def __init__(self, RedDataFilesList, **kargs):
if kargs.has_key('num_modes'):
self.num_modes = kargs['num_modes']
else:
self.num_modes = len(RedDataFilesList)
Nvars = len(RedDataFilesList[0].data[0, 2:])
N = RedDataFilesList[0].params['N']
podInput = np.array([
rr.data[:, 2:].reshape(rr.params['N'] * len(rr.data[0, 2:]))
for rr in RedDataFilesList
])
modes, self.eig_vals = MR.compute_POD_matrices_snaps_method(
podInput.T, range(self.num_modes))
self.modes = list()
for m in modes.T:
self.modes.append(m.reshape((N, Nvars)))
self.baseRedFile = RedDataFilesList[0].copyForWriting()
print self.baseRedFile.data.shape
print RedDataFilesList[0].data.shape
for v in RedDataFilesList[0].variables[2:]:
self.baseRedFile.appendVariable(v + '_mode')
def ro0_POD_base(rigid, deformable, num_modes=11, verbose=False):
if modred.__version__[0] != '1':
print 'modred version 1 is needed!'
import sys
sys.exit(1)
vecs = numpy.transpose(numpy.array(rigid + deformable))
if verbose:
svec = vecs.shape[0]
nvec = vecs.shape[1]
print 'Calculating', num_modes, 'POD modes for', nvec, 'input vectors of size', svec, '...'
modes, eig_vals = modred.compute_POD_matrices_snaps_method(
vecs, list(range(num_modes)))
#modes, eig_vals = modred.compute_POD_matrices_direct_method(vecs, list(range(num_modes)))
if verbose:
print 'POD eigen values:'
print eig_vals[0:num_modes]
mod = numpy.transpose(modes).tolist()
base = [x for vec in mod for x in vec]
val = eig_vals.tolist()
return (val[0:len(mod)], base)
def ro0_POD_base(rigid, deformable, num_modes=11, verbose=False):
if modred.__version__[0] != '1':
print 'modred version 1 is needed!'
import sys
sys.exit(1)
vecs = numpy.transpose(numpy.array(rigid+deformable))
if verbose:
svec = vecs.shape[0]
nvec = vecs.shape[1]
print 'Calculating', num_modes, 'POD modes for', nvec, 'input vectors of size', svec, '...'
modes, eig_vals = modred.compute_POD_matrices_snaps_method(vecs, list(range(num_modes)))
#modes, eig_vals = modred.compute_POD_matrices_direct_method(vecs, list(range(num_modes)))
if verbose:
print 'POD eigen values:'
print eig_vals[0:num_modes]
mod = numpy.transpose(modes).tolist()
base = [x for vec in mod for x in vec]
val = eig_vals.tolist()
return (val[0:len(mod)], base)
def calculate_pod(saved_modes_number):
with open('data_loaded.txt') as file:
data = np.loadtxt(file)
data_accumulate = data - data.mean(axis=1, keepdims=True)
modes, eigen_values = mr.compute_POD_matrices_snaps_method(
data_accumulate, list(range(saved_modes_number)))
np.savetxt('modes.txt', modes)
np.savetxt('eigen_values.txt', eigen_values)
def ro0_POD_base_keep_rigid(rigid, deformable, num_modes=11, verbose=False):
vecs = numpy.transpose(numpy.array(deformable))
if verbose:
svec = vecs.shape[0]
nvec = vecs.shape[1]
print 'Calculating', num_modes - 6, 'POD modes for', nvec, 'input vectors of size', svec, '...'
