decoded(text)
pk = params['pk']
whale = Whale()
hail = Hail()
val = whale.count_now(at=times.now(), **params)
#val = whale.count_now(**params)
uid = g('uid')
if not uid or uid == '_new':
default = random.randrange(10**6,10**9)
uid = str(req.get_cookie('uid', str(default), key))
hail.spy_log(uid, params)
response.set_cookie('uid', uid, key)
return str(uid)
class TestHailWhale(unittest.TestCase):
def setUp(self):
from hail import Hail
from whale import Whale
self.hail = Hail()
self.whale = Whale()
def testGetSubdimensions(self):
t = 'subs_%s' % str(time.time())
self.whale.count_now(t, {'a': 1, 'b': 2})
subs = self.whale.get_subdimensions(t)
assert('a' in subs)
assert('b' in subs)
def testGetAllSubdimensions(self):
t = 'all_subs_%s' % str(time.time())
self.whale.count_now(t, {'a': 1, 'b': 2})
subs = self.whale.all_subdimensions(t)
assert('a' in subs)
assert(['a', '1'] in subs)
assert('b' in subs)
assert(['b', '2'] in subs)
def testPlotpoints(self):
t = str(time.time())
for i in range(5):
self.whale.count_now('test_plotpoints', t, {'hits': 1, 'values': 5})
plotpoints = self.whale.plotpoints('test_plotpoints', t, ['hits', 'values'], points_type=list)
self.assertEqual(plotpoints[t]['hits'][-1][1], 5)
self.assertEqual(plotpoints[t]['values'][-1][1], 25)
def testPlotpointsDepth(self):
t = str(time.time())
self.whale.count_now('test_depth', {t: 'a'})
self.whale.count_now('test_depth', {t: 'b'})
self.whale.count_now('test_depth', {t: 'b'})
self.whale.count_now('test_depth', {t: {'c': 'child'}})
# Test 1 level deep
plotpoints = self.whale.plotpoints('test_depth', t, points_type=list, depth=1)
self.assertEqual(plotpoints[maybe_dumps([t, 'a'])]['hits'][-1][1], 1)
self.assertEqual(plotpoints[maybe_dumps([t, 'b'])]['hits'][-1][1], 2)
self.assertEqual(plotpoints[maybe_dumps([t, 'c'])]['hits'][-1][1], 1)
self.assertEqual(False, maybe_dumps([t, 'c', 'child']) in plotpoints)
# Test 2 levels deep
plotpoints = self.whale.plotpoints('test_depth', t, points_type=list, depth=2)
self.assertEqual(True, maybe_dumps([t, 'c', 'child']) in plotpoints)
self.assertEqual(plotpoints[maybe_dumps([t, 'c', 'child'])]['hits'][-1][1], 1)
# Test ranking and limiting
plotpoints = self.whale.plotpoints('test_depth', t, points_type=list,
depth=1, limit=2)
self.assertEqual(plotpoints[maybe_dumps([t, 'b'])]['hits'][-1][1], 2)
self.assertEqual(True, maybe_dumps([t, 'a']) not in plotpoints)
self.assertEqual(True, maybe_dumps([t, 'c']) not in plotpoints)
def testRatioPlotpoints(self):
t = str(time.time())
for i in range(5):
self.whale.count_now('test_ratio', t, {'hit': 1, 'value': 5})
plotpoints = self.whale.plotpoints('test_ratio', t, ['hit', 'value', 'value/hit'], points_type=list)
self.assertEqual(plotpoints[t]['hit'][-1][1], 5)
self.assertEqual(plotpoints[t]['value'][-1][1], 25)
self.assertEqual(plotpoints[t]['value/hit'][-1][1], 5)
def testRankSubdimensionsScalar(self):
t = str(time.time())
self.whale.count_now('test_rank', [t, 'a', 'asub1'], {'value': 1})
self.whale.count_now('test_rank', [t, 'a', 'asub2'], {'value': 30})
self.whale.count_now('test_rank', [t, 'b'], {'value': 80})
self.whale.count_now('test_rank', [t, 'c'], {'value': 10})
ranked = self.whale.rank_subdimensions_scalar('test_rank', t, 'value')
self.assertEqual(ranked[maybe_dumps([t, 'a'])]['important'], False)
self.assertEqual(ranked[maybe_dumps([t, 'a', 'asub1'])]['important'], False)
