def predicition(dmddataset, days):
hodmd = HODMD(svd_rank=0, exact=True, opt=True,
d=len(dmddataset) / days).fit(dmddataset)
hodmd.reconstructed_data.shape
hodmd.plot_eigs()
hodmd.dmd_time['tend'] = len(dmddataset) / days * (
days + 1) - 1 # since it starts from zero
dmd_output = hodmd.reconstructed_data[0].real
dmd_prediction = dmd_output[-len(dmddataset) / days:]
return dmd_prediction, dmd_output
def _get_modes_from_word_vecs(self, v, concat_avg=True):
list_of_modes = []
time_lags = [1, 2]
for d in time_lags:
dmd = HODMD(svd_rank=2, opt=True, exact=True, d=d)
dmd.fit(v.T)
fmode = dmd.modes.T
list_of_modes.append(np.hstack(np.absolute(fmode)))
if concat_avg:
mean_vec = self._p_mean_vector([1.0], v)
list_of_modes.append(mean_vec)
return list_of_modes
def sentence_to_dmd_vec(to_keep, time_lags, concat_avg, vectors):
v = numpy.asarray(vectors)
list_of_modes = []
for d in time_lags:
dmd = HODMD(svd_rank=to_keep, opt=True, exact=True, d=d)
dmd.fit(v.T)
fmode = dmd.modes.T
list_of_modes.append(numpy.hstack(numpy.absolute(fmode)))
if concat_avg:
mean_vec = p_mean_vector([1.0], v)
list_of_modes.append(mean_vec)
return numpy.hstack(list_of_modes)
def dmddiffer(emasets,
days_to_keep,
days_to_use=25,
pointsperday=288): #get the multiple dmd result
dmds = []
for n in np.arange(days_to_use):
dmddataset = []
dmdset = emasets[-days_to_use + n - days_to_keep:-days_to_use + n]
for loop in np.arange(len(dmdset)):
dmddataset.extend(dmdset[loop])
dmddataset = np.array(dmddataset)
hodmd = HODMD(svd_rank=0, exact=True, opt=True, d=288).fit(dmddataset)
hodmd.reconstructed_data.shape
hodmd.dmd_time['tend'] = (days_to_keep + 1) * pointsperday - 1
dmddataset = hodmd.reconstructed_data[0].real[-pointsperday:]
dmds.append(dmddataset)
return dmds
def emddmd(dataset,
d,
draw=0
): #can we do the dmd again? to see whether there are some miracles
emd = EMD()
imfs = emd(dataset) # do the EMD
nimfs = len(imfs)
result = []
sumresult = []
#firstimf = firstone([imfs[0]])
#result.append(firstimf)
for n in np.arange(nimfs):
dmddataset = imfs[n]
hodmd = HODMD(svd_rank=0, exact=True, opt=True, d=d).fit(dmddataset)
hodmd.reconstructed_data.shape
hodmd.dmd_time['tend'] = len(imfs[n]) + 24
dmdresult = hodmd.reconstructed_data[0].real
#uppercap = np.max(dmddataset)*1.1
#lowercap = np.min(dmddataset)*1.1
#if dmdresult[-2] > uppercap:
# dmdresult[-2] = uppercap
#elif dmdresult[-2] < lowercap:
#dmdresult[-2] = lowercap
if draw == 1:
fig = plt.figure(figsize=(20, 10))
plt.plot(np.arange(len(imfs[n])),