modes, eig_vals = modred.compute_POD_matrices_snaps_method(
vecs, list(range(num_modes - 6)))
#modes, eig_vals = modred.compute_POD_matrices_direct_method(vecs, list(range(num_modes-6)))
if verbose:
print 'POD eigen values:'
print eig_vals[0:num_modes - 6]
defo = numpy.transpose(modes).tolist()
# re-orthogonalize deformable modes with
# respect to the 6 rigid modes and themselves
for i in range(0, len(defo)):
for j in range(0, i): # defo(i) _|_ defo(j<i)
x = defo[i]
y = defo[j]
dot = numpy.dot(x, y)
z = numpy.array(x) - dot * numpy.array(y)
defo[i] = z.tolist()
for y in rigid: # defo(i) _|_ rigid
x = defo[i]
dot = numpy.dot(x, y)
z = numpy.array(x) - dot * numpy.array(y)
defo[i] = z.tolist()
# normalize
x = defo[i]
invlen = 1.0 / numpy.dot(x, x)**0.5
z = numpy.array(x) * invlen
defo[i] = z.tolist()
'''
for i in range(0,len(defo)):
for j in range(0,len(defo)):
x = defo[i]
y = defo[j]
dot = numpy.dot(x, y)
print 'dot(%d,%d) = %g' % (i, j, dot)
for y in rigid:
x = defo[i]
dot = numpy.dot(x, y)
print 'dot(%d,rigid) = %g' % (i, dot)
'''
base = [x for vec in (rigid + defo) for x in vec]
val = eig_vals.tolist()
return ([0.] * 6 + [1.] * (num_modes - 6), base)
def ro0_POD_base_keep_rigid(rigid, deformable, num_modes=11, verbose=False):
vecs = numpy.transpose(numpy.array(deformable))
if verbose:
svec = vecs.shape[0]
nvec = vecs.shape[1]
print 'Calculating', num_modes-6, 'POD modes for', nvec, 'input vectors of size', svec, '...'
modes, eig_vals = modred.compute_POD_matrices_snaps_method(vecs, list(range(num_modes-6)))
#modes, eig_vals = modred.compute_POD_matrices_direct_method(vecs, list(range(num_modes-6)))
if verbose:
print 'POD eigen values:'
print eig_vals[0:num_modes-6]
defo = numpy.transpose(modes).tolist()
# re-orthogonalize deformable modes with
# respect to the 6 rigid modes and themselves
for i in range(0,len(defo)):
for j in range(0, i): # defo(i) _|_ defo(j<i)
x = defo[i]
y = defo[j]
dot = numpy.dot(x, y)
z = numpy.array(x) - dot*numpy.array(y)
defo[i] = z.tolist()
for y in rigid: # defo(i) _|_ rigid
x = defo[i]
dot = numpy.dot(x, y)
z = numpy.array(x) - dot*numpy.array(y)
defo[i] = z.tolist()
# normalize
x = defo[i]
invlen = 1.0/numpy.dot(x, x)**0.5
z = numpy.array(x) * invlen
defo[i] = z.tolist()
'''
for i in range(0,len(defo)):
for j in range(0,len(defo)):
x = defo[i]
y = defo[j]
dot = numpy.dot(x, y)
print 'dot(%d,%d) = %g' % (i, j, dot)
for y in rigid:
x = defo[i]
dot = numpy.dot(x, y)
print 'dot(%d,rigid) = %g' % (i, dot)
'''
base = [x for vec in (rigid+defo) for x in vec]
val = eig_vals.tolist()
return ([0.]*6+[1.]*(num_modes-6), base)
def decompose(self, nMode, method='snap', subtractMean=False):
'''
Compute the Porper Orthogonal Decomposition (POD) from a list of N snapshots.
The snapshots are PIV surfaces of size (surfX,surfY) of field F. F can be any
scalar field (Ux, ux, T, vorticity, R11,...)
Arguments:
*nMode*: python integer.
Number of modes of the POD.
*PODmethod*: python string. Default='snap'
Type of POD algorithm used. For the snapshot method, use PODmethod='snap' and
for the direct method, use PODmewthod='direct'. The default value is 'snap'.
How to choose between 'snap' and 'direct': if surfX*surfY > N**2, use the snap
method, it should increase the coputational speed for only a little loss in
precision.
Returns:
None
'''
nSnap = self.vecs.shape[1]
self.result['nMode'] = nMode
self.result['nSnap'] = nSnap
if np.any(self.vecs):
self.vecs = np.nan_to_num(self.vecs)
if subtractMean:
self.vecs = self.vecs - np.mean(self.vecs, axis=1, keepdims=True)
if method == 'snap':
modesPOD, eigVals = modred.compute_POD_matrices_snaps_method(
self.vecs, range(nMode))
elif method == 'direct':
modesPOD, eigVals = modred.compute_POD_matrices_direct_method(
self.vecs, range(nMode))
else:
print('error: argument ' + str(method) +
' is not valid. Use \'snap\' or \'direct\'')
# convert modesPOD in an numpy array
modesPOD = np.asarray(modesPOD)
self.result['raw_modes'] = modesPOD
self.result['eigVals'] = eigVals
ai = self.projectOnMode(self.vecs)
self.result['ai'] = ai
def decompose(self,nMode,method='snap',subtractMean=False):
'''
Compute the Porper Orthogonal Decomposition (POD) from a list of N snapshots.