self.assertEqual(ranked[maybe_dumps([t, 'a', 'asub2'])]['important'], True)
self.assertEqual(ranked[maybe_dumps([t, 'b'])]['important'], True)
self.assertEqual(ranked[maybe_dumps([t, 'c'])]['important'], False)
def testRankSubdimensionsRatio(self):
t = str(time.time())
pk = 'test_ratio_rank'
# OVERALL STATS: 529,994 value, 50,000 visitors, 10.6 value per visitor
# Not important, too close to overall
self.whale.count_now(pk, [t, 'a', 'asub1'],
{'value': 54989, 'visitors': 4999}) # 11 value per visitor
# Important, high relative ratio
self.whale.count_now(pk, [t, 'a', 'asub2'],
{'value': 375000, 'visitors': 25000}) # 15 value per visitor
# Important, low relative ratio
self.whale.count_now(pk, [t, 'b'],
{'value': 100000, 'visitors': 20000}) # 5 value per visitor
# Not important, not enough visitors
self.whale.count_now(pk, [t, 'c'],
{'value': 5, 'visitors': 1}) # 5 value per visitor
one_level = self.whale.rank_subdimensions_ratio('test_rank_ratio', 'value', 'visitors',
t, recursive=False)
all_levels = self.whale.rank_subdimensions_ratio(pk, 'value', 'visitors', t)
self.assertEqual(True, maybe_dumps([t, 'a', 'asub1']) not in one_level)
self.assertEqual(all_levels[maybe_dumps([t, 'a', 'asub1'])]['important'], False)
self.assertEqual(all_levels[maybe_dumps([t, 'a', 'asub2'])]['important'], True)
self.assertEqual(all_levels[maybe_dumps([t, 'b'])]['important'], True)
self.assertEqual(all_levels[maybe_dumps([t, 'c'])]['important'], False)
def testBasicDecision(self):
pk = 'test_basic_decision'
decision = str(time.time())
# Make a decision, any decision, from no information whatsoever
good, bad, test = self.whale.weighted_reasons(pk, 'random', [1,2,3])
#_print_reasons(good, bad, test)
any_one = self.whale.decide_from_reasons(good, bad, test)
self.assertEqual(True, any_one in [1, 2, 3])
# OK, now how about something somewhat informed?
# This will be easy. Slogan A makes us huge profit. Products B and C suck.
# D looks promissing but isn't yet significant
opts = ['a', 'b', 'c', 'd']
self.whale.count_now([pk, decision, 'a'], None, dict(dollars=5000, visitors=1000))
self.whale.count_now([pk, decision, 'b'], None, dict(dollars=0, visitors=2000))
self.whale.count_now([pk, decision, 'c'], None, dict(dollars=0, visitors=2000))
self.whale.count_now([pk, decision, 'd'], None, dict(dollars=50, visitors=10))
good, bad, test = self.whale.weighted_reasons(pk, decision, opts, formula='dollars/visitors')
#_print_reasons(good, bad, test)
self.assertEqual(True, 'a' in good.keys())
self.assertEqual(True, 'b' in bad.keys())
self.assertEqual(True, 'c' in bad.keys())
self.assertEqual(True, 'd' in test.keys())
which_one = self.whale.decide(pk, decision, opts, formula='dollars/visitors',
bad_idea_threshold=0, test_idea_threshold=0)
self.assertEqual(which_one, 'a')
def testInformedDecision(self):
pk = 'test_informed_decision'
decision = str(time.time())
# A is the clear winner, except when country=UK, in which case B wins
opts = ['a', 'b', 'c', 'd']
self.whale.count_now([pk, decision, 'a'], None, dict(dollars=50000, visitors=10000))
self.whale.count_now([pk, decision, 'b'], None, dict(dollars=0, visitors=2000))
self.whale.count_now([pk, decision, 'b'], {'country': 'uk'}, dict(dollars=10000, visitors=2000))
self.whale.count_now([pk, decision, 'c'], None, dict(dollars=0, visitors=7500))
self.whale.count_now([pk, decision, 'd'], None, dict(dollars=5, visitors=1))