imfs[n],
'-',
label='the practical signal',
color='g')
plt.plot(np.arange(len(imfs[n]) + 25),
dmdresult,
'--',
label='DMD output',
color='r')
plt.show()
result.append(dmdresult[-2])
sumresult.append(np.sum(result))
return sumresult, result
(loop + 1) + points_per_day]
daysdata.append(onedaydata)
return daysdata[-days_to_keep:], daysdata
#moving average definition is done
days = 5
cut = iters
madataset = realwindset[:cut]
maresult = ma(madataset, days, pointsperday)
dmddataset_org = maresult[0]
dmddataset = []
for n in np.arange(len(dmddataset_org)):
dmddataset.extend(dmddataset_org[n])
dmddataset = np.array(dmddataset)
hodmd = HODMD(svd_rank=0, exact=True, opt=True,
d=pointsperday).fit(dmddataset)
hodmd.reconstructed_data.shape
hodmd.dmd_time['tend'] = (
days + 1) * pointsperday - 1 # since it starts from zero
#----one pic about the with DMD one
#fig = plt.figure(figsize=(20, 10))
#plt.plot(hodmd.original_timesteps+cut-days*pointsperday, dmddataset, '.', label='snapshots')
data_practical = windsetoriginal[cut + hodmd.dmd_timesteps -
days * pointsperday]
data_prediction = hodmd.reconstructed_data[0].real
#----some special correction on the decomposition-----
zeroindex = np.where(dmddataset_org[-1] <= 0.01)[0]
data_prediction[days * pointsperday +
zeroindex] = dmddataset_org[-1][zeroindex]
dataGridHalf1 = uGridHalf1
dataGridHalf2 = uGridHalf2
# Joined data
tVector = np.hstack((tVectorHalf1, tVectorHalf2))
xGrid, tGrid = np.meshgrid(xVector1, tVector)
uGrid = np.vstack((uGridHalf1, uGridHalf2))
vGrid = np.vstack((vGridHalf1, vGridHalf2))
aGrid = np.vstack((aGridHalf1, aGridHalf2))
uexGrid = np.vstack((uexGridHalf1, uexGridHalf2))
dataGrid = np.vstack((dataGridHalf1, dataGridHalf2))
# Compute DMDs
d = 5
print('Computing DMD 1...')
hodmd1 = HODMD(svd_rank=rank, opt=True, d=d)
hodmd1.fit(dataGridHalf1.T)
print('Computing DMD 2...')
hodmd2 = HODMD(svd_rank=rank, opt=True, d=d)
hodmd2.fit(dataGridHalf2.T)
print('Computing mixed DMD...')
hodmd = HODMD(svd_rank=rank, opt=True, d=d)
hodmd.fit(dataGrid.T)
# Adjust times for partials
hodmd1.original_time['dt'] = hodmd1.dmd_time[
'dt'] = tVectorHalf1[1] - tVectorHalf1[0]
hodmd1.original_time['t0'] = hodmd1.dmd_time['t0'] = tVectorHalf1[0]
hodmd1.original_time['tend'] = hodmd1.dmd_time['tend'] = tVectorHalf1[-1]
# # Plot read data
# fig = plt.figure(figsize=(17,6))
# plt.pcolor(xGrid, tGrid, uGrid)
# plt.title('Displacement')
# plt.colorbar()
# plt.show()
# # print('Done')
print('Computing DMD...')
