# For Profiling
start = time.time()
for n in num_streams:
for snr in SNRs:
for l in anomaly_lengths:
for m in anomaly_magnitudes:
A = 0
for i in range(initial_conditions):
# Seed random number generator
np.random.seed(i)
# Two ts that have anomalous shift
s0 = Tseries(0)
s1 = Tseries(0)
s0.makeSeries(
[1, 3, 1],
[100, l, 200 - l],
[baseLine, baseLine, baseLine + m],
gradient=float(m) / float(l),
noise_type="none",
)
s1.makeSeries(
[1, 4, 1],
[200, l, 100 - l],
[baseLine, baseLine, baseLine - m],
gradient=float(m) / float(l),
noise_type="none",
)
# -*- coding: utf-8 -*-
"""
Created on Sun Nov 21 13:05:27 2010
@author: musselle
"""
from ControlCharts import Tseries
# import numpy as np
# from matplotlib.pyplot import plot, figure
import scipy
# Initialise stream
series1 = Tseries(0)
series2 = Tseries(0)
series3 = Tseries(0)
series1.makeSeries([3],[5],[2], gradient = 10, noise = 0.000000001)
series2.makeSeries([3],[4],[2], gradient = 10, noise = 0.000000001)
series3.makeSeries([3],[8],[2], gradient = 5, noise = 0.000000001)
series1.makeSeries([2,1,2],[95, 100, 100],[50, 50, 50], amp = 10,
noise = 0.00000001)
series2.makeSeries([2],[296],[40], amp = 10,
noise = 0.000000001, period = 10, phase = 5)
series3.makeSeries([2,4,2],[142, 10, 140],[40, 40, 20], gradient = 2,
amp = 10, noise = 0.00000001)
# For Profiling
start = time.time()
for n in num_streams:
for snr in SNRs:
for l in anomaly_lengths:
for m in anomaly_magnitudes:
A = 0
for i in range(initial_conditions):
# Seed random number generator
np.random.seed(i)
# Two ts that have anomalous shift
s0lin = Tseries(0) # linear component
s0sin = Tseries(0) # Sine component
s1lin = Tseries(0)
s1sin = Tseries(0)
if anomaly_type == 'peak':
s0lin.makeSeries([1,3,4,1], [interval_length, l/2, l/2, 2 * interval_length - l],
[baseLine, baseLine, baseLine + m, baseLine],
gradient = float(m)/float(l/2), noise_type ='none')
s0sin.makeSeries([2], [3 * interval_length], [0.0],
amp = amp, period = period, noise_type ='none')
# sum sin and linear components to get data stream
s0 = np.array(s0lin) + np.array(s0sin)
import time
from PedrosFrahst import frahst_pedro
#===============================================================================
# Runscript
#===============================================================================
#============
# Initialise
#============
'Create time series string'
start = time.time()
# Initialise stream
series1 = Tseries(0)
series2 = Tseries(0)
series3 = Tseries(0)
#series4 = Tseries(0)
series1.makeSeries([1],[300],[5], noise = 1)
series2.makeSeries([1,4,1],[200, 10, 90],[6, 6, -4], gradient = 1, noise = 1)
series3.makeSeries([1,3,1],[100,10,190],[2,2,12], gradient = 1, noise = 1)
#series4.makeSeries([2],[300],[5], period = 10, amp = 1, noise = 1)
streams = scipy.c_[series1, series2, series3] # ,series4]
# Run SPIRIT and Frahst
# Parameters
def genDataMatrix(anomaly_type,
n,
snr,
l,
m,
initial_conditions,
period,
amp,
baseLine=0.0,
baseLine_MA_window=15):
'''Generate dataset matrix with given parameters
Matrix is timesteps x num_strems x initial conditions
'''
A = 0
for i in range(initial_conditions):
# Seed random number generator
np.random.seed(i)
# Two ts that have anomalous shift
s0lin = Tseries(0) # linear component
s0sin = Tseries(0) # Sine component
s1lin = Tseries(0)
s1sin = Tseries(0)
if anomaly_type == 'peak':
s0lin.makeSeries([1, 3, 4, 1], [100, l / 2, l / 2, 200 - l],
[baseLine, baseLine, baseLine + m, baseLine],
gradient=float(m) / float(l / 2),
noise_type='none')
s0sin.makeSeries([2], [300], [0.0],
amp=amp,
period=period,
noise_type='none')
# sum sin and linear components to get data stream
s0 = np.array(s0lin) + np.array(s0sin)
s1lin.makeSeries([1, 4, 3, 1], [200, l / 2, l / 2, 100 - l],
[baseLine, baseLine, baseLine - m, baseLine],
gradient=float(m) / float(l / 2),
noise_type='none')
s1sin.makeSeries([2], [300], [0.0],
amp=amp,
period=period,
noise_type='none')
# sum sin and linear components to get data stream
s1 = np.array(s1lin) + np.array(s1sin)
elif anomaly_type == 'shift':
s0lin.makeSeries([1, 3, 1], [100, l, 200 - l],
[baseLine, baseLine, baseLine + m],
gradient=float(m) / float(l),
noise_type='none')
s0sin.makeSeries([2], [300], [0.0],
amp=amp,
period=period,
noise_type='none')
# sum sin and linear components to get data stream
s0 = np.array(s0lin) + np.array(s0sin)
s1lin.makeSeries([1, 4, 1], [200, l, 100 - l],
[baseLine, baseLine, baseLine - m],
gradient=float(m) / float(l),
noise_type='none')
s1sin.makeSeries([2], [300], [0.0],
amp=amp,
period=period,
noise_type='none')
# sum sin and linear components to get data stream
s1 = np.array(s1lin) + np.array(s1sin)
# The rest of the ts
for k in range(2, n):
name = 's' + str(k)
vars()[name] = Tseries(0)
vars()[name].makeSeries([2], [300], [baseLine],
amp=amp,
period=period,
noise_type='none')
# Concat into one matrix
S = scipy.c_[s0]
for k in range(1, n):
S = scipy.c_[S, vars()['s' + str(k)]]
# Concatonate to 3d array, timesteps x streams x initial condition
if type(A) == int:
A = S # first loop only
else:
A = np.dstack((A, S))