# Here's a visitor with no info -- 'A' should win by far.
good, bad, test = self.whale.weighted_reasons(pk, decision, opts, formula='dollars/visitors')
#_print_reasons(good, bad, test)
self.assertEqual(True, 'a' in good.keys())
self.assertEqual(True, 'b' in bad.keys())
self.assertEqual(True, 'c' in bad.keys())
self.assertEqual(True, 'd' in test.keys())
# How about when we know the country is "UK"?
good, bad, test = self.whale.weighted_reasons(pk, decision, opts, formula='dollars/visitors',
known_data={'country': 'uk'})
#_print_reasons(good, bad, test)
self.assertEqual(True, 'a' in good.keys())
self.assertEqual(True, 'b' in good.keys())
self.assertEqual(True, 'c' in bad.keys())
self.assertEqual(True, 'd' in test.keys())
chosen = {'a': 0, 'b': 0}
for k in range(100):
choose = self.whale.decide(pk, decision, opts, formula='dollars/visitors',
known_data={'country': 'uk'}, bad_idea_threshold=0, test_idea_threshold=0)
chosen[choose] += 1
self.assertEqual(True, chosen['b'] > 70,
"""A decision made 100 times between weights .15 vs .85 should have around 85 votes for 'b',
we got %s, which is unlikely enough to fail a test, but not definitely
indicative of a problem. If this test passes again on the next run, ignore the failure.""" % chosen)
def testTrickyDecision(self):
pk = 'test_tricky_decision'
decision = str(time.time())
opts = ['en', 'sp', 'pt']
def count(geo, lang, dollars, visitors):
self.whale.count_decided_now(pk, decision, lang, geo,
{'dollars': dollars, 'visitors': visitors})
def justify(geo):
#print
#print 'Picking reasons for ', geo
good, bad, test = self.whale.weighted_reasons(pk, decision, opts,
'dollars/visitors', geo)
#print good.keys(), bad.keys(), test.keys()
#_print_reasons(good, bad, test)
return self.whale.decide(pk, decision, opts, 'dollars/visitors', geo,
bad_idea_threshold=0, test_idea_threshold=0)
k = 1000
m = k * k
# Sure, these results seem predictable to a human
# But what will our philosopher whale friend make of it?
count('us', 'en', 1.5 * m, 300 * k) # $5/visitor, alright!
count('us', 'sp', 1 * k, 10 * k) # $.10/visitor, well that is not surprising
count('us', 'pt', 300, 5 * k) # $.06/visitor, :(
count('mx', 'en', 100 * k, 100 * k) # $1/visitor, this almost works
count('mx', 'sp', 200 * k, 100 * k) # $2/visitor aww yah!
