def plot_piecewise(measure_type, df_all, b, s):
npar = 2
if measure_type == 'gas':
df_reg = df_all.rename(columns={'Temperature_F': '{0}'.format(s), '{0}'.format(b): 'eui_gas', 'Timestamp': 'timestamp'})
ltm.piecewise_reg_one(b, s, npar, 'eui_gas', False, None, df_reg)
elif measure_type == 'electric':
df_reg = df_all.rename(columns={'Temperature_F': '{0}'.format(s), '{0}'.format(b): 'eui_elec', 'Timestamp': 'timestamp'})
ltm.piecewise_reg_one(b, s, npar, 'eui_elec', False, None, df_reg)
def compute_piecewise(measure_type, df_all, b, s):
npar = 2
if measure_type == 'gas':
df_reg = df_all.rename(columns={'Temperature_F': '{0}'.format(s), 'eui': 'eui_gas', 'Timestamp': 'timestamp'})
d = ltm.piecewise_reg_one(b, s, npar, 'eui_gas', False, None, df_reg)
elif measure_type == 'electric':
df_reg = df_all.rename(columns={'Temperature_F': '{0}'.format(s), 'eui': 'eui_elec', 'Timestamp': 'timestamp'})
d = ltm.piecewise_reg_one(b, s, npar, 'eui_elec', False, None, df_reg)
return d
def cv(method, theme, kind):
no_invest = gbs.get_no_invest_set()
hdd_pair = get_pair('HDD')
print hdd_pair[0]
hdd_pair = [x for x in hdd_pair if x[0] in no_invest]
print len(hdd_pair)
n_par_temp = 2
lines = []
if theme == 'eui_elec':
base_temp = '55F'
else:
base_temp = '65F'
lines.append('Building Number,CVRMSE_{0}'.format(method))
for p in hdd_pair:
b = p[0]
s = p[1]
df = pd.read_csv(weatherdir + 'dd_temp_eng/{0}_{1}_{2}.csv'.format(kind, b, s))
if len(df) < 36:
continue
df = df.tail(n=36)
df_pairs = get_partitions(3, df)
errs = []
for x in df_pairs:
df_train = x[0]
if df_train[theme].sum() == 0:
continue
df_test = x[1]
y = np.array(df_test[theme])
if method == 'temperature':
x = np.array(df_test[s])
d = ltm.piecewise_reg_one(b, s, n_par_temp, theme, None, df_train)
y_hat = d['fun'](x, *(d['regression_par']))
elif method == 'dd':
par_list = ld.opt_lireg(b, s, df_train, kind, theme, None)
x = df_test[par_list[-1]]
fun = lambda x: par_list[0] * x + par_list[1]
y_hat = x.map(fun)
elif method == 'vv':
par_list = vv.lean(b, s, df_train, theme, base_temp)
df_temp = df_test.copy()
df_temp = df_temp[['month', base_temp]]
df_perdd = par_list[0]
df_perdd.reset_index(inplace=True)
df_mg = pd.merge(df_test, df_perdd, left_on='month', right_index=True)
y = df_mg[theme]
y_hat = df_mg[base_temp] * df_mg['y_per_dd'] + par_list[-1]
cvrmse = util.CVRMSE(y, y_hat, n_par_temp)
print cvrmse
errs.append(cvrmse)
line = '{0},{1}'.format(b, (np.array(errs)).mean())
print line
lines.append(line)
with open(weatherdir + 'cv_{0}_{1}.csv'.format(method, theme), 'w+') as wt:
wt.write('\n'.join(lines))
return
def load_data_ion(b, s, npar, theme, timestep, timerange):
if timestep == 'D':
df_e = pd.read_csv(interval_daily + '{0}_{1}_D.csv'.format(b, theme))
elif timestep == 'H':
df_e = pd.read_csv(interval_dir + 'single_hourly/{0}.csv'.format(b))