import matplotlib.pyplot as plt from Frahst_v3_1 import FRAHST_V3_1 from SPIRIT import SPIRIT from utils import analysis, QRsolveA, pltSummary # =============================================================================== # Runscript # =============================================================================== # ============ # Initialise # ============ "Create time series string" series_1 = Tseries(0) series_2 = Tseries(0) series_3 = Tseries(0) series_4 = Tseries(0) series_5 = Tseries(0) # S 1 # CYCLIC - Normal - Cyclic series_1.cyclicEt(100, base=6, noise=0.1, amp=2, period=5, phase=0, noise_type="gauss") series_1.normalEt(100, base=6, noise=0.00000000000000001, noise_type="gauss") series_1.cyclicEt(100, base=6, noise=0.1, amp=2, period=5, phase=0, noise_type="gauss") # S 2 # Cyclic series_2.cyclicEt(300, base=6, noise=0.1, amp=2, period=5, phase=1.5, noise_type="gauss") # S 3 # Cyclic
return data
if __name__ == '__main__' :
first = 1
if first:
#s1 = Tseries(0)
s2 = Tseries(0)
#s1.makeSeries([2,1,2], [300, 300, 300], noise = 0.5, period = 50, amp = 5)
#s2.makeSeries([2], [900], noise = 0.5, period = 50, amp = 5)
#data = sp.r_['1,2,0', s1, s2]
s0lin = Tseries(0)
s0sin = Tseries(0)
s2lin = Tseries(0)
s2sin = Tseries(0)
interval_length = 300
l = 10
m = 10
baseLine = 0
amp = 5
period = 50
s0lin.makeSeries([1,3,4,1], [interval_length, l/2, l/2, 2 * interval_length - l],
[baseLine, baseLine, baseLine + m, baseLine],
gradient = float(m)/float(l/2), noise = 0.5)
s0sin.makeSeries([2], [3 * interval_length], [0.0],
amp = amp, period = period, noise = 0.5)
def fractional_decay_MA(data, alpha):
""" MA based on incremetal time fractions """
aved_data = np.zeros_like(data)
u = np.zeros(data.shape[1])
for i in range(data.shape[0]):
t = i + 1
new_data_vec = data[i]
u = ((t - 1) / float(t)) * alpha * u + (1 / float(t)) * new_data_vec
aved_data[i] = u
return aved_data
if __name__ == '__main__':
# Create Series
s0 = Tseries(0)
s1 = Tseries(0)
s2 = Tseries(0)
s0.makeSeries([1, 3, 1], [100, 10, 190],
base=[0, 0, 2],
gradient=0.2,
noise=0.5)
s1.makeSeries([1, 4, 1], [200, 10, 90],
base=[0, 0, -2],
gradient=0.2,
noise=0.5)
s2.makeSeries([1], [300], noise=0.5)
streams = scipy.c_[s0, s1, s2]
# -*- coding: utf-8 -*-
"""
Created on Sat Mar 05 23:19:11 2011
THIS IS THE VERSION TO RUN ON MAC, NOT WINDOWS
@author: musselle
"""
from numpy import eye, zeros, dot, sqrt, log10, trace, arccos, nan, arange
import numpy as np
import numpy.random as npr
import scipy as sp
from numpy.random import rand
from numpy.linalg import qr, eig, norm, solve
from matplotlib.pyplot import plot, figure, title, step
from artSigs import genCosSignals_no_rand , genCosSignals
import scipy.io as sio
from utils import analysis, QRsolveA, pltSummary2
from PedrosFrahst import frahst_pedro_original
from Frahst_v3_1 import FRAHST_V3_1
from load_syn_ping_data import load_n_store
def FRAHST_V3_3(data, r=1, alpha=0.96, e_low=0.96, e_high=0.98, fix_init_Q = 0, holdOffTime=0,
evalMetrics = 'F', static_r = 0, r_upper_bound = None, ignoreUp2 = 0):
"""
Fast Rank Adaptive Householder Subspace Tracking Algorithm (FRAHST)
Version 3.3 - Add decay of S and in the event of vanishing inputs
- Make sure rank of S does not drop (and work out what that means!) - stops S going singular
Version 3.2 - Added ability to fix r to a static value., and also give it an upper bound.
If undefined, defaults to num of data streams.
def genDataMatrix(anomaly_type, n, snr, l, m, initial_conditions, period, amp, baseLine = 0.0, baseLine_MA_window = 15):
'''Generate dataset matrix with given parameters
Matrix is timesteps x num_strems x initial conditions
'''
A = 0
for i in range(initial_conditions):
# Seed random number generator
np.random.seed(i)
# Two ts that have anomalous shift
s0lin = Tseries(0) # linear component
s0sin = Tseries(0) # Sine component
s1lin = Tseries(0)
s1sin = Tseries(0)
if anomaly_type == 'peak':
s0lin.makeSeries([1,3,4,1], [100, l/2, l/2, 200 - l], [baseLine, baseLine, baseLine + m, baseLine],
gradient = float(m)/float(l/2), noise_type ='none')
s0sin.makeSeries([2], [300], [0.0],
amp = amp , period = period, noise_type ='none')
# sum sin and linear components to get data stream
s0 = np.array(s0lin) + np.array(s0sin)
s1lin.makeSeries([1,4,3,1],[200, l/2, l/2, 100 - l],[baseLine, baseLine, baseLine - m, baseLine],
gradient = float(m)/float(l/2), noise_type ='none')
s1sin.makeSeries([2], [300], [0.0],
amp = amp , period = period, noise_type ='none')
# sum sin and linear components to get data stream
s1 = np.array(s1lin) + np.array(s1sin)
elif anomaly_type == 'shift':
s0lin.makeSeries([1,3,1], [100, l, 200 - l], [baseLine, baseLine, baseLine + m],
gradient = float(m)/float(l), noise_type ='none')
s0sin.makeSeries([2], [300], [0.0],
amp = amp, period = period, noise_type ='none')
# sum sin and linear components to get data stream
s0 = np.array(s0lin) + np.array(s0sin)
s1lin.makeSeries([1,4,1],[200, l, 100 - l],[baseLine, baseLine, baseLine - m],
gradient = float(m)/float(l), noise_type ='none')
s1sin.makeSeries([2], [300], [0.0],
amp = amp , period = period, noise_type ='none')
# sum sin and linear components to get data stream
s1 = np.array(s1lin) + np.array(s1sin)
# The rest of the ts
for k in range(2, n) :
name = 's'+ str(k)
vars()[name] = Tseries(0)
vars()[name].makeSeries([2], [300], [baseLine],