count('mx', 'pt', 200, 10 * k) # $.02/visitor lol
count('br', 'en', 300 * k, 100 * k) # $3/visitor is good
count('br', 'sp', 150 * k, 50 * k) # $3/visitor as well
count('br', 'pt', 500 * k, 50 * k) # $10 JACKPOT
self.assertEqual('en', justify('us'))
self.assertEqual(True, justify('mx') in ['sp', 'en'])
self.assertEqual('pt', justify('br'))
def testWhaleCacheWrapper(self):
t = str(time.time())
count = lambda: self.whale.count_now('test_cached', t)
cached_sum = lambda clear=False: sum(self.whale.cached_plotpoints('test_cached',
t, period='fivemin', unmemoize=clear)[t]['hits'].values())
# Set hits to 1
count()
self.assertEqual(cached_sum(), 1)
# Should stay 1 for a while
for i in range(3):
count()
self.assertEqual(cached_sum(), 1)
self.assertEqual(cached_sum(clear=True), 4)
def count_now():
from datetime import datetime
whale = Whale()
val = whale.count_now(at=datetime.utcnow(), **default_params())
return 'OK'
def count_now():
whale = Whale()
val = whale.count_now(at=datetime.now(), **default_params())
return 'OK'
class TestHailWhale(unittest.TestCase):
def setUp(self):
from hail import Hail
from whale import Whale
self.hail = Hail()
self.whale = Whale()
def testGetSubdimensions(self):
t = 'subs_%s' % str(time.time())
self.whale.count_now(t, {'a': 1, 'b': 2})
subs = self.whale.get_subdimensions(t)
assert('a' in subs)
assert('b' in subs)
def testGetAllSubdimensions(self):
t = 'all_subs_%s' % str(time.time())
self.whale.count_now(t, {'a': 1, 'b': 2})
subs = self.whale.all_subdimensions(t)
assert('a' in subs)
assert(['a', '1'] in subs)
assert('b' in subs)
assert(['b', '2'] in subs)
def testPlotpoints(self):
t = str(time.time())
for i in range(5):
self.whale.count_now('test_plotpoints', t, {'hits': 1, 'values': 5})
plotpoints = self.whale.plotpoints('test_plotpoints', t, ['hits', 'values'], points_type=list)
self.assertEqual(plotpoints[t]['hits'][-1][1], 5)
self.assertEqual(plotpoints[t]['values'][-1][1], 25)
def testPlotpointsDepth(self):
t = str(time.time())
self.whale.count_now('test_depth', {t: 'a'})
self.whale.count_now('test_depth', {t: 'b'})
self.whale.count_now('test_depth', {t: 'b'})
self.whale.count_now('test_depth', {t: {'c': 'child'}})
# Test 1 level deep
plotpoints = self.whale.plotpoints('test_depth', t, points_type=list, depth=1)
self.assertEqual(plotpoints[maybe_dumps([t, 'a'])]['hits'][-1][1], 1)
self.assertEqual(plotpoints[maybe_dumps([t, 'b'])]['hits'][-1][1], 2)
self.assertEqual(plotpoints[maybe_dumps([t, 'c'])]['hits'][-1][1], 1)
self.assertEqual(False, maybe_dumps([t, 'c', 'child']) in plotpoints)
# Test 2 levels deep
plotpoints = self.whale.plotpoints('test_depth', t, points_type=list, depth=2)
self.assertEqual(True, maybe_dumps([t, 'c', 'child']) in plotpoints)
self.assertEqual(plotpoints[maybe_dumps([t, 'c', 'child'])]['hits'][-1][1], 1)
# Test ranking and limiting i.e assign rank on the basis of value and then extract top limit candidate
plotpoints = self.whale.plotpoints('test_depth', t, points_type=list,depth=1, limit=2)
#self.assertEqual(plotpoints[maybe_dumps([t, 'b'])]['hits'][-1][1], 2)
self.assertEqual(True, maybe_dumps([t, 'a']) not in plotpoints)
self.assertEqual(True, maybe_dumps([t, 'c']) not in plotpoints)
def testRatioPlotpoints(self):
t = str(time.time())
for i in range(5):
self.whale.count_now('test_ratio', t, {'hit': 1, 'value': 5})
plotpoints = self.whale.plotpoints('test_ratio', t, ['hit', 'value', 'value/hit'], points_type=list)
self.assertEqual(plotpoints[t]['hit'][-1][1], 5)
self.assertEqual(plotpoints[t]['value'][-1][1], 25)
self.assertEqual(plotpoints[t]['value/hit'][-1][1], 5)
def testRankSubdimensionsScalar(self):
t = str(time.time())
self.whale.count_now('test_rank', [t, 'a', 'asub1'], {'value': 1})
self.whale.count_now('test_rank', [t, 'a', 'asub2'], {'value': 30})
