Homec4science
Diffusion ML_students_works (master) Browse

ML_students/proj_anomaly_detection_timeseries/helpers_postprocessing.py
ab71824feb99master

Last Change
Raw File

helpers_postprocessing.py
View Options

import numpy as np
from copy import deepcopy
def compute_output(regression_result, df_atypical_weeks, data_per_day=96, data_per_week=672, total_weeks=98):
"""
Create a DataFrame summarizing the results.
For every week (row) count atypical days and number of anomalies in non-atypical days.
:param DataFrame regression_result: DataFrame containing the constructed model
Index is of type DatetimeIndex,
column 'consumption' contains the measurement,
column 'model' contains the prediction made by the model
column 'anomaly' contains a bool indicating if the point is considered as anomalous
column 'atypical day' contains a bool indicating if that measurament
is part of a day detected as atypical in the first phase of the analysis.
:param DataFrame df_atypical_weeks: contains numbers of atypical days for every week
:param int data_per_day: number of measurement for every day (default value is 96, a measurement every 15 minutes)
:param int data_per_week: number of measurement for every week (default value is 672, a measurement every 15 minutes)
:param int total_weeks: number of weeks in the dataset (default value is 98)
:output DataFrame final_table: DataFrame summarizing the results of anomaly detection
Index is the index of the building,
column 'Relative percentage error' contains the realtive error computed on non atypical days,
column 'Percentage of anomalies' contains the percentage of pointwise anaomalies
with respect to all the measurements non contained in atypical days
column 'Percentage of atypical days ' contains the percentage of atypical days with respect to the total
"""
final_table = deepcopy(df_atypical_weeks)
final_table['Anomalies number'] = 0
for row_index in range(regression_result.shape[0]):
# Compute week index (in [0, total_weeks-1])
week_index = np.floor(row_index / data_per_week)
# Skip atypical days from anomalies count
if regression_result.loc[regression_result.index[row_index], 'atypical day'] == True:
continue
# Update counter when anomaly is found
elif regression_result.loc[regression_result.index[row_index], 'anomaly'] == True:
final_table.loc[week_index, 'Anomalies number'] += 1
return final_table
def compute_error(regression_dataframe):
"""
Return the error of the model:mean of the relative error (real_value - predicted_value)/real_value on all the
measurements that are not in atypical days
:param DataFrame regression_result: DataFrame containing the constructed model
Index is of type DatetimeIndex,
column 'consumption' contains the measurement,
column 'model' contains the prediction made by the model
column 'anomaly' contains a bool indicating if the point is considered as anomalous
column 'atypical day' contains a bool indicating if that measurament
is part of a day detected as atypical in the first phase of the analysis.
:output double error_real: percentage relative error
"""
atypical_bool = regression_dataframe.iloc[:, 3].values
normal_values = regression_dataframe.iloc[np.where(atypical_bool == False)[0],0].values
normal_prediction = regression_dataframe.iloc[np.where(atypical_bool == False)[0], 1].values
indexes = np.where(normal_values == 0)[0]
relative_error_real = np.abs(normal_values - normal_prediction)/normal_values
relative_error_real[indexes] = np.abs(normal_values[indexes] - normal_prediction[indexes])/normal_prediction[indexes]
error_real = np.mean(relative_error_real)
return error_real*100
c4science · Help