def drop(feature_name, month, fname):
col_names = np.genfromtxt(paths.my_path + 'col_names{0}.csv'.format(month),
delimiter=',',
dtype=str)
col_number = col_names.shape[1]
X = Matrix(
np.genfromtxt(paths.my_path + 'X{0}.csv'.format(month),
delimiter=',',
dtype=str), list(col_names[0, :]),
["%s" for i in xrange(col_names.shape[1])])
X.drop(feature_name)
def gen_user_buy_shop(since, to, month):
assert to == month - 1
offline = Matrix(np.genfromtxt('offline_{0}_month_{1}{2}.csv'.format(to - since + 1, since, to), delimiter=',', dtype=str), ["uid", "mid", "cid", "dis_rate", "dist", "date_rec", "date"], ["%s" for i in xrange(7)])
online = Matrix(np.genfromtxt('offline_{0}_month_{1}{2}.csv'.format(to - since + 1, since, to), delimiter=',', dtype=str), ["uid", "mid", "act", "cid", "dis_rate", "date_rec", "date"], ["%s" for i in xrange(7)])
col_names = np.genfromtxt(paths.my_path + 'col_names_{0}_{1}{2}.csv'.format(month, since, to), delimiter=',', dtype=str)
col_number = col_names.shape[1]
X = Matrix(np.genfromtxt(paths.my_path + 'X_{0}_{1}{2}.csv'.format(month, since, to), delimiter=',', dtype=str), list(col_names[0, :]), ["%s" for i in xrange(col_names.shape[1])])
user_buy_shop = HashSet()
for i in xrange(offline.ndata):
def append_feature_by_month(func, month, since, to):
col_names = np.genfromtxt(
paths.my_path + 'col_names_{0}_{1}{2}.csv'.format(month, since, to),
delimiter=',',
dtype=str)
col_number = col_names.shape[1]
X = Matrix(
np.genfromtxt(paths.my_path +
'X_{0}_{1}{2}.csv'.format(month, since, to),
delimiter=',',
dtype=str), list(col_names[0, :]),
["%s" for i in xrange(col_names.shape[1])])
func(X, month)
X.check_point(month, since, to)
def gen_user_get_shop_coupon(offline_source, online_source, X):
offline = Matrix(
np.genfromtxt(paths.ccf_path + offline_source,
delimiter=',',
dtype=str),
["uid", "mid", "cid", "dis_rate", "dist", "date_rec", "date"],
["%s" for i in xrange(7)])
online = Matrix(
np.genfromtxt(paths.ccf_path + online_source, delimiter=',',
dtype=str),
["uid", "mid", "act", "cid", "dis_rate", "date_rec", "date"],
["%s" for i in xrange(7)])
user_get_shop_coupon = HashSet()
for i in xrange(offline.ndata):
uid_str = offline.get_cell(i, "uid")
cid_str = offline.get_cell(i, "cid")
mid_str = offline.get_cell(i, "mid")
if cid_str != 'null':
if not user_get_shop_coupon.has(uid_str):
user_get_shop_coupon.set(uid_str, HashSet())
user_get_shop_coupon.get(uid_str).add_one(mid_str)
for i in xrange(online.ndata):
uid_str = online.get_cell(i, "uid")
cid_str = online.get_cell(i, "cid")
mid_str = online.get_cell(i, "mid")
act_str = online.get_cell(i, "act")
if cid_str != 'null' and act_str != '0':
if not user_get_shop_coupon.has(uid_str):
user_get_shop_coupon.set(uid_str, HashSet())
user_get_shop_coupon.get(uid_str).add_one(mid_str)
X.join_by_double_key("uid",
"mid",
"user_get_shop_coupon",
user_get_shop_coupon,
"%s",
dft=0.0)
def divide(x, y):
assert float(x) <= float(y)
if float(y) == 0:
return 0.0
else:
return float(x) * 1.0 / float(y)
X.gen_arith_feature("user_buy_with_coupon_in_shop",
"user_get_shop_coupon",
"user_use_shop_coupon_freq",
divide,
"%s",
dft=0.0)
def append_feature_by_month(func, month):
col_names = np.genfromtxt(
paths.my_path +
'X_{0}_{1}{2}_shop_month_col_names.csv'.format(month, 1, month - 1),
delimiter=',',
dtype=str)
col_number = col_names.shape[1]
X = Matrix(
np.genfromtxt(
paths.my_path +
'X_{0}_{1}{2}_shop_month.csv'.format(month, 1, month - 1),
delimiter=',',
dtype=str), list(col_names[0, :]),
["%s" for i in xrange(col_names.shape[1])])
func(X, month)
def drop_multiple(feature_names, month):