amp = amp , period = period, noise_type ='none')
# Concat into one matrix
S = scipy.c_[s0]
for k in range(1, n) :
S = scipy.c_[ S, vars()['s'+ str(k)] ]
# Concatonate to 3d array, timesteps x streams x initial condition
if type(A) == int:
A = S # first loop only
else:
A = np.dstack((A, S))
# Author: C Musselle --<> # Purpose: Generate Synthetic data streams to test for anomaly detection and monitouring applications # Created: 11/18/11 import numpy as np import scipy as sp import matplotlib.pyplot as plt import sys import os from ControlCharts import Tseries """ Code Description: . """ s0 = Tseries(0) s1 = Tseries(0) s2 = Tseries(0) s3 = Tseries(0) s4 = Tseries(0) s5 = Tseries(0) s6 = Tseries(0) s7 = Tseries(0) s8 = Tseries(0) s9_lin = Tseries(0) s9_sin = Tseries(0) s0.makeSeries([2], [1000], amp=1, noise=0.25) s1.makeSeries([2], [1000], amp=1, noise=0.25) s2.makeSeries([2], [1000], amp=1, noise=0.25) s3.makeSeries([2], [1000], amp=1, noise=0.25)
def gen_a_peak_dip(N, T, L, M, pA, k = 5, interval = [10,121], seed = None, noise_sig = 0.1, L2 = None, periods = None, periods2 = None):
""" Adds short peak or dip to data
interval used is (low (Inclusive), High (exclusive)]
"""
if seed:
print 'Setting seed to %i' % (seed)
npr.seed(seed)
if periods is None:
periods = []
num_left = float(interval[1] - interval[0])
num_needed = float(k)
for i in xrange(interval[0], interval[1]+1):
# probability of selection = (number needed)/(number left)
p = num_needed / num_left
if npr.rand() <= p:
periods.append(i)
num_needed -= 1
else:
num_left -=1
#if seed is not None:
## Code to make random sins combination
#A, sins = sin_rand_combo(N, T, periods, seed = seed, noise_scale = noise_sig)
#else:
## Code to make random sins combination
# Seed already set at start of function
A, sins = sin_rand_combo(N, T, periods, noise_scale = noise_sig)
# Anomaly will occur (and finish) Between time points 50 and T - 10
start_point = npr.randint(50, T - L - 10)
# Code to make linear Anomalous trend
baseLine = 0
# Randomly choose peak or dip
if npr.rand() < 0.5:
anom_type = 'Peak'
s0lin = Tseries(0)
s0lin.makeSeries([1,3,4,1], [start_point, L/2, L/2, T - start_point - L],
[baseLine, baseLine, baseLine + M, baseLine],
gradient = float(M)/float(L/2), noise_type ='none')
else:
anom_type = 'Dip'
s0lin = Tseries(0)
s0lin.makeSeries([1,4,3,1], [start_point, L/2, L/2, T - start_point - L],
[baseLine, baseLine, baseLine - M, baseLine],
gradient = float(M)/float(L/2), noise_type ='none')
# Select stream(s) to be anomalous
if type(pA) == int:
num_anom = pA
elif pA < 1.0:
num_anom = np.floor(pA * N)
if num_anom > 1:
anoms = []
num_left = float(N)
num_needed = float(num_anom)
for i in xrange(N):
# probability of selection = (number needed)/(number left)
p = num_needed / num_left
if npr.rand() <= p:
anoms.append(i)
num_needed -= 1
num_left -= 1
else:
num_left -= 1
A[:,anoms] = A[:,anoms] + np.atleast_2d(s0lin).T
else:
anoms = npr.randint(N)
A[:,anoms] = A[:,anoms] + np.array(s0lin)
''' Grount truth Table '''
if anoms.__class__ == int:
anoms = [anoms]
gt_table = np.zeros(num_anom, dtype = [('start','i4'),('loc','i4'),('len','i4'),('mag','i4'),('type','a10')])
for i, a in enumerate(anoms):
gt_table[i] = (start_point, a, L, M, anom_type)
output = dict(data = A, gt = gt_table, trends = sins, periods = periods)
return output
def gen_a_grad_persist(N, T, L, M, pA, k = 5, interval = [10,101], seed = None, noise_sig = 0.1, L2 = None, periods = None, periods2 = None):
""" Adds longer persisted anomaly. gradient up/down to it
and vice versa after L steps """
if seed:
print 'Setting seed to %i' % (seed)
npr.seed(seed)
if periods is None:
periods = []
num_left = float(interval[1] - interval[0])
num_needed = float(k)
for i in xrange(interval[0], interval[1]+1):
# probability of selection = (number needed)/(number left)
p = num_needed / num_left
if npr.rand() <= p:
periods.append(i)
num_needed -= 1
else:
num_left -=1
# Seed already set at start of function
A, sins = sin_rand_combo(N, T, periods, noise_scale = noise_sig)
# Anomaly will occur (and finish) Between time points 50 and T - 10
start_point = npr.randint(50, T - L - L2 - 10)
# Code to make linear Anomalous trend
baseLine = 0
# Randomly choose peak or dip
if npr.rand() < 0.5:
anom_type = 'Grad Up--Down'
s0lin = Tseries(0)
s0lin.makeSeries([1,3,1,4,1], [start_point, L/2, L2, L/2, T - start_point - L - L2],
[baseLine, baseLine, baseLine + M, baseLine + M, baseLine],
gradient = float(M)/float(L/2), noise_type ='none')
else:
anom_type = 'Grad Down--Up'
s0lin = Tseries(0)
s0lin.makeSeries([1,4,1,3,1], [start_point, L/2, L2, L/2, T - start_point - L - L2],
[baseLine, baseLine, baseLine - M, baseLine - M, baseLine],
gradient = float(M)/float(L/2), noise_type ='none')
# Select stream(s) to be anomalous
if type(pA) == int:
num_anom = pA
elif pA < 1.0:
num_anom = np.floor(pA * N)
if num_anom > 1:
anoms = []
num_left = float(N)
num_needed = float(num_anom)
for i in xrange(N):
# probability of selection = (number needed)/(number left)
p = num_needed / num_left
if npr.rand() <= p:
anoms.append(i)
num_needed -= 1
num_left -=1
else:
num_left -=1
A[:,anoms] = A[:,anoms] + np.atleast_2d(s0lin).T
else:
anoms = npr.randint(N)
