def evaluate_submission(submission, intermediate_data_dir):
if isinstance(submission, str):
df_solution = pd.read_csv(submission)
df_solution = df_solution.set_index(pd.to_datetime(df_solution.datetime, utc=True))
else:
assert isinstance(submission, pd.DataFrame), '`submission` must either be a valid submission dataframe or a filepath to the submission'
df_solution = submission
df_real = clean.combine_training_datasets(intermediate_data_dir)
df_real = df_real[df_real.index.isin(df_solution.index)]
df_solution_charge = df_solution.between_time('00:00', '15:00')
df_solution_discharge = df_solution.between_time('15:30', '20:30')
df_real_charge = df_real.between_time('00:00', '15:00')
df_real_discharge = df_real.between_time('15:30', '20:30')
total_score, charge_score, discharge_score = calculate_score_s(df_solution_discharge.charge_MW, df_solution_charge.charge_MW, df_real_discharge.demand_MW, df_real_charge.pv_power_mw)
df_results = pd.DataFrame({
'total_score': total_score,
'charge_score': charge_score,
'discharge_score': discharge_score,
'max_charge_score': max_charge_score(df_real_charge.pv_power_mw)
})
return df_results
def prepare_training_input_data(intermediate_data_dir, start_hour=4):
# Loading input data
df = clean.combine_training_datasets(intermediate_data_dir).interpolate(
limit=1)
df_features = construct_df_charge_features(df)
# Filtering for overlapping feature and target data
dt_idx = pd.date_range(df_features.index.min(),
df['pv_power_mw'].dropna().index.max() -
pd.Timedelta(minutes=30),
freq='30T')
s_pv = df.loc[dt_idx, 'pv_power_mw']
print(s_pv)
df_features = df_features.loc[dt_idx]
# Constructing the charge series
s_charge = construct_charge_s(s_pv,
start_time=f'0{start_hour}:00',
end_time='15:00')
# Filtering for evening datetimes
charging_datetimes = extract_charging_datetimes(df_features,
start_hour=start_hour)
X = df_features.loc[charging_datetimes]
y = s_charge.loc[charging_datetimes]
return X, y
def prepare_training_input_data(intermediate_data_dir):
# Loading input data
df = clean.combine_training_datasets(intermediate_data_dir).interpolate(
limit=1)
df_features = construct_df_discharge_features(df)
# Filtering for overlapping feature and target data
dt_idx = pd.date_range(df_features.index.min(),
df['demand_MW'].dropna().index.max() -
pd.Timedelta(minutes=30),
freq='30T')
s_demand = df.loc[dt_idx, 'demand_MW']
df_features = df_features.loc[dt_idx]
# Constructing the discharge series
s_discharge = construct_discharge_s(s_demand,
start_time='15:30',
end_time='20:30')
# Filtering for evening datetimes
evening_datetimes = extract_evening_datetimes(df_features)
X = df_features.loc[evening_datetimes]
y = s_discharge.loc[evening_datetimes]
return X, y
def prepare_test_feature_data(raw_data_dir, intermediate_data_dir, test_start_date=None, test_end_date=None, start_time='08:00', end_time='23:59'):
# Loading input data
df = clean.combine_training_datasets(intermediate_data_dir).interpolate(limit=1)
df_features = construct_df_charge_features(df)
# Loading default index (latest submission)
if test_end_date is None or test_start_date is None:
index = discharge.load_latest_submission_template(raw_data_dir).index
else:
index = df_features[test_start_date:test_end_date].index
# Filtering feature data on submission datetimes
df_features = df_features.loc[index].between_time(start_time, end_time)
return df_features
def prepare_test_feature_data(raw_data_dir,
intermediate_data_dir,
test_start_date=None,
test_end_date=None):
# Loading input data
df_features = (
clean.combine_training_datasets(intermediate_data_dir).interpolate(
limit=1).pipe(construct_df_discharge_features))
# Loading default index (latest submission)
if test_end_date is None or test_start_date is None:
index = load_latest_submission_template(raw_data_dir).index
else:
index = df_features[test_start_date:test_end_date].index
# Filtering feature data on submission datetimes
df_features = df_features.loc[index]
return df_features