kats_mean_kshift.py

# -*- coding: utf-8 -*-
# """
# Created on 2021/07/25
# @author: Ivan Y.W.Chiu
# """

import pandas as pd
import numpy as np
import os, sys, time, random
from datetime import datetime
from typing import Tuple
import matplotlib as mpl
# mpl.use('Qt5Agg')
from matplotlib import pyplot as plt
plt.ion()
from spc_data_generator import SPCDataConfig, SPCDataGenerator
from spc_chart_plotter import SpcPlotter
from src.anomaly_maker import SimpleOneThingMeanMismatchMaker, SimpleOneThingMeanKShiftMaker
from src.anomaly_detector import SimpleOneThingMeanMismatchDetector

mpl.rcParams.update({"font.size": "10",
                     'lines.linewidth': 1,
                     'lines.markersize': 4})


cfg = SPCDataConfig(
    '2020/11/17 00:00:00.0', '2020/12/18 00:00:00.0', 
    stage_id='stageE11', ctype='MEAN', cnt=300, data_order=0.01, sl=0.03, cl=0.025, target=0)
SPC = SPCDataGenerator(cfg)
SPC.gen()
spc_df = SPC.df.copy()

# %%
mismatch_maker = SimpleOneThingMeanKShiftMaker(spc_df, 'global', 0.333)
new_spc_df = mismatch_maker.gen()

# %%
Plotter = SpcPlotter(spc_df, figsize=(10, 4.5))
Plotter.plot()
fig, ax = Plotter.fig, Plotter.ax

Plotter = SpcPlotter(new_spc_df, figsize=(10, 4.5))
Plotter.plot()
fig1, ax1 = Plotter.fig, Plotter.ax
# new_spc_df.to_csv("data/treding_data_01.csv", index=False)
# %%
# import packages
from kats.consts import TimeSeriesData, TimeSeriesIterator
from kats.detectors.cusum_detection import CUSUMDetector

# %%
# detect increase
df = new_spc_df[['Data_Time', 'Value']]
df.rename(columns={'Data_Time': 'time', 'Value': 'value'}, inplace=True)

timeseries = TimeSeriesData(
    df.loc[:,['time','value']]
)
detector = CUSUMDetector(timeseries)

# run detector
# delta_std_ratio: means how much times std you want to detect for |mu0-mu1|.
change_points = detector.detector(change_directions=["increase"], delta_std_ratio=0.5)

# plot the results
fig3 = plt.figure()
plt.xticks(rotation=45)
detector.plot(change_points)
plt.show()

# %%
#CUMSUM
df['value_residule'] = df['value']-df['value'].mean()
df['value_cumsum'] = np.cumsum(df['value_residule'])

fig4, ax4 = plt.subplots(1, 1)
ax4.plot(df.index, df['value'], 'b', label='value')
ax4.plot(df.index, df['value_residule'], 'y', label='residule')
ax4.plot(df.index, np.ones(len(df))*df['value'].mean(), color='k', linestyle='--', label='mean')
ax4.legend()
fig4.savefig('output/value_residule.png', dpi=300)

# %%
ax4.plot(df.index, df['value_cumsum'], label='cusum')
ax4.legend()
fig4.savefig('output/value_cusum.png', dpi=300)

# %%