df_e.info()
dates = pd.to_datetime(df_e['Date'])
dayOfWeek = pd.DatetimeIndex(dates).dayofweek
days = ['Mon', 'Tue', 'Wed', 'Thu', 'Fri', 'Sat', 'Sun']
fun = lambda x: days[x]
fun = np.vectorize(fun)
dayOfWeek_str = fun(dayOfWeek)
if timestep == 'D':
df_e.set_index(dates, inplace=True)
elif timestep == 'H':
time_str = df_e['Timestamp'].map(lambda x: x[:18])
time_idx = pd.to_datetime(time_str)
df_e.set_index(time_idx, inplace=True)
print len(df_e)
# fixme, timerange
# df_e = df_e[df_e['Date'] < np.datetime64('2013-09-01')]
print len(df_e)
minDate = dates.min()
maxDate = dates.max()
print minDate, maxDate
minDate_str = minDate.strftime('%Y-%m-%d %H:%M:%S')
maxDate_str = maxDate.strftime('%Y-%m-%d %H:%M:%S')
print minDate_str, maxDate_str
df_w = ltm.get_weather_data(s, minDate_str, maxDate_str, timestep)
df = pd.merge(df_e, df_w, left_index=True, right_index=True, how='left')
df['day'] = dayOfWeek_str
df['hour'] = df['Timestamp'].map(lambda x: int(x[11:13]))
df['year'] = df['Timestamp'].map(lambda x: int(x[:4]))
df['month'] = df['Timestamp'].map(lambda x: int(x[5:7]))
df_plot = df
df_plot = df_plot[df_plot[theme + ' (kBtu)'] >= 0]
if timestep == 'H':
sns.lmplot(x=s, y='{0} (kBtu)'.format(theme), data=df_plot,
fit_reg=False)
plt.ylim((0, 4000))
# plt.gca().set_ylim(bottom=0)
# plt.show()
image_output_dir = os.getcwd() + '/plot_FY_weather/html/single_building/lean_interval/'
P.savefig('{0}scatter_{1}_{2}_{3}_{4}_ori.png'.format(image_output_dir, b, s, theme, timestep), dpi = my_dpi, figsize = (2000/my_dpi, 500/my_dpi))
plt.close()
sns.lmplot(x=s, y='{0} (kBtu)'.format(theme), hue='day',
col='hour', col_wrap=6, size=3, data=df_plot,
fit_reg=False)
plt.ylim((0, 4000))
# plt.gca().set_ylim(bottom=0)
P.savefig('{0}scatter_{1}_{2}_{3}_{4}.png'.format(image_output_dir, b, s, theme, timestep), dpi = my_dpi, figsize = (2000/my_dpi, 500/my_dpi))
plt.close()
# for year in [2014, 2015]:
# df_plot = df[(df['day'] == 'Tue') & (df['year'] == year)]
# sns.lmplot(x=s, y='{0} (kBtu)'.format(theme), hue='month',
# col='hour', col_wrap=6, size=3, data=df_plot, fit_reg=False)
# plt.gca().set_ylim(bottom=0)
# P.savefig('{0}scatter_{1}_{2}_{3}_{4}_{5}_Tue.png'.format(image_output_dir, b, s, theme, timestep, year), dpi = my_dpi, figsize = (2000/my_dpi, 500/my_dpi))
# plt.close()
if theme == 'Gas':
df_reg = df.rename(columns={'{0} (kBtu)'.format(theme): 'eui_gas', 'Date': 'timestamp'})
df_reg['day'] = dayOfWeek_str
ltm.piecewise_reg_one(b, s, npar, 'eui_gas', False, timerange, df_reg)
elif theme == 'Electric':
df_reg = df.rename(columns={'{0} (kBtu)'.format(theme): 'eui_elec', 'Date': 'timestamp'})
df_reg['day'] = dayOfWeek_str
df_reg = df_reg[df_reg['day'] != 'Sat']
df_reg = df_reg[df_reg['day'] != 'Sun']
ltm.piecewise_reg_one(b, s, npar, 'eui_elec', False, timerange, df_reg)
# # plt.show()
return