self.whale.count_now('test_rank', [t, 'b'], {'value': 80})
self.whale.count_now('test_rank', [t, 'c'], {'value': 10})
ranked = self.whale.rank_subdimensions_scalar('test_rank', t, 'value')
self.assertEqual(ranked[maybe_dumps([t, 'a'])]['important'], False)
self.assertEqual(ranked[maybe_dumps([t, 'a', 'asub1'])]['important'], False)
self.assertEqual(ranked[maybe_dumps([t, 'a', 'asub2'])]['important'], True)
self.assertEqual(ranked[maybe_dumps([t, 'b'])]['important'], True)
self.assertEqual(ranked[maybe_dumps([t, 'c'])]['important'], False)
def testRankSubdimensionsRatio(self):
t = str(time.time())
pk = 'test_ratio_rank'
# OVERALL STATS: 529,994 value, 50,000 visitors, 10.6 value per visitor
# Not important, too close to overall
self.whale.count_now(pk, [t, 'a', 'asub1'],
{'value': 54989, 'visitors': 4999}) # 11 value per visitor
# Important, high relative ratio
self.whale.count_now(pk, [t, 'a', 'asub2'],
{'value': 375000, 'visitors': 25000}) # 15 value per visitor
# Important, low relative ratio
self.whale.count_now(pk, [t, 'b'],
{'value': 100000, 'visitors': 20000}) # 5 value per visitor
# Not important, not enough visitors
self.whale.count_now(pk, [t, 'c'],
{'value': 5, 'visitors': 1}) # 5 value per visitor
one_level = self.whale.rank_subdimensions_ratio('test_rank_ratio', 'value', 'visitors',
t, recursive=False)
all_levels = self.whale.rank_subdimensions_ratio(pk, 'value', 'visitors', t)
self.assertEqual(True, maybe_dumps([t, 'a', 'asub1']) not in one_level)
self.assertEqual(all_levels[maybe_dumps([t, 'a', 'asub1'])]['important'], False)
self.assertEqual(all_levels[maybe_dumps([t, 'a', 'asub2'])]['important'], True)
self.assertEqual(all_levels[maybe_dumps([t, 'b'])]['important'], True)
self.assertEqual(all_levels[maybe_dumps([t, 'c'])]['important'], False)
def testBasicDecision(self):
pk = 'test_basic_decision'
decision = str(time.time())
# Make a decision, any decision, from no information whatsoever
good, bad, test = self.whale.weighted_reasons(pk, 'random', [1,2,3])
#_print_reasons(good, bad, test)
any_one = self.whale.decide_from_reasons(good, bad, test)
self.assertEqual(True, any_one in [1, 2, 3])
# OK, now how about something somewhat informed?
# This will be easy. Slogan A makes us huge profit. Products B and C suck.
# D looks promissing but isn't yet significant
opts = ['a', 'b', 'c', 'd']
self.whale.count_now([pk, decision, 'a'], None, dict(dollars=5000, visitors=1000))
self.whale.count_now([pk, decision, 'b'], None, dict(dollars=0, visitors=2000))
self.whale.count_now([pk, decision, 'c'], None, dict(dollars=0, visitors=2000))
self.whale.count_now([pk, decision, 'd'], None, dict(dollars=50, visitors=10))
good, bad, test = self.whale.weighted_reasons(pk, decision, opts, formula='dollars/visitors')
#_print_reasons(good, bad, test)
self.assertEqual(True, 'a' in good.keys())
self.assertEqual(True, 'b' in bad.keys())
self.assertEqual(True, 'c' in bad.keys())
self.assertEqual(True, 'd' in test.keys())
which_one = self.whale.decide(pk, decision, opts, formula='dollars/visitors',
bad_idea_threshold=0, test_idea_threshold=0)
self.assertEqual(which_one, 'a')
def testInformedDecision(self):
pk = 'test_informed_decision'
decision = str(time.time())
# A is the clear winner, except when country=UK, in which case B wins
opts = ['a', 'b', 'c', 'd']
self.whale.count_now([pk, decision, 'a'], None, dict(dollars=50000, visitors=10000))
self.whale.count_now([pk, decision, 'b'], None, dict(dollars=0, visitors=2000))
self.whale.count_now([pk, decision, 'b'], {'country': 'uk'}, dict(dollars=10000, visitors=2000))
self.whale.count_now([pk, decision, 'c'], None, dict(dollars=0, visitors=7500))
self.whale.count_now([pk, decision, 'd'], None, dict(dollars=5, visitors=1))