The snapshots are PIV surfaces of size (surfX,surfY) of field F. F can be any
scalar field (Ux, ux, T, vorticity, R11,...)
Arguments:
*nMode*: python integer.
Number of modes of the POD.
*PODmethod*: python string. Default='snap'
Type of POD algorithm used. For the snapshot method, use PODmethod='snap' and
for the direct method, use PODmewthod='direct'. The default value is 'snap'.
How to choose between 'snap' and 'direct': if surfX*surfY > N**2, use the snap
method, it should increase the coputational speed for only a little loss in
precision.
Returns:
None
'''
nSnap = self.vecs.shape[1]
self.result['nMode']=nMode
self.result['nSnap']=nSnap
if np.any(self.vecs):
self.vecs = np.nan_to_num(self.vecs)
if subtractMean:
self.vecs=self.vecs-np.mean(self.vecs,axis=1,keepdims=True)
if method=='snap':
modesPOD, eigVals = modred.compute_POD_matrices_snaps_method(self.vecs, range(nMode))
elif method=='direct':
modesPOD, eigVals = modred.compute_POD_matrices_direct_method(self.vecs, range(nMode))
else:
print('error: argument '+str(method)+' is not valid. Use \'snap\' or \'direct\'')
# convert modesPOD in an numpy array
modesPOD = np.asarray(modesPOD)
self.result['raw_modes']=modesPOD
self.result['eigVals']=eigVals
ai=self.projectOnMode(self.vecs)
self.result['ai']=ai
def compare_modred_sparse():
"""Compare output from modred with our sparse method."""
import modred as mr
import gc_turbulence as g
run = g.ProcessedRun(g.default_processed + 'r13_12_16a.hdf5')
# slice with no nans
# ( or use # complement(find_nan_slice(run.Uf_[:])) )
good_slice = (slice(None), slice(None), slice(46L, None))
data = run.Uf_[good_slice]
snapshots = sparse_dmd.SparseDMD.to_snaps(data, decomp_axis=1)
modes, ritz_values, norms \
= mr.compute_DMD_matrices_snaps_method(snapshots, slice(None))
pmodes, pnorms \
= mr.compute_POD_matrices_snaps_method(snapshots, slice(None))
dmd = sparse_dmd.SparseDMD(snapshots)
def compare_modred_sparse():
"""Compare output from modred with our sparse method."""
import modred as mr
import gc_turbulence as g
run = g.ProcessedRun(g.default_processed + 'r13_12_16a.hdf5')
# slice with no nans
# ( or use # complement(find_nan_slice(run.Uf_[:])) )
good_slice = (slice(None), slice(None), slice(46L, None))
data = run.Uf_[good_slice]
snapshots = sparse_dmd.SparseDMD.to_snaps(data, decomp_axis=1)
modes, ritz_values, norms \
= mr.compute_DMD_matrices_snaps_method(snapshots, slice(None))
pmodes, pnorms \
= mr.compute_POD_matrices_snaps_method(snapshots, slice(None))
dmd = sparse_dmd.SparseDMD(snapshots)
def __init__(self, RedDataFilesList, **kargs):
if kargs.has_key('num_modes'):
self.num_modes = kargs['num_modes']
else:
self.num_modes = len(RedDataFilesList)
Nvars = len(RedDataFilesList[0].data[0,2:])
N = RedDataFilesList[0].params['N']
podInput = np.array([ rr.data[:,2:].reshape(rr.params['N']*len(rr.data[0,2:])) for rr in RedDataFilesList ])
modes, self.eig_vals = MR.compute_POD_matrices_snaps_method(podInput.T, range(self.num_modes))
self.modes = list()
for m in modes.T:
self.modes.append( m.reshape((N,Nvars)) )
self.baseRedFile = RedDataFilesList[0].copyForWriting()
print self.baseRedFile.data.shape
print RedDataFilesList[0].data.shape
for v in RedDataFilesList[0].variables[2:]:
self.baseRedFile.appendVariable(v+'_mode')
datarray = np.array([data(np.random.random(4)) for i in range(nr_data)])
