def test_sigmoid():
''' Tests difference between relu and sigmoid activation function'''
print('Testing test_sigmoid()')
lstm_dataset = lstmInputDataset(main_dataset, df, num_past_periods=12)
normal_predictions = wahrsager(lstm_dataset, power_dem_df,
TYPE = 'NORMAL',
NAME ='_test_sigmoid',
activation_hidden = 'sigmoid',
activation_end = 'sigmoid',
PLOTTING = True,
).train()
def test_max():
''' Tests difference between different numbers of past periods'''
print('Testing test_max()')
past_periods_list = [6,12,24]
for past_periods in past_periods_list:
lstm_dataset = lstmInputDataset(main_dataset, df, num_past_periods=past_periods)
normal_predictions = wahrsager(lstm_dataset, power_dem_df,
TYPE = 'MAX',
NAME ='_test_max_{}'.format(past_periods),
#PLOTTING = True,
).train()
def test_seq():
''' Tests difference between different numbers of past periods'''
print('Testing test_seq()')
output_periods_list = [6,12,24]
for output_periods in output_periods_list:
lstm_dataset = lstmInputDataset(main_dataset, df, num_past_periods=12)
normal_predictions = wahrsager(lstm_dataset, power_dem_df,
TYPE = 'SEQ',
NAME ='_test_seq_{}'.format(output_periods),
num_outputs = output_periods,
#PLOTTING = True,
).train()
def test_hidden_layers():
'''Tests difference between different numbers of hidden layers'''
print('Testing test_hidden_layers()')
hidden_layers_list = [1,2]
lstm_dataset = lstmInputDataset(main_dataset, df, num_past_periods=12)
for hidden_layers in hidden_layers_list:
normal_predictions = wahrsager(lstm_dataset, power_dem_df,
TYPE = 'NORMAL',
NAME ='_test_hidden_layers_{}'.format(hidden_layers),
num_hidden = hidden_layers,
PLOTTING = True,
).train()
def test_learning_rate():
'''Tests different learning rates'''
print('Testing test_learning_rate()')
learning_rate_list = [0.00001,0.0001,0.001,0.01,0.1,0.5]
lstm_dataset = lstmInputDataset(main_dataset, df, num_past_periods=12)
for learning_rate in learning_rate_list:
normal_predictions = wahrsager(lstm_dataset, power_dem_df,
TYPE = 'NORMAL',
NAME = '_test_learning_rate_{}'.format(learning_rate),
lr = learning_rate,
num_epochs = 1000,
PLOTTING = True,
).train()
def test_lstm_layers():
'''Tests difference between different numbers of lstm layers'''
print('Testing test_lstm_layers()')
lstm_layers_list = [16,32,64,256,512,1028]
lstm_dataset = lstmInputDataset(main_dataset, df, num_past_periods=12)
for lstm_layers in lstm_layers_list:
normal_predictions = wahrsager(lstm_dataset, power_dem_df,
TYPE = 'NORMAL',
NAME ='_test_lstm_layers_{}'.format(lstm_layers),
# Model-Parameter:
lstm_size = lstm_layers,
first_hidden_size = lstm_layers,
PLOTTING = True,
).train()
def test_dropout():
'''Tests overfitting by comparing training loss to validation loss'''
print('Testing test_dropout()')
dropout_list = [0.1,0.2,0.4,0.6]
lstm_dataset = lstmInputDataset(main_dataset, df, num_past_periods=12)
for dropout in dropout_list:
normal_predictions = wahrsager(lstm_dataset, power_dem_df,
TYPE = 'NORMAL',
NAME ='_test_dropout_{}'.format(dropout),
dropout = dropout,
recurrent_dropout = dropout,
num_epochs = 1000,
PLOTTING = True,
).train()
def standart_settings():
'''Creates standart results to compare the tests'''
