trnslator/tests/test_schedules.py
2020-06-23 11:06:05 -04:00

162 lines
4.6 KiB
Python

import os
import pytest
from translater import (
Schedule,
load_idf,
copy_file,
run_eplus,
config,
get_eplus_dirs,
settings,
)
def test_schedules_in_necb_specific(config):
files = [
"tests/input_data/necb/NECB 2011-MediumOffice-NECB HDD Method-CAN_PQ_Montreal.Intl.AP.716270_CWEC.epw.idf"
]
idfs = {os.path.basename(file): load_idf(file) for file in files}
import matplotlib.pyplot as plt
for key in idfs:
idf = idfs[key]
s = Schedule(
Name="NECB-A-Thermostat Setpoint-Heating", idf=idf, start_day_of_the_week=0
)
s.plot(slice=("2018/01/02", "2018/01/03"), drawstyle="steps-post")
plt.show()
def test_constant_schedule():
const = Schedule.constant_schedule()
assert const.__class__.__name__ == "Schedule"
idf_file = "tests/input_data/schedules/test_multizone_EP.idf"
def schedules_idf():
config(cache_folder="tests/.temp/cache")
idf = load_idf(
idf_file,
include=[
get_eplus_dirs(settings.ep_version)
/ "DataSets"
/ "TDV"
/ "TDV_2008_kBtu_CTZ06.csv"
],
)
return idf
idf = schedules_idf()
schedules = list(idf.get_all_schedules(yearly_only=True).keys())
ids = [i.replace(" ", "_") for i in schedules]
schedules = [
pytest.param(
schedule,
marks=pytest.mark.xfail(
reason="Can't quite capture all possibilities " "with special days"
),
)
if schedule == "POFF"
else pytest.param(schedule, marks=pytest.mark.xfail(raises=NotImplementedError))
if schedule == "Cooling Setpoint Schedule"
else schedule
for schedule in schedules
]
@pytest.fixture(params=schedules, ids=ids)
def test_schedules(request, run_schedules_idf):
"""Create the test_data"""
import pandas as pd
# read original schedule
idf = schedules_idf()
schName = request.param
orig = Schedule(Name=schName, idf=idf)
print(
"{name}\tType:{type}\t[{len}]\tValues:{"
"values}".format(
name=orig.Name,
type=orig.schType,
values=orig.all_values,
len=len(orig.all_values),
)
)
# create year:week:day version
new_eps = orig.to_year_week_day()
new = Schedule(Name=new_eps[0].Name, idf=idf)
index = orig.series.index
epv = pd.read_csv(run_schedules_idf)
epv.columns = epv.columns.str.strip()
epv = epv.loc[:, schName.upper() + ":Schedule Value [](Hourly)"].values
expected = pd.Series(epv, index=index)
print("Year: {}".format(new_eps[0].Name))
print("Weeks: {}".format([obj.Name for obj in new_eps[1]]))
print("Days: {}".format([obj.Name for obj in new_eps[2]]))
yield orig, new, expected
@pytest.fixture(scope="module")
def run_schedules_idf(config):
files = run_eplus(
idf_file,
weather_file="tests/input_data/CAN_PQ_Montreal.Intl.AP" ".716270_CWEC.epw",
annual=True,
readvars=True,
include=[
get_eplus_dirs(settings.ep_version)
/ "DataSets"
/ "TDV"
/ "TDV_2008_kBtu_CTZ06.csv"
],
return_files=True,
)
cache_dir = files[1][0].dirname()
csv = next(iter(cache_dir.glob("*out.csv")))
yield csv
def test_ep_versus_schedule(test_schedules):
"""Main test. Will run the idf using EnergyPlus, retrieve the csv file,
create the schedules and compare"""
orig, new, expected = test_schedules
# slice_ = ('2018/01/01 00:00', '2018/01/08 00:00') # first week
# slice_ = ('2018/05/20 12:00', '2018/05/22 12:00')
slice_ = ("2018-10-05 12:00", "2018-10-07 12:00") # Holiday
# slice_ = ('2018/01/01 00:00', '2018/12/31 23:00') # all year
# slice_ = ('2018/04/30 12:00', '2018/05/01 12:00') # break
mask = expected.values.round(3) != orig.all_values.round(3)
# # region Plot
# fig, ax = plt.subplots(1, 1, figsize=(5, 4))
# orig.plot(slice=slice_, ax=ax, legend=True, drawstyle='steps-post',
# linestyle='dashed')
# new.plot(slice=slice_, ax=ax, legend=True, drawstyle='steps-post',
# linestyle='dotted')
# expected.loc[slice_[0]:slice_[1]].plot(label='E+', legend=True, ax=ax,
# drawstyle='steps-post',
# linestyle='dashdot')
# ax.set_title(orig.Name.capitalize())
# plt.show()
# # endregion
print(orig.series[mask])
assert (
orig.all_values.round(3)[0 : 52 * 7 * 24] == expected.round(3)[0 : 52 * 7 * 24]
).all()
assert (
new.all_values.round(3)[0 : 52 * 7 * 24] == expected.round(3)[0 : 52 * 7 * 24]
).all()