A[:,anoms] = A[:,anoms] + np.array(s0lin)
''' Ground Truth Table '''
if anoms.__class__ == int:
anoms = [anoms]
gt_table = np.zeros(num_anom*2, dtype = [('start','i4'),('loc','i4'),('len','i4'),('mag','i4'),('type','a10')])
count = 0
for i, a in enumerate(anoms):
a1_type = anom_type[5:].split('--')[0]
a2_type = anom_type[5:].split('--')[1]
gt_table[i+count] = (start_point, a, L, M, 'Grad ' + a1_type)
gt_table[i+count+1] = (start_point+L2+(L/2.0), a, L, M, 'Grad ' + a2_type)
count += 1
output = dict(data = A, gt = gt_table, trends = sins, periods = periods)
return output
# -*- coding: utf-8 -*-
"""
Created on Sat Mar 05 23:19:11 2011
THIS IS THE VERSION TO RUN ON MAC, NOT WINDOWS
@author: musselle
"""
from numpy import eye, zeros, dot, sqrt, log10, trace, arccos, nan, arange
import numpy as np
import numpy.random as npr
import scipy as sp
from numpy.random import rand
from numpy.linalg import qr, eig, norm, solve
from matplotlib.pyplot import plot, figure, title, step
from artSigs import genCosSignals_no_rand, genCosSignals
import scipy.io as sio
from utils import analysis, QRsolveA, pltSummary2
from PedrosFrahst import frahst_pedro_original
from Frahst_v3_1 import FRAHST_V3_1
from load_syn_ping_data import load_n_store
def FRAHST_V3_3(data,
r=1,
alpha=0.96,
e_low=0.96,
e_high=0.98,
fix_init_Q=0,
holdOffTime=0,
evalMetrics='F',
static_r=0,
r_upper_bound=None,
z22 = np.sin(2 * np.pi * t / p22) + npr.randn(t.shape[0]) * noise_scale
z33 = np.sin(2 * np.pi * t / p33) + npr.randn(t.shape[0]) * noise_scale
data = sp.r_['1,2,0', sp.r_[z1, z11], sp.r_[z2, z22], sp.r_[z3, z33]]
return data
if __name__ == '__main__':
first = 1
if first:
#s1 = Tseries(0)
s2 = Tseries(0)
#s1.makeSeries([2,1,2], [300, 300, 300], noise = 0.5, period = 50, amp = 5)
#s2.makeSeries([2], [900], noise = 0.5, period = 50, amp = 5)
#data = sp.r_['1,2,0', s1, s2]
s0lin = Tseries(0)
s0sin = Tseries(0)
s2lin = Tseries(0)
s2sin = Tseries(0)
s3 = Tseries(0)
s4 = Tseries(0)
interval_length = 300
l = 10
m = 10
baseLine = 0
from ControlCharts import Tseries from CUSUM import cusum import numpy as np import matplotlib.pyplot as plt #=============================================================================== # Runscript #=============================================================================== #============ # Initialise #============ 'Create time series string' series_1 = Tseries(0) # NORMAL - normalEt(self, size, base=0, noise, noise_type = 'gauss') series_1.normalEt(50, 0, 1,'gauss') series_1.normalEt(50, 3, 1,'gauss') # CYCLIC - cyclicEt(self, size, base=0, noise=2, amp=10, period=25, \ # noise_type = 'gauss'): series_1.cyclicEt(100, 3, 1, 5, 50, 'gauss') # NORMAL - series_1.normalEt(50, 3, 1,'gauss') series_1.normalEt(50, -3, 1,'gauss') # UP - upEt(self,size,base=0,noise=1,gradient=0.2, noise_type = 'gauss') series_1.upEt(100, -3, 1, 0.2, 'gauss')
# -*- coding: utf-8 -*-
"""
Created on Sat Mar 05 23:19:11 2011
THIS IS THE VERSION TO RUN ON MAC, NOT WINDOWS
@author: musselle
"""
from numpy import eye, zeros, dot, sqrt, log10, trace, arccos, nan, arange, ones
import numpy as np
import scipy as sp
import numpy.random as npr
from numpy.linalg import qr, eig, norm, solve
from matplotlib.pyplot import plot, figure, title, step, ylim
from artSigs import genCosSignals_no_rand, genCosSignals
import scipy.io as sio
from utils import analysis, QRsolveA, pltSummary2
from PedrosFrahst import frahst_pedro_original
from Frahst_v3_1 import FRAHST_V3_1
from Frahst_v3_3 import FRAHST_V3_3
from Frahst_v3_4 import FRAHST_V3_4
from Frahst_v4_0 import FRAHST_V4_0
from load_syn_ping_data import load_n_store
from QR_eig_solve import QRsolve_eigV
from create_Abilene_links_data import create_abilene_links_data
from MAfunctions import MA_over_window
from ControlCharts import Tseries
def FRAHST_V6_3(data,
r=1,
alpha=0.96,
L=1,
from ControlCharts import Tseries from CUSUM import cusum import numpy as np import matplotlib.pyplot as plt #=============================================================================== # Runscript #=============================================================================== #============ # Initialise #============ 'Create time series string' series_1 = Tseries(0) # NORMAL - normalEt(self, size, base=0, noise, noise_type = 'gauss') series_1.normalEt(50, 0, 1, 'gauss') series_1.normalEt(50, 3, 1, 'gauss') # CYCLIC - cyclicEt(self, size, base=0, noise=2, amp=10, period=25, \ # phase = 0, noise_type = 'gauss'): series_1.cyclicEt(100, 3, 1, 5, 50) # NORMAL - series_1.normalEt(50, 3, 1, 'gauss') series_1.normalEt(50, -3, 1, 'gauss') # UP - upEt(self,size,base=0,noise=1,gradient=0.2, noise_type = 'gauss') series_1.upEt(100, -3, 1, 0.2, 'gauss')
count = 1
total = len(num_streams) * len(SNR) * len(anomaly_length) * len(anomaly_magnitude)
for a in num_streams:
for b in SNR:
for c in anomaly_length:
for d in anomaly_magnitude:
A = 0
for e in range(initial_conditions):
# Seed random number generator
np.random.seed(e)
# Two ts that have anomalous shift
s0 = Tseries(0)
s1 = Tseries(0)
s0.makeSeries([1,3,1],[100, c, 200 - c],[baseLine, baseLine, baseLine + d],
gradient = d/c, noise_type ='none')
s1.makeSeries([1,4,1],[200, c, 100 - c],[baseLine, baseLine, baseLine - d],