col_names = np.genfromtxt(paths.my_path + 'col_names{0}.csv'.format(month),
delimiter=',',
dtype=str)
col_number = col_names.shape[1]
X = Matrix(
np.genfromtxt(paths.my_path + 'X{0}.csv'.format(month),
delimiter=',',
dtype=str), list(col_names[0, :]),
["%s" for i in xrange(col_names.shape[1])])
for name in feature_names:
X.drop(name)
X.check_point(month)
def gen_user_get_coupon(offline_source, online_source, X, month):
offline = Matrix(
np.genfromtxt(paths.ccf_path + offline_source,
delimiter=',',
dtype=str),
["uid", "mid", "cid", "dis_rate", "dist", "date_rec", "date"],
["%s" for i in xrange(7)])
online = Matrix(
np.genfromtxt(paths.ccf_path + online_source, delimiter=',',
dtype=str),
["uid", "mid", "act", "cid", "dis_rate", "date_rec", "date"],
["%s" for i in xrange(7)])
user_get_coupon = HashSet()
for i in xrange(offline.ndata):
uid_str = offline.get_cell(i, "uid")
cid_str = offline.get_cell(i, "cid")
if cid_str != 'null':
user_get_coupon.add_one(uid_str)
for i in xrange(online.ndata):
uid_str = online.get_cell(i, "uid")
cid_str = online.get_cell(i, "cid")
act_str = online.get_cell(i, "act")
if cid_str != 'null':
assert act_str != '0'
user_get_coupon.add_one(uid_str)
X.join("uid", ["user_get_coupon"], user_get_coupon, ["%s"], dft=0.0)
def divide(x, y):
assert float(x) != -9999 and float(y) != -9999
assert float(x) <= float(y), "{0} {1}".format(x, y)
if float(y) == 0:
return 0.0
else:
return float(x) * 1.0 / float(y)
X.gen_arith_feature("user_buy_with_coupon",
"user_get_coupon",
"user_use_coupon_freq",
divide,
"%s",
dft=0.0)
def gen_merchant_share(X, month):
offline = Matrix(
np.genfromtxt(paths.ccf_path + 'offline_train_{0}.csv'.format(month),
delimiter=',',
dtype=str),
["uid", "mid", "cid", "dis_rate", "dist", "date_rec", "date"],
["%s" for i in xrange(7)])
online = Matrix(
np.genfromtxt(paths.ccf_path + 'online_train_{0}.csv'.format(month),
delimiter=',',
dtype=str),
["uid", "mid", "act", "cid", "dis_rate", "date_rec", "date"],
["%s" for i in xrange(7)])
merchant_user_buy = HashSet()
merchant_user_use_coupon = HashSet()
merchant_user_buy_counter = HashSet()
merchant_user_use_coupon_counter = HashSet()
for i in xrange(offline.ndata):
if i % 100000 == 0:
print i
mid_str = offline.get_cell(i, "mid")
uid_str = offline.get_cell(i, "uid")
date_str = offline.get_cell(i, "date")
cid_str = offline.get_cell(i, "cid")
if date_str != 'null':
if not merchant_user_buy.has(mid_str):
merchant_user_buy.set(mid_str, HashSet())
if not merchant_user_buy.get(mid_str).has(uid_str):
merchant_user_buy.get(mid_str).add_one(uid_str)
merchant_user_buy_counter.add_one(mid_str)
if cid_str != 'null':
if not merchant_user_use_coupon.has(mid_str):
merchant_user_use_coupon.set(mid_str, HashSet())
if not merchant_user_use_coupon.get(mid_str).has(uid_str):
merchant_user_use_coupon.get(mid_str).add_one(uid_str)
merchant_user_use_coupon_counter.add_one(mid_str)
for i in xrange(online.ndata):
if i % 100000 == 0:
print i
mid_str = online.get_cell(i, "mid")
uid_str = online.get_cell(i, "uid")
act_str = online.get_cell(i, "act")
cid_str = online.get_cell(i, "cid")
if act_str == '1':
if not merchant_user_buy.has(mid_str):
merchant_user_buy.set(mid_str, HashSet())
if not merchant_user_buy.get(mid_str).has(uid_str):
merchant_user_buy.get(mid_str).add_one(uid_str)
merchant_user_buy_counter.add_one(mid_str)
if cid_str != 'null':
if not merchant_user_use_coupon.has(mid_str):
merchant_user_use_coupon.set(mid_str, HashSet())
if not merchant_user_use_coupon.get(mid_str).has(uid_str):
merchant_user_use_coupon.get(mid_str).add_one(uid_str)
merchant_user_use_coupon_counter.add_one(mid_str)