# ----------------------------------------------------------------------------------------------------------------------
# Compute DMD on displacement
d = 5
dmd = HODMD(svd_rank=rank, opt=True, d=d)
# dmd = MrDMD(svd_rank=rank, max_level = 3, max_cycles = 1)
dmd.fit(dataGrid.T)
# Show eigenvalues
for eig in dmd.eigs:
print('Eigenvalue {}: distance from unit circle {}'.format(eig, np.abs(eig.imag**2+eig.real**2 - 1)))
dmd.plot_eigs(show_axes=True, show_unit_circle=True)
# Show modes
fig = plt.figure(figsize=(8,3))
fig.subplots_adjust(top=0.8, bottom=0.2)
plt.subplot(121)
for mode in dmd.modes.T:
plt.plot(xVector, mode.real)
import numpy as np
import time
from pydmd import HODMD
x = int(input("predicted timestep-\n"))
z = int(input("data cutoff-\n"))
b = slice(0, z)
A = np.loadtxt('dmd.txt', delimiter=",")
B = A.T
C = B[:, b]
original = C
print(C.shape)
hodmd = HODMD(svd_rank=0, exact=True, opt=True, d=30).fit(C)
print(hodmd.reconstructed_data.shape)
hodmd.plot_eigs()
hodmd.original_time['dt'] = hodmd.dmd_time['dt']
hodmd.original_time['t0'] = hodmd.dmd_time['t0']
hodmd.original_time['tend'] = hodmd.dmd_time['tend']
plt.plot(hodmd.original_timesteps, C[0, :], '.', label='snapshots')
plt.plot(hodmd.original_timesteps,
original[0, :],
'-',
label='original function')
plt.plot(hodmd.dmd_timesteps,
hodmd.reconstructed_data[0].real,
'--',
label='DMD output')
composite = np.arange(
int(composite[1]) - 1, int(composite[2]), 1)
else:
composite = np.array(map(int, composite))
newsignal = np.sum(imfs[composite], axis=0)
plt.figure()
plt.plot(x, newsignal)
plt.show()
else:
print(colored("You have to do at least one EMD first.", 'red'))
continue
if name == 'dmd': # do a DMD after a composition among IMFs
if 'newsignal' in locals():
d = raw_input('what is the d in DMD-d?')
x = np.linspace(1, len(newsignal), len(newsignal))
hodmd = HODMD(svd_rank=0, exact=True, opt=True,
d=int(d)).fit(newsignal)
hodmd.reconstructed_data.shape
hodmd.plot_eigs()
hodmd.original_time['dt'] = hodmd.dmd_time['dt'] = x[1] - x[0]
hodmd.original_time['t0'] = hodmd.dmd_time['t0'] = x[0]
hodmd.original_time['tend'] = hodmd.dmd_time['tend'] = x[-1]
#draw
plt.plot(hodmd.dmd_timesteps,
hodmd.reconstructed_data[0].real,
'--',
label='DMD output')
plt.legend()
plt.show()
farseer = raw_input('want to see a forecast?(y/n)')
if farseer == 'y':
# # Plot read data
# fig = plt.figure(figsize=(17,6))
# plt.pcolor(xGrid, tGrid, uGrid)
# plt.title('Displacement')
# plt.colorbar()
# plt.show()
# # print('Done')
print('Computing DMD...')
# ----------------------------------------------------------------------------------------------------------------------
# Compute DMD on displacement
d = 5
dmd = HODMD(svd_rank=rank, opt=True, d=d)
# dmd = MrDMD(svd_rank=rank, max_level=2, max_cycles=2)
dmd.fit(dataGrid.T)
# Show eigenvalues
for eig in dmd.eigs:
print('Eigenvalue {}: distance from unit circle {}'.format(
eig, np.abs(eig.imag**2 + eig.real**2 - 1)))
dmd.plot_eigs(show_axes=True, show_unit_circle=True)
# Show modes
for mode in dmd.modes.T:
plt.plot(xVector, mode.real)
plt.title('Modes')
plt.show()
fucresult = np.cos(x) * np.sin(np.cos(x)) + np.cos(
x * .2) * np.sin(x) + x * (25 - x) / 100
return fucresult
import plotcheck
#just test
#path = "/home/bwei/PycharmProjects/data lib/long wind"
#windset = rd(path)
#name = raw_input('the name of data set?')
