def compute_mean_by_serial_number(self, X):
Xcopy = X.copy()
self.replace_error_with_nan(Xcopy)
self.equipment_average = {}
srl_num_list = Xcopy['UNIT_SRL_NUM'].drop_duplicates().tolist()
print("\t{}: Computing average value of each equipment".format(mark_timestamp()))
for srl_num in srl_num_list:
Xsub = Xcopy[Xcopy['UNIT_SRL_NUM']==srl_num][self.features]
self.equipment_average[srl_num] = Xsub.mean()
print("\t{}: Computing average value of all equipment".format(mark_timestamp()))
self.all_equipment_average = Xcopy[self.features].mean()
del Xcopy
def replace_error_with_nan(self, X):
print("\t{}: Separate sensor value from mean calculation".format(mark_timestamp()))
for c in self.features:
if c in X.columns:
abnormal_idx = X[(X[c]<0) | (X[c]>1E4)].index
if len(abnormal_idx) > 0:
X.loc[abnormal_idx, c] = np.nan
else:
X[c] = np.nan
return X
def fit(self, X, y=None):
print("{}: Fitting Oil Analyisis data for references".format(mark_timestamp()))
X = X[(X['MODL_NUM']==self.unit_model) &
(X['COMPONENT']==self.component) &
(X['HRS_KM_TOT']<self.standard_lifetime) &
(X['HRS_KM_TOT']>=self.unit_hours_bin)
][['HRS_KM_TOT']+self.features]
X['UNIT_HOURS_GROUP'] = X['HRS_KM_TOT'].map(
lambda x: int(self.map_unit_hours(x)/self.unit_hours_bin)*self.unit_hours_bin)
self.reference = X.groupby(['UNIT_HOURS_GROUP'])[self.features].mean()
return self
def replace_error_with_average_serial_number(self, X):
def equipment_exist(srl_num):
if srl_num in self.equipment_catalogue:
return True
else:
return False
print("\t{}: Replacing error value with average for each serial number".format(mark_timestamp()))
for c in self.features:
if c in X.columns:
# find index which data is anomaly
abnormal_idx = X[(X[c].isnull()) | (X[c]<0) | (X[c]>1E4)].index
# replace abnormal datum with mean value
X.loc[abnormal_idx, c] = X.loc[abnormal_idx, 'UNIT_SRL_NUM']\
.map(lambda x: self.get_equipment_average(x).loc[c]
if equipment_exist(x) else self.all_equipment_average.loc[c])
else:
X[c] = self.all_equipment_average.loc[c]
X[c] = X[c].astype(np.double)
return X
def transform(self, X, y=None):
print("{}: Transforming data".format(mark_timestamp()))
self.replace_error_with_average_serial_number(X)
return X
def fit(self, X, y=None):
print("{}: Fitting sensor error scaler with data".format(mark_timestamp()))
self.equipment_catalogue = X['UNIT_SRL_NUM'].drop_duplicates().tolist()
self.compute_mean_by_serial_number(X)
print("\t{}: Finish fitting scaler".format(mark_timestamp()))
return self