X.join("mid", ["merchant_user_buy"],
merchant_user_buy_counter, ["%s"],
dft=0.0)
X.join("mid", ["merchant_user_use_coupon"],
merchant_user_use_coupon_counter, ["%s"],
dft=0.0)
X.check_point(month)
def gen_merchant_buy(X, month):
offline = Matrix(
np.genfromtxt(paths.ccf_path + 'offline_train_{0}.csv'.format(month),
delimiter=',',
dtype=str),
["uid", "mid", "cid", "dis_rate", "dist", "date_rec", "date"],
["%s" for i in xrange(7)])
online = Matrix(
np.genfromtxt(paths.ccf_path + 'online_train_{0}.csv'.format(month),
delimiter=',',
dtype=str),
["uid", "mid", "act", "cid", "dis_rate", "date_rec", "date"],
["%s" for i in xrange(7)])
merchant_buy = HashSet()
merchant_buy_with_coupon = HashSet()
merchant_distribute_coupon = HashSet()
for i in xrange(offline.ndata):
if i % 100000 == 0:
print i
mid_str = offline.get_cell(i, "mid")
date_str = offline.get_cell(i, "date")
cid_str = offline.get_cell(i, "cid")
if date_str != 'null':
merchant_buy.add_one(mid_str)
if cid_str != 'null':
merchant_buy_with_coupon.add_one(mid_str)
if cid_str != 'null':
merchant_distribute_coupon.add_one(mid_str)
for i in xrange(online.ndata):
if i % 100000 == 0:
print i
mid_str = online.get_cell(i, "mid")
act_str = online.get_cell(i, "act")
cid_str = online.get_cell(i, "cid")
if act_str == '1':
merchant_buy.add_one(mid_str)
if cid_str != 'null':
merchant_buy_with_coupon.add_one(mid_str)
if cid_str != 'null':
assert act_str != '0'
merchant_distribute_coupon.add_one(mid_str)
X.join("mid", ["merchant_buy"], merchant_buy, ["%s"], 0.0)
X.join("mid", ["merchant_buy_with_coupon"], merchant_buy_with_coupon,
["%s"], 0.0)
X.join("mid", ["merchant_distribute_coupon"], merchant_distribute_coupon,
["%s"], 0.0)
def divide(x, y):
assert float(x) <= float(y)
if float(y) == 0:
return 0.0
else:
return float(x) * 1.0 / float(y)
X.gen_arith_feature("merchant_buy_with_coupon",
"merchant_buy",
"merchant_buy_with_coupon_ratio",
divide,
"%s",
dft=0.0)
X.gen_arith_feature("merchant_buy_with_coupon",
"merchant_distribute_coupon",
"merchant_coupon_ratio",
divide,
"%s",
dft=0.0)
X.check_point(month)
def gen_user_recent_behavior(X, month):
offline = Matrix(
np.genfromtxt(paths.ccf_path + 'offline_train_{0}.csv'.format(month),
delimiter=',',
dtype=str),
["uid", "mid", "cid", "dis_rate", "dist", "date_rec", "date"],
["%s" for i in xrange(7)])
online = Matrix(
np.genfromtxt(paths.ccf_path + 'online_train_{0}.csv'.format(month),
delimiter=',',
dtype=str),
["uid", "mid", "act", "cid", "dis_rate", "date_rec", "date"],
["%s" for i in xrange(7)])
behaviors = {
1: HashSet(),
2: HashSet(),
3: HashSet(),
4: HashSet(),
5: HashSet(),
6: HashSet()
}
for i in xrange(offline.ndata):
if i % 100000 == 0:
print i
uid_str = offline.get_cell(i, "uid")
date_str = offline.get_cell(i, "date")
cid_str = offline.get_cell(i, "cid")
date_rec_str = offline.get_cell(i, "date_rec")
if date_str != 'null':
act_month = get_month(date_str)
if not behaviors[act_month].has(uid_str):
behaviors[act_month].set(uid_str, np.zeros(3))
if cid_str != 'null':
behaviors[act_month].get(uid_str)[2] += 1
else:
behaviors[act_month].get(uid_str)[1] += 1
else:
act_month = get_month(date_rec_str)
if not behaviors[act_month].has(uid_str):
behaviors[act_month].set(uid_str, np.zeros(3))
behaviors[act_month].get(uid_str)[0] += 1
for i in xrange(online.ndata):
if i % 100000 == 0:
print i
uid_str = online.get_cell(i, "uid")
date_str = online.get_cell(i, "date")
cid_str = online.get_cell(i, "cid")
date_rec_str = online.get_cell(i, "date_rec")
act_str = online.get_cell(i, "act")
if act_str == '0':
continue
if date_str != 'null':
assert act_str == '1'
act_month = get_month(date_str)