gradient = d/c, noise_type ='none')
# The rest of the ts
for i in range(2, a) :
name = 's'+ str(i)
vars()[name] = Tseries(0)
vars()[name].makeSeries([1],[300],[5], noise_type ='none')
# concat into one matrix
streams = scipy.c_[s0]
for i in range(1, a) :
streams = scipy.c_[streams, vars()[name] ]
total = len(num_streams) * len(SNR) * len(anomaly_length) * len(
anomaly_magnitude)
for a in num_streams:
for b in SNR:
for c in anomaly_length:
for d in anomaly_magnitude:
A = 0
for e in range(initial_conditions):
# Seed random number generator
np.random.seed(e)
# Two ts that have anomalous shift
s0 = Tseries(0)
s1 = Tseries(0)
s0.makeSeries([1, 3, 1], [100, c, 200 - c],
[baseLine, baseLine, baseLine + d],
gradient=d / c,
noise_type='none')
s1.makeSeries([1, 4, 1], [200, c, 100 - c],
[baseLine, baseLine, baseLine - d],
gradient=d / c,
noise_type='none')
# The rest of the ts
for i in range(2, a):
name = 's' + str(i)
vars()[name] = Tseries(0)
vars()[name].makeSeries([1], [300], [5],
noise_type='none')
score = [] # reset score
tagMetric.append(tag) # list of cumulative sums of Smax
tag = np.zeros((len(stream))) # reset tag for next win size
# Plot results
x = range(1, len(stream))
plt.figure()
plt.subplot(311)
plt.plot(stream)
plt.title('Input Stream')
plt.subplot(312)
for i in range(len(win_sizes)): # i = 1:size(win_sizes,2)
plt.plot(sMaxScore[i]) #'Color', ColOrd(i,:))
plt.title('Smax Score')
plt.subplot(313)
for i in range(len(win_sizes)):
plt.plot(tagMetric[i])
plt.title('Tag')
if __name__ == '__main__':
series = Tseries(0)
series.makeSeries([1,2,3,4] , [100,100,100,100], noise = 4)
cusum_alg(series, [30, 60, 90])
def singleShiftRerun(parameter_string):
params = parameter_string.split(', ')
# Data Sets
baseLine = 0.0
num_streams = [int(params[0].split()[-1])] # n
SNRs = [int(params[1].split()[-1])] # snr
anomaly_lengths = [int(params[2].split()[-1])] # l
anomaly_magnitudes = [int(params[3].split()[-1])] # m
initial_conditions = 20 # i
# Algorithm Parameters
thresholds = params[4].split()[-1].split(',')
e_highs = [float(thresholds[1][0:-1])] # eh
e_lows = [float(thresholds[0][1:])] # el
alphas = [float(params[5].split()[-1])] # a
holdOffs = [int(params[6].split()[-1])] # h
# Algorithm flags
run_spirit = 0
run_frahst = 1
run_frahst_pedro = 0
#===============================================================================
# Initialise Data sets
#========-=======================================================================
# For Profiling
start = time.time()
for n in num_streams:
for snr in SNRs:
for l in anomaly_lengths:
for m in anomaly_magnitudes:
A = 0
for i in range(initial_conditions):
# Seed random number generator
np.random.seed(i)
# Two ts that have anomalous shift
s0 = Tseries(0)
s1 = Tseries(0)
s0.makeSeries([1, 3, 1], [100, l, 200 - l],
[baseLine, baseLine, baseLine + m],
gradient=float(m) / float(l),
noise_type='none')
s1.makeSeries([1, 4, 1], [200, l, 100 - l],
[baseLine, baseLine, baseLine - m],
gradient=float(m) / float(l),
noise_type='none')
# The rest of the ts
for k in range(2, n):
name = 's' + str(k)
vars()[name] = Tseries(0)
vars()[name].makeSeries([1], [300], [baseLine],
noise_type='none')
# Concat into one matrix
S = scipy.c_[s0]
for k in range(1, n):
S = scipy.c_[S, vars()['s' + str(k)]]
# Concatonate to 3d array, timesteps x streams x initial condition
if type(A) == int:
A = S
else:
A = np.dstack((A, S))
# Calculate the noise
Ps = np.sum(A[:, :, 0]**2)
Pn = Ps / (10.**(snr / 10.))
scale = Pn / (n * 300.)
noise = np.random.randn(A.shape[0], A.shape[1],
A.shape[2]) * np.sqrt(scale)
A = A + noise
#===============================================================================
# Ground Truths
#===============================================================================
# # time step | length
ground_truths = np.array([[100, l], [200, l]])
#==============================================================================
# Run Algorithm
#==============================================================================
alg_count = 1
for eh in e_highs:
for el in e_lows:
for a in alphas:
for h in holdOffs:
print 'Running Algorithm(s) with:\nE_Thresh = (' + str(el) + ',' + str(eh) + ')\n' + \
'alpha = ' + str(a) + '\nHoldOff = ' + str(h)
SPIRIT_metricList = []
FRAHST_metricList = []
PEDRO_FRAHST_metricList = []
for i in range(initial_conditions):
# Load Data
streams = A[:, :, i]
if run_spirit == 1:
# SPIRIT
res_sp = SPIRIT(streams,
a, [el, eh],
evalMetrics='F',
reorthog=False,
holdOffTime=h)
res_sp['Alg'] = 'SPIRIT: alpha = ' + str(a) + \
' ,E_Thresh = (' + str(el) + ',' + str(eh) + ')'
pltSummary2(
res_sp, streams, (el, eh))
SPIRIT_metricList.append(
analysis(
res_sp, ground_truths,
300))
# data = res_sp
# Title = 'SPIRIT alpha = ' + str(a) + ' ,E_Thresh = (' + str(el) + ',' + str(eh) + ')'
#
# plot_4x1(streams, data['hidden'], data['e_ratio'], data['orthog_error'],
# ['Input Data','Hidden\nVariables',
# 'Energy Ratio', 'Orthogonality\nError (dB)'] , 'Time Steps', Title)
# plot_4x1(streams, data['hidden'], data['orthog_error'], data['subspace_error'],
# ['Input Data','Hidden\nVariables', 'Orthogonality\nError (dB)','Subspace\nError (dB)'],
# 'Time Steps', Title)
#
if run_frahst == 1:
# My version of Frahst
res_fr = FRAHST_V3_1(
streams,
alpha=a,
e_low=el,
e_high=eh,
holdOffTime=h,
fix_init_Q=1,
r=1,
evalMetrics='F')
res_fr[
'Alg'] = 'MyFrahst: alpha = ' + str(
a
) + ' ,E_Thresh = (' + str(
el) + ',' + str(eh) + ')'
FRAHST_metricList.append(
analysis(
res_fr, ground_truths,
300))
pltSummary2(
res_fr, streams, (el, eh))
#
# data = res_fr
# Title = 'My Frahst with alpha = ' + str(a) + ' ,E_Thresh = (' + str(el) + ',' + str(eh) + ')'
# plot_4x1(streams, data['hidden'], data['orthog_error'], data['subspace_error'],
# ['Input Data','Hidden\nVariables', 'Orthogonality\nError (dB)','Subspace\nError (dB)'],
# 'Time Steps', Title)
if run_frahst_pedro == 1:
# Pedros version of Frahst
res_frped = frahst_pedro(
streams,
alpha=a,
e_low=el,
e_high=eh,
holdOffTime=h,
r=1,
evalMetrics='F')
res_frped['Alg'] = 'Pedros Frahst: alpha = ' + str(a) + \
' ,E_Thresh = (' + str(el) + ',' + str(eh) + ')'
PEDRO_FRAHST_metricList.append(
analysis(
res_frped, ground_truths,
300))
pltSummary2(
res_frped, streams, (el, eh))
#
# data = res_frped
# Title = 'Pedros Frahst: alpha = ' + str(a) + \
# ' ,E_Thresh = (' + str(el) + ',' + str(eh) + ')'
#
# plot_4x1(streams, data['hidden'], data['orthog_error'], data['subspace_error'],
# ['Input Data','Hidden\nVariables', 'Orthogonality\nError (dB)','Subspace\nError (dB)'],
# 'Time Steps', Title)
finish = time.time() - start
print 'Runtime = ' + str(finish) + 'seconds\n'
print 'In H:M:S = ' + GetInHMS(finish)
def singleShiftRerun(parameter_string):
params = parameter_string.split(', ')
# Data Sets
baseLine = 0.0
num_streams = [int(params[0].split()[-1])] # n
SNRs = [int(params[1].split()[-1])] # snr
anomaly_lengths = [int(params[2].split()[-1])] # l
anomaly_magnitudes = [int(params[3].split()[-1])] # m
initial_conditions = 20 # i
# Algorithm Parameters
thresholds = params[4].split()[-1].split(',')
e_highs = [float(thresholds[1][0:-1])] # eh
e_lows = [float(thresholds[0][1:])] # el
alphas = [float(params[5].split()[-1])] # a
holdOffs = [int(params[6].split()[-1])] # h
# Algorithm flags
run_spirit = 0
run_frahst = 1
run_frahst_pedro = 0
#===============================================================================
# Initialise Data sets
#========-=======================================================================
# For Profiling
start = time.time()
for n in num_streams:
for snr in SNRs:
for l in anomaly_lengths:
for m in anomaly_magnitudes:
A = 0
for i in range(initial_conditions):
# Seed random number generator
np.random.seed(i)
# Two ts that have anomalous shift
s0 = Tseries(0)
s1 = Tseries(0)
s0.makeSeries([1,3,1],[100, l, 200 - l],[baseLine, baseLine, baseLine + m],
gradient = float(m)/float(l), noise_type ='none')
s1.makeSeries([1,4,1],[200, l, 100 - l],[baseLine, baseLine, baseLine - m],
gradient = float(m)/float(l), noise_type ='none')
# The rest of the ts
for k in range(2, n) :
name = 's'+ str(k)
vars()[name] = Tseries(0)
vars()[name].makeSeries([1], [300], [baseLine], noise_type ='none')
# Concat into one matrix
S = scipy.c_[s0]
for k in range(1, n) :
S = scipy.c_[ S, vars()['s'+ str(k)] ]
# Concatonate to 3d array, timesteps x streams x initial condition
if type(A) == int:
A = S
else:
A = np.dstack((A, S))
# Calculate the noise
Ps = np.sum(A[:,:,0] ** 2)
Pn = Ps / (10. ** (snr/10.))
scale = Pn / (n * 300.)
noise = np.random.randn(A.shape[0], A.shape[1], A.shape[2]) * np.sqrt(scale)
A = A + noise
#===============================================================================
# Ground Truths
#===============================================================================
# # time step | length
ground_truths = np.array([[100, l],
[200, l]])
#==============================================================================
# Run Algorithm
#==============================================================================
alg_count = 1
for eh in e_highs :
for el in e_lows :
for a in alphas :
for h in holdOffs :
print 'Running Algorithm(s) with:\nE_Thresh = (' + str(el) + ',' + str(eh) + ')\n' + \
'alpha = ' + str(a) + '\nHoldOff = ' + str(h)
SPIRIT_metricList = []
FRAHST_metricList = []
PEDRO_FRAHST_metricList = []
for i in range(initial_conditions):
# Load Data
streams = A[:,:,i]
if run_spirit == 1:
# SPIRIT
res_sp = SPIRIT(streams, a, [el, eh], evalMetrics = 'F',
reorthog = False, holdOffTime = h)
res_sp['Alg'] = 'SPIRIT: alpha = ' + str(a) + \
' ,E_Thresh = (' + str(el) + ',' + str(eh) + ')'
pltSummary2(res_sp, streams, (el, eh))
SPIRIT_metricList.append(analysis(res_sp, ground_truths, 300 ))
# data = res_sp
# Title = 'SPIRIT alpha = ' + str(a) + ' ,E_Thresh = (' + str(el) + ',' + str(eh) + ')'
#
# plot_4x1(streams, data['hidden'], data['e_ratio'], data['orthog_error'],
# ['Input Data','Hidden\nVariables',
# 'Energy Ratio', 'Orthogonality\nError (dB)'] , 'Time Steps', Title)
# plot_4x1(streams, data['hidden'], data['orthog_error'], data['subspace_error'],
# ['Input Data','Hidden\nVariables', 'Orthogonality\nError (dB)','Subspace\nError (dB)'],
# 'Time Steps', Title)
#
if run_frahst == 1:
# My version of Frahst
res_fr = FRAHST_V3_1(streams, alpha=a, e_low=el, e_high=eh,
holdOffTime=h, fix_init_Q = 1, r = 1, evalMetrics = 'F')
