def load_ts_in_range_flattern(start_date=None,
end_date=None,
diff_between=60):
if start_date is None:
start_date = datetime.datetime(2018, 6, 15)
if end_date is None:
end_date = datetime.datetime(2018, 8, 15)
file_name = "data/flattern_ts_in_range-" + start_date.strftime(
"%d_%m_%Y") + "-" + end_date.strftime("%d_%m_%Y") + ".npy"
try:
return np.load(Config.base_dir() +
file_name) # "data/flattern_ts_in_range.npy"
except FileNotFoundError:
pass
configs = DataFactory.load_ts_in_range(start_date=start_date,
end_date=end_date,
diff_between=diff_between)
train = []
for c in configs:
train = np.concatenate([train, c.train.y.values])
np.save(Config.base_dir() + file_name,
train) #"data/flattern_ts_in_range.npy"
return train
def _prebuild_data():
statuses = ["EF", "EO", "EP", "F", "NA", "NF", "O", "P", "PF", "PO"]
channels = ["A", "I", "B"]
data = pd.read_csv(Config.base_dir() + "data/reservations_all.csv",
parse_dates=[
"DATUM_KREIRANJA", "DATUM_OD", "DATUM_DO",
"DATUM_STORNA", "VRIJEME_ZAMRZAVANJA"
],
low_memory=False)
g = data.groupby(["HOTEL", "GODINA", "SIF_REZERVACIJE"])
mx = g["VRIJEME_ZAMRZAVANJA"].transform(max)
data = data[data["VRIJEME_ZAMRZAVANJA"] == mx]
data.STATUS_REZERVACIJE.fillna("F", inplace=True)
data = data[[x in statuses for x in data.STATUS_REZERVACIJE]]
data = data[[x in channels for x in data.KANAL_ID]]
columns = [
"DATUM_KREIRANJA", "DATUM_OD", "DATUM_DO", "BROJ_SOBA_BOOK",
"HOTEL"
]
data = data[columns]
data.to_csv(Config.base_dir() + "data/reservations.csv")
def _folder():
return Config.base_dir() + "data/stats/"
def load_data():
return pd.read_csv(
Config.base_dir() + "data/reservations.csv",
parse_dates=["DATUM_KREIRANJA", "DATUM_OD", "DATUM_DO"],
low_memory=False)
def load_ts(end_date=None, target_date=None, data=None, use_cache=True):
if target_date is None:
target_date = datetime.datetime(2018, 7, 1)
if end_date is None:
end_date = datetime.datetime(2018, 5, 3)
temp_file_path = Config.base_dir(
) + ".temp/config_" + end_date.strftime(
"%d_%m_%Y") + "_" + target_date.strftime("%d_%m_%Y") + ".pkl"
def load_from_temp():
with open(temp_file_path, 'rb') as input:
return pickle.load(input)
def save_to_temp(c):
import os
if not os.path.exists(temp_file_path.split("config")[0]):
os.mkdir(temp_file_path.split("config")[0])
with open(temp_file_path, 'wb') as output:
pickle.dump(c, output, pickle.HIGHEST_PROTOCOL)
def load_data():
return pd.read_csv(
Config.base_dir() + "data/reservations.csv",
parse_dates=["DATUM_KREIRANJA", "DATUM_OD", "DATUM_DO"],
low_memory=False)
try:
if use_cache:
return load_from_temp()
except:
pass
try:
if data is None:
data = load_data()
except FileNotFoundError:
DataFactory._prebuild_data()
data = load_data()
data.DATUM_OD = data.DATUM_OD.dt.normalize()
data.DATUM_DO = data.DATUM_DO.dt.normalize()
data.DATUM_KREIRANJA = data.DATUM_KREIRANJA.dt.normalize()
min_date = target_date.replace(year=target_date.year -
3) #datetime.datetime(2015, 7, 1)
max_date = target_date
data["DATUM_OD_C"] = pd.to_datetime([
datetime.date(2018, 2, 28)
if x == datetime.date(2016, 2, 29) else x.replace(year=2018)
for x in data.DATUM_OD.dt.date.values
])
data["DATUM_DO_C"] = pd.to_datetime([
datetime.date(2018, 2, 28)
if x == datetime.date(2016, 2, 29) else x.replace(year=2018)
for x in data.DATUM_DO.dt.date.values
])
target_date_data = data[(data.DATUM_OD_C <= target_date)
& (data.DATUM_DO_C > target_date)]
range_dates = [
min_date + datetime.timedelta(days=x)
for x in range(0, (max_date - min_date).days + 1)
]
assert min_date == range_dates[0] and max_date == range_dates[
len(range_dates) - 1], "Range of dates generated incorrectly"
elem_list = []
for index in range(len(range_dates)):
obs_date = range_dates[index]
date_data = target_date_data[target_date_data.DATUM_KREIRANJA ==
obs_date]
# Ignore 29.2. and add it to 28.2.
if obs_date.date() == datetime.date(2016, 2, 29):
prev = list(
filter(
lambda s: s["X"].date() == datetime.date(2016, 2, 28),
elem_list))[0]
prev["y"] += date_data.BROJ_SOBA_BOOK.sum()
continue
ts_elem = dict()
ts_elem["X"] = obs_date
ts_elem["y"] = date_data.BROJ_SOBA_BOOK.sum()
elem_list.append(ts_elem)
ts = pd.DataFrame(elem_list)
ts = ts.set_index("X", drop=True)
config = Config.build(ts, end_date, target_date)
if use_cache:
save_to_temp(config)
return config
def load_ts2(target_date=None, end_date=None, use_cache=True):
if target_date is None:
target_date = datetime.datetime(2018, 7, 1)
if end_date is None:
end_date = datetime.datetime(2018, 5, 3)
#end_date = target_date - datetime.timedelta(60)
temp_file_path = Config.base_dir(
) + ".temp/ts2_method_configs_" + end_date.strftime(
"%d_%m_%Y") + "_" + target_date.strftime("%d_%m_%Y") + ".pkl"
def load_from_temp():
with open(temp_file_path, 'rb') as input:
return pickle.load(input)
def save_to_temp(c):
import os
if not os.path.exists(temp_file_path.split("ts2_method")[0]):
os.mkdir(temp_file_path.split("ts2_method")[0])
with open(temp_file_path, 'wb') as output:
pickle.dump(c, output, pickle.HIGHEST_PROTOCOL)
def load():
return pd.read_csv(
Config.base_dir() + "data/reservations.csv",
parse_dates=["DATUM_KREIRANJA", "DATUM_OD", "DATUM_DO"],
low_memory=False)
try:
if use_cache:
return load_from_temp()
except:
pass
try:
data = load()
except:
DataFactory._prebuild_data()
data = load()
#hotels = data.HOTEL.unique()
hotels = ["H" + str(i + 1) for i in range(12)]
configs = list()
for h in hotels:
filtered_data = data[data.HOTEL == h].copy()
config = DataFactory.load_ts(end_date=end_date,
target_date=target_date,
use_cache=False,
data=filtered_data)
"""
plt.plot(config.all)
plt.title(h + " (Total: " + str(config.all.y.sum()) + ")")
plt.savefig(Config.base_dir() + "figures/tmp/" + h + ".png")
plt.close("all")
"""
configs.append({"hotel": h, "config": config})
if use_cache:
save_to_temp(configs)
return configs