if not behaviors[act_month].has(uid_str):
behaviors[act_month].set(uid_str, np.zeros(3))
if cid_str != 'null':
behaviors[act_month].get(uid_str)[2] += 1
else:
behaviors[act_month].get(uid_str)[1] += 1
else:
assert act_str == '2'
act_month = get_month(date_rec_str)
if not behaviors[act_month].has(uid_str):
behaviors[act_month].set(uid_str, np.zeros(3))
behaviors[act_month].get(uid_str)[0] += 1
def gen_user_recent_behavior_func(row):
date_rec_str = row['date_rec']
uid_str = row['uid']
act_month = get_month(date_rec_str)
behavior_history = np.zeros(0)
for i in xrange(4):
behavior_vector = np.array([-9999, -9999, -9999])
if date_rec_str != 'null' and act_month - 1 - i > 0 and behaviors[
act_month - 1 - i].has(uid_str):
behavior_vector = behaviors[act_month - 1 - i].get(uid_str)
elif act_month - 1 - i > 0 and date_rec_str != 'null':
behavior_vector = np.array([0, 0, 0])
behavior_history = np.hstack((behavior_history, behavior_vector))
return behavior_history
names = []
for i in xrange(4):
names += [
"recent_user_unuse{0}".format(i), "recent_user_buy{0}".format(i),
"recent_user_use{0}".format(i)
]
X.gen_features(names, gen_user_recent_behavior_func,
["%s" for i in xrange(12)])
X.check_point(month)
def gen_by_month(offline_source,
online_source,
target_file,
month,
since=None):
offline = Matrix(
np.genfromtxt(paths.ccf_path + offline_source,
delimiter=',',
dtype=str),
["uid", "mid", "cid", "dis_rate", "dist", "date_rec", "date"],
["%s" for i in xrange(7)])
online = Matrix(
np.genfromtxt(paths.ccf_path + online_source, delimiter=',',
dtype=str),
["uid", "mid", "act", "cid", "dis_rate", "date_rec", "date"],
["%s" for i in xrange(7)])
if month < 7:
X = Matrix(
np.genfromtxt(paths.ccf_path + 'offline_train_test_' + str(month) +
'.csv',
delimiter=',',
dtype=str),
["uid", "mid", "cid", "dis_rate", "dist", "date_rec", "date"],
["%s" for i in xrange(7)])
else:
X = Matrix(
np.genfromtxt(paths.ccf_path + 'offline_train_test_' + str(month) +
'.csv',
delimiter=',',
dtype=str),
["uid", "mid", "cid", "dis_rate", "dist", "date_rec"],
["%s" for i in xrange(6)])
print "generating month {0}".format(month)
print "gen_basic_features" #11
gen_basic_features(X)
print "gen_user_buy_with_coupon"
gen_user_buy_with_coupon(offline, online, X) #14
print "gen_user_shop_features"
#X.join("uid", feature_names, features, formats, dft=0.0)
gen_user_shop_features(
paths.my_path + "u_features_10_{0}.csv".format(month),
paths.my_path + "i_features_10_{0}.csv".format(month), X) #35
print "gen_user_buy_in_shop"
gen_user_buy_in_shop(offline, online, X) #37
print "gen_user_buy_coupon_in_shop"
gen_user_buy_coupon_in_shop(offline, online, X)
if month < 7:
print "gen_label"
gen_label(X)
print "gen_user_get_shop_coupon"
gen_user_get_shop_coupon(offline_source, online_source, X)
print "gen_unused_shop_coupon"
gen_unused_shop_coupon(X)
print "gen_no_penalty_user_shop_features"
gen_no_penalty_user_shop_features(X, month)
X.check_point(target_file)
def gen_coupon_gap(X, month):
get_coupon_history = HashSet()
get_coupon_history_user = HashSet()
for i in xrange(X.ndata):
if i % 100000 == 0:
print i
uid_str = X.get_cell(i, "uid")
mid_str = X.get_cell(i, "mid")
date_rec_str = X.get_cell(i, "date_rec")
if date_rec_str != 'null':
history_list = get_coupon_history.get(uid_str,
HashSet()).get(mid_str, [])
#if date_rec_str not in history_list:
history_list.append(date_rec_str)
user_history_list = get_coupon_history_user.get(uid_str, [])
#if date_rec_str not in user_history_list:
user_history_list.append(date_rec_str)
if month > 1:
last_X = Matrix(
np.genfromtxt(paths.ccf_path +