print('Testing standart_settings()')
lstm_dataset = lstmInputDataset(main_dataset, df, num_past_periods=12)
normal_predictions = wahrsager(lstm_dataset, power_dem_df,
TYPE = 'NORMAL',
NAME ='_test_standart',
#PLOTTING = True,
# Model-Parameter:
num_outputs = 1,
dropout = 0.1,
recurrent_dropout = 0.1,
num_hidden = 3,
lstm_size = 128,
first_hidden_size = 128,
neuron_num_change = 0.5,
activation_hidden = 'relu',
activation_end = 'relu',
lr = 0.001,
# Trainings-Parameter
val_data_size = 2000,
num_epochs = 200,
).train()
def run_wahrsager(NAME,
TYPE='NORMAL',
num_outputs=1,
dropout=0.1,
num_hidden=3,
lstm_size=256,
num_past_periods=24,
activation_hidden='relu',
activation_end='relu',
lr=0.001,
num_epochs=1000):
lstm_dataset = lstmInputDataset(main_dataset,
df,
num_past_periods=num_past_periods)
normal_predictions = wahrsager(
lstm_dataset,
power_dem_df,
TYPE=TYPE,
NAME=NAME,
PLOTTING=True,
# Model-Parameter:
num_outputs=num_outputs,
dropout=dropout,
recurrent_dropout=dropout,
num_hidden=num_hidden,
lstm_size=lstm_size,
first_hidden_size=lstm_size,
neuron_num_change=0.5,
activation_hidden=activation_hidden,
activation_end=activation_end,
lr=lr,
# Trainings-Parameter
val_data_size=2000,
num_epochs=num_epochs,
).train()
def use_heuristic(HEURISTIC_TYPE='Perfekt-Pred-Heuristic',
epochs=1,
threshold_dem=50,
deactivate_SMS=True,
deactivate_LION=True):
# Naming the agent and setting up the directory path:
now = datetime.now()
NAME = str(
round(threshold_dem)) + '_NO_BATTERY_' + HEURISTIC_TYPE + now.strftime(
"_%d-%m-%Y_%H-%M-%S")
D_PATH = '_small_d/'
# Load the dataset:
main_dataset = mainDataset(D_PATH=D_PATH,
period_string_min='15min',
full_dataset=True)
# Normalized dataframe:
df = main_dataset.make_input_df(drop_main_terminal=False,
use_time_diff=True,
day_diff='holiday-weekend')
# Sum of the power demand dataframe (nor normalized):
power_dem_df = main_dataset.load_total_power()[24:-12]
# Load the LSTM input dataset:
lstm_dataset = lstmInputDataset(main_dataset, df, num_past_periods=12)
# Making predictions:
normal_predictions = wahrsager(lstm_dataset, power_dem_df,
TYPE='NORMAL').pred()[:-12]
seq_predictions = wahrsager(lstm_dataset,
power_dem_df,
TYPE='SEQ',
num_outputs=12).pred()
# Adding the predictions to the dataset:
df = df[24:-12]
df['normal'] = normal_predictions
df['seq_max'] = max_seq(seq_predictions)
logger = Logger(NAME, D_PATH)
# Setup reward_maker
r_maker = reward_maker(LOGGER=logger,
COST_TYPE='exact_costs',
R_TYPE='savings_focus',
R_HORIZON='single_step',
cost_per_kwh=0.2255,
LION_Anschaffungs_Preis=34100,
LION_max_Ladezyklen=1000,
SMS_Anschaffungs_Preis=115000 / 3,
SMS_max_Nutzungsjahre=20,
Leistungspreis=102,
logging_list=[
'cost_saving', 'exact_costs', 'sum_exact_costs',
'sum_cost_saving'
],
deactivate_SMS=deactivate_SMS,
deactivate_LION=deactivate_LION)
# Lade Environment:
env = common_env(reward_maker=r_maker,
df=df,
power_dem_df=power_dem_df,
input_list=['norm_total_power', 'normal', 'seq_max'],
max_SMS_SoC=12 / 3,
max_LION_SoC=54,
PERIODEN_DAUER=15,
ACTION_TYPE='contin',
OBS_TYPE='contin',
AGENT_TYPE='heuristic')
agent = heurisitc(env=env,
HEURISTIC_TYPE=HEURISTIC_TYPE,
threshold_dem=threshold_dem)
return agent.calculate(epochs=epochs)