x = np.linspace(0, 30, 128)
realwindset = fuc(x)
realwindset.shape = (len(realwindset), )
#realwindset = realwindset + np.random.rand(128)
#x = np.linspace(1, len(realwindset), len(realwindset))
hodmd = HODMD(svd_rank=0, exact=True, opt=True, d=58).fit(realwindset)
hodmd.reconstructed_data.shape
hodmd.plot_eigs()
hodmd.original_time['dt'] = hodmd.dmd_time['dt'] = x[1] - x[0]
hodmd.original_time['t0'] = hodmd.dmd_time['t0'] = x[0]
hodmd.original_time['tend'] = hodmd.dmd_time['tend'] = x[-1]
plt.plot(hodmd.original_timesteps, realwindset, '.', label='realwindset')
plt.plot(x, realwindset, '-', label='original function')
plt.plot(hodmd.dmd_timesteps,
hodmd.reconstructed_data[0].real,
'--',
label='DMD output')
plt.legend()
plt.show()
from plotcheck import pl
import GPy
from matplotlib import pyplot as plt
from pydmd import HODMD
aaa = np.array([1, 2, 3, 4, 5, 6, 8, 19, 30])
bbb = np.array([2, 10, 4, 5, 12, 7])
ccc = np.array([3, 4, 5, 6, 7, 8])
imfs = np.array([])
kkkk = np.array([aaa, bbb, ccc])
Percentile = np.percentile(aaa, [0, 25, 50, 75, 100])
IQR = Percentile[3] - Percentile[1]
UpLimit = Percentile[3] + IQR * 1.5
DownLimit = Percentile[1] - IQR * 1.5
aaa[np.where(aaa > UpLimit)] = UpLimit
avvv = [{}] * 3
path = "/home/bwei/PycharmProjects/data lib/PVhourly6months.csv"
realwindset = readcsv.rd(path, datawashflag=1)
pointsperday = 24
days = 2
hodmd = HODMD(svd_rank=0, exact=True, opt=True,
d=pointsperday).fit(realwindset[-48:])
hodmd.reconstructed_data.shape
hodmd.plot_eigs()
hodmd.dmd_time['tend'] = (days +
1) * pointsperday - 1 # since it starts from zero
import time
from pydmd import HODMD
def myfunc(x):
return np.cos(x)*np.sin(np.cos(x)) + np.cos(x*.2)
x = np.linspace(0, 10, 64)
y = myfunc(x)
snapshots = y
plt.plot(x, snapshots, '.')
plt.show()
hodmd = HODMD(svd_rank=0, exact=True, opt=True, d=30).fit(snapshots)
hodmd.reconstructed_data.shape
hodmd.plot_eigs()
hodmd.original_time['dt'] = hodmd.dmd_time['dt'] = x[1] - x[0]
hodmd.original_time['t0'] = hodmd.dmd_time['t0'] = x[0]
hodmd.original_time['tend'] = hodmd.dmd_time['tend'] = x[-1]
plt.plot(hodmd.original_timesteps, snapshots, '.', label='snapshots')
plt.plot(hodmd.original_timesteps, y, '-', label='original function')
plt.plot(hodmd.dmd_timesteps, hodmd.reconstructed_data[0].real, '--', label='DMD output')
plt.legend()
plt.show()
hodmd.dmd_time['tend'] = 50
x = np.linspace(0, 10, 128)
realwindset = fuc(x)
# toy end
#realwindset = windset[name]
rounds = raw_input('from ?')
rounds = int(rounds)
roundt = raw_input('to?')
roundt = int(roundt)
drange = np.arange(rounds, roundt + 1, 1)
realwindset.shape = (len(realwindset), )
x = np.linspace(1, len(realwindset), len(realwindset))
#new:
for loop in drange:
hodmd = HODMD(svd_rank=0, exact=True, opt=True,
d=int(loop)).fit(realwindset) # key functional sentence
#hodmd.reconstructed_data.shape
errors = realwindset - hodmd.reconstructed_data[0]
sumerror = np.sum(abs(errors))
errorvector.append(sumerror)
y = np.arange(1, len(errorvector) + 1, 1)
plt.plot(y, errorvector, '--', label='errors')
#plt.plot(hodmd.dmd_timesteps, realwindset, '-', label='realwindset')
#plt.plot(hodmd.dmd_timesteps, errors, '.', label='errors')
plt.legend()
plt.show()