'offline_1_month{0}.csv'.format(month - 1),
delimiter=',',
dtype=str),
["uid", "mid", "cid", "dis_rate", "dist", "date_rec", "date"],
["%s" for i in xrange(7)])
for i in xrange(last_X.ndata):
if i % 100000 == 0:
print i
uid_str = last_X.get_cell(i, "uid")
mid_str = last_X.get_cell(i, "mid")
date_rec_str = last_X.get_cell(i, "date_rec")
if date_rec_str != 'null':
history_list = get_coupon_history.get(uid_str, HashSet()).get(
mid_str, [])
#if date_rec_str not in history_list:
history_list.append(date_rec_str)
user_history_list = get_coupon_history_user.get(uid_str, [])
#if date_rec_str not in user_history_list:
user_history_list.append(date_rec_str)
if month < 7:
next_X = Matrix(
np.genfromtxt(paths.ccf_path +
'offline_1_month{0}.csv'.format(month + 1),
delimiter=',',
dtype=str),
["uid", "mid", "cid", "dis_rate", "dist", "date_rec", "date"],
["%s" for i in xrange(7)])
for i in xrange(next_X.ndata):
if i % 100000 == 0:
print i
uid_str = next_X.get_cell(i, "uid")
mid_str = next_X.get_cell(i, "mid")
date_rec_str = next_X.get_cell(i, "date_rec")
if date_rec_str != 'null':
history_list = get_coupon_history.get(uid_str, HashSet()).get(
mid_str, [])
#if date_rec_str not in history_list:
history_list.append(date_rec_str)
user_history_list = get_coupon_history_user.get(uid_str, [])
#if date_rec_str not in user_history_list:
user_history_list.append(date_rec_str)
for i in xrange(X.ndata):
if i % 100000 == 0:
print i
uid_str = X.get_cell(i, "uid")
mid_str = X.get_cell(i, "mid")
history_list = get_coupon_history.get(uid_str,
HashSet()).get(mid_str, [])
get_coupon_history.get(uid_str).set(mid_str, np.sort(history_list))
user_history_list = get_coupon_history_user.get(uid_str, [])
get_coupon_history_user.set(uid_str, np.sort(user_history_list))
col_names = [
"prev_gap", "next_gap", "user_prev_gap", "user_next_gap", "in_between",
"prev_gap_prev", "next_gap_prev"
]
gaps = np.zeros((X.ndata, len(col_names)))
for i in xrange(X.ndata):
if i % 100000 == 0:
print i
uid_str = X.get_cell(i, "uid")
mid_str = X.get_cell(i, "mid")
date_rec_str = X.get_cell(i, "date_rec")
history_list = get_coupon_history.get(uid_str).get(mid_str)
user_history_list = get_coupon_history_user.get(uid_str)
k = -1
for j in xrange(len(history_list)):
##if history_list[j] == date_rec_str:
if j == len(history_list) - 1 or history_list[j +
1] > date_rec_str:
break
else:
k += 1
in_between = 1
if k > 0:
prev_gap = days_dis(history_list[k], date_rec_str)
else:
prev_gap = 0
in_between = 0
if k + 2 < len(history_list):
next_gap = days_dis(date_rec_str, history_list[k + 2])
else:
next_gap = 0
in_between = 0
k = -1
for j in xrange(len(user_history_list)):
##if user_history_list[j] == date_rec_str:
if j == len(history_list) - 1 or history_list[j +
1] > date_rec_str:
break
else:
k += 1
if k > 0:
user_prev_gap = days_dis(user_history_list[k], date_rec_str)
else:
user_prev_gap = 0
if k + 2 < len(user_history_list):
user_next_gap = days_dis(date_rec_str, user_history_list[k + 2])
else:
user_next_gap = 0
k = -1
for j in xrange(len(history_list)):
if history_list[j] == date_rec_str:
##if j == len(history_list) - 1 or history_list[j + 1] > date_rec_str:
break
else:
k += 1
if k > 0:
prev_gap_prev = days_dis(history_list[k], date_rec_str)
else:
prev_gap_prev = 0
if k + 2 < len(history_list):
next_gap_prev = days_dis(date_rec_str, history_list[k + 2])
else:
next_gap_prev = 0
in_between = 0
gaps[i, :] = np.array([
prev_gap, next_gap, user_prev_gap, user_next_gap, in_between,
prev_gap_prev, next_gap_prev
])
X.cat_col(gaps, col_names, ["%s" for i in xrange(len(col_names))])
X.check_point("X_{0}_{1}{2}_gap".format(month, 1, month - 1))