def probYearGivenWeightedNames(wnames, r):
'''
Now the first parameter, the set of names, is provided with the associated
weights with names. Our task is to calculate p(year|names), where year is
between r[0] and r[1], inclusive.
The data structure of the wnames is as follows:
[(name, weight), (name, weight), ...]
'''
assert(len(r) == 2)
assert(r[0] <= r[1])
names, weights = zip(*wnames)
# normalize the weights
sum_weights = float(sum(weights))
weights = [x / sum_weights for x in weights]
wnames = zip(names, weights)
# calculation part
sum_prob_year = None
for name, weight in wnames:
prob_year = probYearGivenName(name, r[0], r[1])
prob_year = vector_scalar_product(prob_year, weight)
if sum_prob_year is None:
sum_prob_year = prob_year
else:
sum_prob_year = vector_sum(sum_prob_year, prob_year)
# Finally, we need to normalize the prob array to ensure that the sum is 1.
factor = 1 / float(sum(sum_prob_year))
sum_prob_year = vector_scalar_product(sum_prob_year, factor)
return sum_prob_year
def probYearGivenNames(names, r):
'''
Given a set of names (first names), we return p(year|names), i.e. the
probability of a random user from the set might be born in that year. Since
the year is a variable, we return an array in which each element holds a
probability for corresponding year.
The range specified by 'r' is inclusive. The output array is from r[0]
through r[1].
'''
assert(len(r) == 2) # r must be a tuple or a list of length 2
assert(r[0] <= r[1])
prob_year_sum = None
for name in names:
prob_year = probYearGivenName(name, r[0], r[1])
if prob_year_sum is None:
prob_year_sum = prob_year
else:
prob_year_sum = vector_sum(prob_year_sum, prob_year)
# Finally, we need to normalize the prob array to ensure that the sum is 1.
factor = 1 / float(sum(sum_prob_year))
sum_prob_year = vector_scalar_product(sum_prob_year, factor)
return prob_year_sum
def run_allName():
con = mdb.connect('localhost', 'yongjoo', 'Fgla4Zp0', 'yongjoo')
with con:
cur = con.cursor()
sql = '''SELECT name, SUM(num) AS total
FROM babyname
GROUP BY name
ORDER BY total DESC
LIMIT 1000
'''
cur.execute(sql)
bins = [0 for x in range(1950, 2001)]
debugout = open('debug.txt', 'w')
numrows = int(cur.rowcount)
for i in range(numrows):
row = cur.fetchone()
name = row[0]
firstname = name.split(' ')[0]
prob_year = babyname.probYearGivenName(firstname, 1950, 2000)
try:
assert(len(bins) == len(prob_year))
except AssertionError:
print len(bins), len(prob_year)
raise RuntimeError
if sum(prob_year) != 0:
bins = [bins[i] + prob_year[i] for i in range(len(bins))]
debugout.close()
print bins
def probYearGivenDomainInYear(start_year, end_year):
'''
We compute in the function y(year) = avg_i p(year|name_i) where name_i is
the i-th element in the set of users who were born between start_year
(inclusive) and end_year (exclusive).
The function avg is not a simple average since we consider the weights
between different names, thus it's rather a weighted average. Weights are
determined by considering the sum of the counts of the names between the
periods specified in the arguments.
'''
assert(start_year < end_year)
return_range = (2012-50, 2012-11) # this is inclusive
q = '''SELECT b.name AS name, sum(b.num) AS count
FROM babyname b
INNER JOIN popularname p
ON b.name = p.name
WHERE year BETWEEN %s AND %s
GROUP BY name
ORDER BY count
''' % (start_year, end_year - 1) # sql between is inclusive
name_weights = {} # key: name, value: weight
con = mdb.connect('localhost', 'yongjoo', 'Fgla4Zp0', 'yongjoo');
with con:
cur = con.cursor()
cur.execute(q)
numrows = int(cur.rowcount)
for i in range(numrows):
row = cur.fetchone()
name = row[0]
count = row[1]
name_weights[name] = count
# noramlize the weights for the sake of the calucation safety
sum_name_weights = sum(name_weights.values())
for n, w in name_weights.iteritems():
name_weights[n] /= sum_name_weights
# Get prob distribution for every name
name_prob_year = {}
for n in name_weights.keys():
prob_year = probYearGivenName(n, return_range[0], return_range[1])
name_prob_year[n] = prob_year
# Get weighted average of them
assert(len(name_prob_year) > 0) # sanity check
d = name_prob_year
sum_prob_year = [0] * len(d[random.choice(d.keys())])
for n in name_prob_year.keys():
weighted = vector_scalar_product(name_prob_year[n], name_weights[n])
sum_prob_year = vector_sum(sum_prob_year, weighted)
# We don't need to average since the summation of the weights is one.
# However, we need to normalize the prob array to ensure that the sum is 1.
factor = 1 / float(sum(sum_prob_year))
sum_prob_year = vector_scalar_product(sum_prob_year, factor)
return sum_prob_year
prob_year_sum[age] = []
skip_count[age] = 0
# Start to collect prob for every age.
for line in datain:
tweet = json.loads(line)
name = tweet['user']['name'].encode('ascii', 'ignore')
firstname = name.split(' ')[0]
age = int(tweet['user']['age'])
if not age in given_ages:
continue;
start_year = 2012 - 50
end_year = 2012 - 11
prob_year = probYearGivenName(firstname, start_year, end_year)
if sum(prob_year) == 0:
skip_count[age] += 1
continue
prob_year_sum[age].append(prob_year)
for age in sorted(prob_year_sum.keys()):
print "For %d, collected %d number of users. Skipped %d" % (age,
len(prob_year_sum[age]), skip_count[age])
# Average the prob array
def elementWiseSum(x, y):
assert(len(x) == len(y))
