random_seed = int(random_seed)
start = int(start)
stop = int(stop)
if static_use == "True":
# Use non-zero value of penalty
lambda_cv_range = [0, 0.001, 0.01, 0.1]
else:
lambda_cv_range = [0]
A_store = pickle.load(
open(
os.path.expanduser(
'~/git/scalable-nilm/aaai18/predictions/case-{}-graph_{}_{}_{}_{}_As.pkl'
.format(case, source, constant_use, start, stop)), 'r'))
source_df, source_dfc, source_tensor, source_static = create_region_df_dfc_static(
source, year, start, stop)
target_df, target_dfc, target_tensor, target_static = create_region_df_dfc_static(
target, year, start, stop)
# # using cosine similarity to compute L
source_L = get_L(source_static)
target_L = get_L(target_static)
if setting == "transfer":
name = "{}-{}-{}-{}".format(source, target, random_seed, train_percentage)
else:
name = "{}-{}-{}".format(target, random_seed, train_percentage)
# Seasonal constant constraints
if constant_use == 'True':
T_constant = np.ones(stop - start).reshape(-1, 1)
"""
Run all the code on HCDM
"""
import os
import sys
import pickle
import pandas as pd
from common import APPLIANCES_ORDER, compute_rmse_fraction, create_region_df_dfc_static
source, target, start, stop = sys.argv[1:]
start = int(start)
stop = int(stop)
year=2014
target_df, target_dfc, target_tensor, target_static = create_region_df_dfc_static(target, year, start, stop)
df = pd.DataFrame(target_static, index=target_df.index)
idx = df.dropna(how='any').index
out = {}
params = {}
for case in [2, 4]:
out[case] = {}
params[case] = {}
for constant_use in ['True','False']:
out[case][constant_use] = {}
params[case][constant_use] = {}
for static_use in ['True', 'False']:
out[case][constant_use][static_use] = {}
global source, source_df, source_dfc, source_tensor, source_static
global case
global T_constant
global start, stop
global source_L
appliance_index = {appliance: APPLIANCES_ORDER.index(appliance) for appliance in APPLIANCES_ORDER}
APPLIANCES = ['fridge', 'hvac', 'wm', 'mw', 'oven', 'dw']
year = 2014
case, source, constant_use, start, stop = sys.argv[1:]
case = int(case)
start = int(start)
stop = int(stop)
source_df, source_dfc, source_tensor, source_static = create_region_df_dfc_static(source, year, start, stop)
# # using cosine similarity to compute L
source_L = get_L(source_static)
# Seasonal constant constraints
if constant_use == 'True':
T_constant = np.ones(stop-start).reshape(-1,1)
else:
T_constant = None
# End
pred = {}
n_splits = 10
algo = 'adagrad'