true_vx = np.array([0.45, 0.55, 0.75, 0.35])
datb_true = data(true_vx)
datb = datb_true.copy()
datb[30:70] *= np.NaN
#
#
# plot the data (snapshots in black and missing in red)
pl.figure(1)
pl.plot(datarray.T, 'k', alpha=0.5)
pl.plot(datb, 'r', lw=3)
pl.title('Data snapshots used (black) and gappy data (red)')
#
#
# decompose datarray using the method of snapshots
modes, eig_vals, eig_vecs = mr.compute_POD_matrices_snaps_method(
datarray.T, range(nr_reconstr), return_all=True)[:3]
# modes used for reconstruction
scaled_modes = modes.copy()
for i in range(nr_reconstr):
scaled_modes[:, i] *= np.sqrt(eig_vals[i])
# plot the scaled modes used in reconstruction:
pl.figure(2)
for i in range(nr_reconstr):
pl.plot(scaled_modes[:, i], label='Mode %i' % (i + 1))
pl.legend()
pl.title('Scaled (proper orthogonal) modes')
#
#
#
# used masked modes in reconstruction:
mask = np.where(np.isfinite(datb))[0]
import numpy as N import modred as MR num_vecs = 30 # Arbitrary data vecs = N.random.random((100, num_vecs)) num_modes = 5 modes, eig_vals = MR.compute_POD_matrices_snaps_method(vecs, range(num_modes))
from future.builtins import range import numpy as np import modred as mr num_vecs = 30 # Arbitrary data vecs = np.random.random((100, num_vecs)) num_modes = 5 modes, eig_vals = mr.compute_POD_matrices_snaps_method(vecs, list(range(num_modes)))
def main(argv):
import vtk
from vtk.numpy_interface import dataset_adapter as dsa
from vtk.numpy_interface import algorithms as algs
import numpy as np
### get parameters
import os
if not os.path.exists(OD):
os.makedirs(OD)
print '!!!Output to DIR: ',OD
if not read_fields_from_file:
### Readin stage
# using parallel openfoam reader
ofr = vtk.vtkPOpenFOAMReader()
# set reader's options
ofr.SetFileName(ID+IF)
print '!!!open file: ',ID+IF
ofr.SetDecomposePolyhedra(0)
ofr.CacheMeshOn()
ofr.SetCreateCellToPoint(0)
ofr.DisableAllCellArrays()
ofr.SetCellArrayStatus(fieldname,1)
ofr.Update()
# VTKArray is same as numpy array
times = dsa.vtkDataArrayToVTKArray( ofr.GetTimeValues() ,ofr)
# select the timestep between t0 and tf
times = [t for t in times if t>=t0 and t<=tf]
print '!!!available time steps: ',times
N = len(times)
np.save(times_filename,times)
# using CellQuality to get cell's volumes as weight
cq = vtk.vtkCellQuality()
cq.SetInputConnection(0,ofr.GetOutputPort(0))
cq.SetQualityMeasureToVolume()
cq.Update()
# cq is a composite dataset so I need GetBlock(0)
geom = cq.GetOutputDataObject(0).GetBlock(0)
# get volumes of cells V, size = L (number of cells)
V = np.copy(dsa.WrapDataObject(geom).CellData['CellQuality'])
# normalize it as weight
Vtotal = sum(V)
V /= Vtotal
# delete all other CellDataArray in geom DataSet, preserve its mesh structure and topology structure
for i in range(geom.GetCellData().GetNumberOfArrays()):
geom.GetCellData().RemoveArray(0)
# add volume weight to it for saving
geom.GetCellData().AddArray(dsa.numpyTovtkDataArray(V,'vol_weight'))
# using *.vtu file format to save the vol_weight
ugw = vtk.vtkXMLUnstructuredGridWriter()
ugw.SetInputDataObject(geom)
print '!!!Output vol_weight to file: ',geom_filename
ugw.SetFileName(geom_filename)
# using binary format
ugw.SetDataModeToBinary()
# enable compression
ugw.SetCompressorTypeToZLib()
# write to the file
ugw.Write()
# disconnect cq and ofr in order to isolate this dataset object from Update()
cq.RemoveAllInputConnections(0)