# Here's a visitor with no info -- 'A' should win by far.
good, bad, test = self.whale.weighted_reasons(pk, decision, opts, formula='dollars/visitors')
#_print_reasons(good, bad, test)
self.assertEqual(True, 'a' in good.keys())
self.assertEqual(True, 'b' in bad.keys())
self.assertEqual(True, 'c' in bad.keys())
self.assertEqual(True, 'd' in test.keys())
# How about when we know the country is "UK"?
good, bad, test = self.whale.weighted_reasons(pk, decision, opts, formula='dollars/visitors',
known_data={'country': 'uk'})
#_print_reasons(good, bad, test)
self.assertEqual(True, 'a' in good.keys())
self.assertEqual(True, 'b' in good.keys())
self.assertEqual(True, 'c' in bad.keys())
self.assertEqual(True, 'd' in test.keys())
chosen = {'a': 0, 'b': 0}
for k in range(100):
choose = self.whale.decide(pk, decision, opts, formula='dollars/visitors',
known_data={'country': 'uk'}, bad_idea_threshold=0, test_idea_threshold=0)
chosen[choose] += 1
self.assertEqual(True, chosen['b'] > 70,
"""A decision made 100 times between weights .15 vs .85 should have around 85 votes for 'b',
we got %s, which is unlikely enough to fail a test, but not definitely
indicative of a problem. If this test passes again on the next run, ignore the failure.""" % chosen)
def testTrickyDecision(self):
pk = 'test_tricky_decision'
decision = str(time.time())
opts = ['en', 'sp', 'pt']
def count(geo, lang, dollars, visitors):
self.whale.count_decided_now(pk, decision, lang, geo,
{'dollars': dollars, 'visitors': visitors})
def justify(geo):
#print
#print 'Picking reasons for ', geo
good, bad, test = self.whale.weighted_reasons(pk, decision, opts,
'dollars/visitors', geo)
#print good.keys(), bad.keys(), test.keys()
#_print_reasons(good, bad, test)
return self.whale.decide(pk, decision, opts, 'dollars/visitors', geo,
bad_idea_threshold=0, test_idea_threshold=0)
k = 1000
m = k * k
# Sure, these results seem predictable to a human
# But what will our philosopher whale friend make of it?
count('us', 'en', 1.5 * m, 300 * k) # $5/visitor, alright!
count('us', 'sp', 1 * k, 10 * k) # $.10/visitor, well that is not surprising
count('us', 'pt', 300, 5 * k) # $.06/visitor, :(
count('mx', 'en', 100 * k, 100 * k) # $1/visitor, this almost works
count('mx', 'sp', 200 * k, 100 * k) # $2/visitor aww yah!
count('mx', 'pt', 200, 10 * k) # $.02/visitor lol
count('br', 'en', 300 * k, 100 * k) # $3/visitor is good
count('br', 'sp', 150 * k, 50 * k) # $3/visitor as well
count('br', 'pt', 500 * k, 50 * k) # $10 JACKPOT
self.assertEqual('en', justify('us'))
self.assertEqual(True, justify('mx') in ['sp', 'en'])
self.assertEqual('pt', justify('br'))
def testWhaleCacheWrapper(self):
t = str(time.time())
count = lambda: self.whale.count_now('test_cached', t)
cached_sum = lambda clear=False: sum(self.whale.cached_plotpoints('test_cached',
t, period='fivemin', unmemoize=clear)[t]['hits'].values())
# Set hits to 1
count()
self.assertEqual(cached_sum(), 1)
# Should stay 1 for a while
for i in range(3):
count()
self.assertEqual(cached_sum(), 1)
self.assertEqual(cached_sum(clear=True), 4)