res_fr['Alg'] = 'MyFrahst: alpha = ' + str(a) + ' ,E_Thresh = (' + str(el) + ',' + str(eh) + ')'
FRAHST_metricList.append(analysis(res_fr, ground_truths, 300 ))
pltSummary2(res_fr, streams, (el, eh))
#
# data = res_fr
# Title = 'My Frahst with alpha = ' + str(a) + ' ,E_Thresh = (' + str(el) + ',' + str(eh) + ')'
# plot_4x1(streams, data['hidden'], data['orthog_error'], data['subspace_error'],
# ['Input Data','Hidden\nVariables', 'Orthogonality\nError (dB)','Subspace\nError (dB)'],
# 'Time Steps', Title)
if run_frahst_pedro == 1:
# Pedros version of Frahst
res_frped = frahst_pedro(streams, alpha=a, e_low=el, e_high=eh,
holdOffTime=h, r = 1, evalMetrics = 'F')
res_frped['Alg'] = 'Pedros Frahst: alpha = ' + str(a) + \
' ,E_Thresh = (' + str(el) + ',' + str(eh) + ')'
PEDRO_FRAHST_metricList.append(analysis(res_frped, ground_truths, 300 ))
pltSummary2(res_frped, streams, (el, eh))
#
# data = res_frped
# Title = 'Pedros Frahst: alpha = ' + str(a) + \
# ' ,E_Thresh = (' + str(el) + ',' + str(eh) + ')'
#
# plot_4x1(streams, data['hidden'], data['orthog_error'], data['subspace_error'],
# ['Input Data','Hidden\nVariables', 'Orthogonality\nError (dB)','Subspace\nError (dB)'],
# 'Time Steps', Title)
finish = time.time() - start
print 'Runtime = ' + str(finish) + 'seconds\n'
print 'In H:M:S = ' + GetInHMS(finish)
# For Profiling
start = time.time()
for n in num_streams:
for snr in SNRs:
for l in anomaly_lengths:
for m in anomaly_magnitudes:
A = 0
for i in range(initial_conditions):
# Seed random number generator
np.random.seed(i)
# Two ts that have anomalous shift
s0lin = Tseries(0) # linear component
s0sin = Tseries(0) # Sine component
s1lin = Tseries(0)
s1sin = Tseries(0)
if anomaly_type == 'peak':
s0lin.makeSeries([1, 3, 4, 1], [
interval_length, l / 2, l / 2,
2 * interval_length - l
], [baseLine, baseLine, baseLine + m, baseLine],
gradient=float(m) / float(l / 2),
noise_type='none')
s0sin.makeSeries([2], [3 * interval_length], [0.0],
amp=amp,
period=period,
# -*- coding: utf-8 -*-
"""
Created on Wed May 18 11:23:36 2011
Test SPIRIT and Frahst on Contol Time Series: Cyclic
@author: - Musselle
"""
from ControlCharts import Tseries
from CUSUM import cusum
import numpy as np
import matplotlib.pyplot as plt
from Frahst_v3_1 import FRAHST_V3_1
from SPIRIT import SPIRIT
from utils import analysis, QRsolveA, pltSummary
#===============================================================================
# Runscript
#===============================================================================
#============
# Initialise
#============
'Create time series string'
series_1 = Tseries(0)
series_2 = Tseries(0)
series_3 = Tseries(0)
series_4 = Tseries(0)
series_5 = Tseries(0)
# S 1
# CYCLIC - Normal - Cyclic
series_1.cyclicEt(100, base=6, noise=0.1, amp=2, period=5, phase=0, \
noise_type = 'gauss')
series_1.normalEt(100, base=6, noise=0.00000000000000001, noise_type='gauss')
# -*- coding: utf-8 -*-
"""
Created on Wed May 18 11:23:36 2011
Test SPIRIT and Frahst on Contol Time Series: Cyclic
@author: - Musselle
"""
from ControlCharts import Tseries
from CUSUM import cusum
import numpy as np
import matplotlib.pyplot as plt
from Frahst_v3_1 import FRAHST_V3_1
from SPIRIT import SPIRIT
from utils import analysis, QRsolveA, pltSummary
#===============================================================================
# Runscript
#===============================================================================
#============
# Initialise
#============
'Create time series string'
series_1 = Tseries(0)
series_2 = Tseries(0)
series_3 = Tseries(0)
series_4 = Tseries(0)
series_5 = Tseries(0)
# S 1
# CYCLIC - Normal - Cyclic
series_1.cyclicEt(100, base=6, noise=0.1, amp=2, period=5, phase=0, \
noise_type = 'gauss')
series_1.normalEt(100, base=6, noise = 0.00000000000000001, noise_type ='gauss')
# -*- coding: utf-8 -*-
"""
Created on Sat Mar 05 23:19:11 2011
THIS IS THE VERSION TO RUN ON MAC, NOT WINDOWS
@author: musselle
"""
from numpy import eye, zeros, dot, sqrt, log10, trace, arccos, nan, arange, ones
import numpy as np
import scipy as sp
import numpy.random as npr
from numpy.linalg import qr, eig, norm, solve
from matplotlib.pyplot import plot, figure, title, step, ylim
from artSigs import genCosSignals_no_rand , genCosSignals
import scipy.io as sio
from utils import analysis, QRsolveA, pltSummary2
from PedrosFrahst import frahst_pedro_original
from Frahst_v3_1 import FRAHST_V3_1
from Frahst_v3_3 import FRAHST_V3_3
from Frahst_v3_4 import FRAHST_V3_4
from Frahst_v4_0 import FRAHST_V4_0
from load_syn_ping_data import load_n_store
from QR_eig_solve import QRsolve_eigV
from create_Abilene_links_data import create_abilene_links_data
from MAfunctions import MA_over_window
from ControlCharts import Tseries
def FRAHST_V6_3(data, r=1, alpha=0.96, L = 1, holdOffTime=0, evalMetrics = 'F',
EW_mean_alpha = 0.1, EWMA_filter_alpha = 0.3, residual_thresh = 0.1,
F_min = 0.9, epsilon = 0.05,
static_r = 0, r_upper_bound = None,
import scipy import time from PedrosFrahst import frahst_pedro #=============================================================================== # Runscript #=============================================================================== #============ # Initialise #============ 'Create time series string' start = time.time() # Initialise stream series1 = Tseries(0) series2 = Tseries(0) series3 = Tseries(0) #series4 = Tseries(0) series1.makeSeries([1], [300], [5], noise=1) series2.makeSeries([1, 4, 1], [200, 10, 90], [6, 6, -4], gradient=1, noise=1) series3.makeSeries([1, 3, 1], [100, 10, 190], [2, 2, 12], gradient=1, noise=1) #series4.makeSeries([2],[300],[5], period = 10, amp = 1, noise = 1) streams = scipy.c_[series1, series2, series3] # ,series4] # Run SPIRIT and Frahst # Parameters
return aved_data
def fractional_decay_MA(data, alpha):
""" MA based on incremetal time fractions """
aved_data = np.zeros_like(data)
u = np.zeros(data.shape[1])
for i in range(data.shape[0]):
t = i + 1
new_data_vec = data[i]
u = ((t - 1)/float(t)) * alpha * u + (1/float(t)) * new_data_vec
aved_data[i] = u
return aved_data
if __name__ == '__main__':
# Create Series
s0 = Tseries(0)
s1 = Tseries(0)
s2 = Tseries(0)
s0.makeSeries([1,3,1], [100,10,190], base = [0, 0, 2], gradient = 0.2, noise = 0.5)
s1.makeSeries([1,4,1], [200,10,90], base = [0, 0, -2], gradient = 0.2, noise = 0.5)
s2.makeSeries([1], [300], noise = 0.5)
streams = scipy.c_[s0, s1, s2]
N = 25
# Test Moving Averages
MA1 = MA_over_window1(streams, N)
MA2 = MA_over_window2(streams, N)
MA3 = MA_over_window3(streams, N)
# Purpose: Generate Synthetic data streams to test for anomaly detection and monitouring applications # Created: 11/18/11 import numpy as np import scipy as sp import matplotlib.pyplot as plt import sys import os from ControlCharts import Tseries """ Code Description: . """ s0 = Tseries(0) s1 = Tseries(0) s2 = Tseries(0) s3 = Tseries(0) s4 = Tseries(0) s5 = Tseries(0) s6 = Tseries(0) s7 = Tseries(0) s8 = Tseries(0) s9_lin = Tseries(0) s9_sin = Tseries(0) s0.makeSeries([2], [1000], amp = 1, noise = 0.25) s1.makeSeries([2], [1000], amp = 1, noise = 0.25) s2.makeSeries([2], [1000], amp = 1, noise = 0.25) s3.makeSeries([2], [1000], amp = 1, noise = 0.25)
# For Profiling
start = time.time()
for n in num_streams:
for snr in SNRs:
for l in anomaly_lengths:
for m in anomaly_magnitudes:
A = 0
for i in range(initial_conditions):
# Seed random number generator
np.random.seed(i)
# Two ts that have anomalous shift
s0lin = Tseries(0) # linear component
s0sin = Tseries(0) # Sine component
s1lin = Tseries(0)
s1sin = Tseries(0)
if anomaly_type == 'peak':
s0lin.makeSeries([1,3,4,1], [100, l/2, l/2, 200 - l], [baseLine, baseLine, baseLine + m, baseLine],
gradient = float(m)/float(l/2), noise_type ='none')
s0sin.makeSeries([2], [300], [0.0],
amp = amp , period = period, noise_type ='none')
# sum sin and linear components to get data stream
s0 = np.array(s0lin) + np.array(s0sin)
s1lin.makeSeries([1,4,3,1],[200, l/2, l/2, 100 - l],[baseLine, baseLine, baseLine - m, baseLine],
else:
# Repeat last entries
count = count + 1
Q_t.append(Q_t[-1])
S_t.append(S_t[-1])
rr.append(rr[-1])
hiddenV[t-1,:len(hh[0])] = hh[0]
return Q_t, S_t, rr, E_t, E_dash_t, hiddenV, count
if __name__ == '__main__':
# Initialise stream
series1 = Tseries(0)
series2 = Tseries(0)
series3 = Tseries(0)
series1.makeSeries([3],[5],[2], gradient = 10, noise = 0.000000001)
series2.makeSeries([3],[4],[2], gradient = 10, noise = 0.000000001)
series3.makeSeries([3],[8],[2], gradient = 5, noise = 0.000000001)
series1.makeSeries([2,1,2],[95, 100, 100],[50, 50, 50], amp = 10,
noise = 0.00000001)
series2.makeSeries([2],[296],[40], amp = 10,
noise = 0.000000001, period = 10, phase = 5)
series3.makeSeries([2,4,2],[142, 10, 140],[40, 40, 2], gradient = 20,
count = 1
total = len(num_streams) * len(SNR) * len(anomaly_length) * len(anomaly_magnitude)
for n in num_streams:
for snr in SNR:
for l in anomaly_length:
for m in anomaly_magnitude:
A = 0
for e in range(initial_conditions):
# Seed random number generator
np.random.seed(e)
# Two ts that have anomalous shift
s0 = Tseries(0)
s1 = Tseries(0)
if anomaly_type == 'peak':
s0.makeSeries([1,3,4,1], [100, l/2, l/2, 200 - l], [baseLine, baseLine, baseLine + m, baseLine],
gradient = float(m)/float(l/2), noise_type ='none')
s1.makeSeries([1,4,3,1],[200, l/2, l/2, 100 - l],[baseLine, baseLine, baseLine - m, baseLine],
gradient = float(m)/float(l/2), noise_type ='none')
elif anomaly_type == 'shift':
s0.makeSeries([1,3,1],[100, l, 200 - l],[baseLine, baseLine, baseLine + m],
gradient = float(m)/float(l), noise_type ='none')
s1.makeSeries([1,4,1],[200, l, 100 - l],[baseLine, baseLine, baseLine - m],
gradient = float(m)/float(l), noise_type ='none')
# The rest of the ts
for i in range(2, n) :
name = 's'+ str(i)
vars()[name] = Tseries(0)
score = [] # reset score
tagMetric.append(tag) # list of cumulative sums of Smax
tag = np.zeros((len(stream))) # reset tag for next win size
# Plot results
x = range(1, len(stream))
plt.figure()
plt.subplot(311)
plt.plot(stream)
plt.title('Input Stream')
plt.subplot(312)
for i in range(len(win_sizes)): # i = 1:size(win_sizes,2)
plt.plot(sMaxScore[i]) #'Color', ColOrd(i,:))
plt.title('Smax Score')
plt.subplot(313)
for i in range(len(win_sizes)):
plt.plot(tagMetric[i])
plt.title('Tag')
if __name__ == '__main__':
series = Tseries(0)
series.makeSeries([1, 2, 3, 4], [100, 100, 100, 100], noise=4)
cusum_alg(series, [30, 60, 90])