L = V.size # number of cells
N = len(times) #number of timesteps
# vector data is larger in size
if field_is_vector == True:
fields = np.zeros([N,L,3])
else:
fields = np.zeros([N,L])
pipepout = ofr
for i in range(N):
t = times[i]
print '!!!reading time:{}'.format(t)
# set time value
pipepout.GetOutputInformation(0).Set(vtk.vtkStreamingDemandDrivenPipeline.UPDATE_TIME_STEP(),t)
# read in field data of new timestep
pipepout.Update()
#
d = dsa.WrapDataObject(pipepout.GetOutput().GetBlock(0))
print '!!!reading field:{}'.format(fieldname)
field = d.CellData[fieldname]
# get the first component of composite dataset, it is the internalField
fields[i]=np.copy(field)
# write data to file
print '!!!write field data to file:',fields_filename
np.savez(fields_filename,fields)
else: #read fields from file
fields = np.load(fields_filename)['arr_0']
ugr = vtk.vtkXMLUnstructuredGridReader()
ugr.SetFileName(geom_filename)
ugr.Update()
geom = ugr.GetOutputDataObject(0)
V = np.copy(dsa.WrapDataObject(geom).CellData['vol_weight'])
times = np.load(times_filename)
assert times.shape[0] == fields.shape[0]
assert fields.shape[1] == V.shape[0]
N = times.shape[0]
L = fields.shape[1]
print 'Read in dataset complete'
### POD section
# calculate average
field_avg = np.average(fields, axis=0)
if subtractAvg:
fields = fields - field_avg
import modred as mr
if do_POD:
# if field is a vector, reshape the fields and corresponding volument weight
if field_is_vector:
shp_vec = fields.shape
shp_flat = (fields.shape[0],fields.shape[1]*fields.shape[2])
fields = fields.reshape(shp_flat)
V = np.tile(V,shp_vec[2])
# POD
print '!!!Doing POD analysis'
modes, eigen_vals, eigen_vecs, correlation_mat = mr.compute_POD_matrices_snaps_method(fields.T,range(M),inner_product_weights=V,return_all=True)
# if field is a vector, reshape the output matrix
if field_is_vector:
fields = fields.reshape(shp_vec)
modes = np.asarray(modes).T.reshape((modes.shape[1],shp_vec[1],shp_vec[2]))
V = V[:shp_vec[1]]
if output_correlation_matrix:
print "!!!output POD correlation matrix",POD_cm_filename
np.savetxt(POD_cm_filename,correlation_mat,delimiter=',')
if output_POD_temporal_modes:
print "!!!output POD temporal modes",POD_tm_filename
# output temporal modes
singular_vals = eigen_vals**0.5
POD_mode_energy_normalized = eigen_vals/correlation_mat.trace()[0,0]
cumsum_POD_mode_energy_normalized = np.cumsum(POD_mode_energy_normalized)
# generate header string
header_str = 'temporal modes\n'
header_str += 'time,eigen value,singular value,normalized eigen value,accumulated normalized eigen value'
for i in range(N-1):
header_str += ',Mode{}'.format(i)
header_str += '\n'
for i in range(N-1):
header_str += ',SV ={}'.format(singular_vals[i])
header_str += '\n'
for i in range(N-1):
header_str += ',EV ={}'.format(eigen_vals[i])
header_str += '\n'
for i in range(N-1):
header_str += ',NEnergy ={}'.format(POD_mode_energy_normalized[i])
header_str += '\n'
for i in range(N-1):
header_str += ',CumsumEnergy ={}'.format(cumsum_POD_mode_energy_normalized[i])
header_str += '\n'
np.savetxt(POD_tm_filename, \
np.c_[times, \
eigen_vecs], \
delimiter = ',', \
header = header_str)
if output_POD_spatial_modes:
print "!!!output POD spatial Modes to ",POD_sm_filename
#output to xml vtk unstructured grid file
ugcd = geom.GetCellData()
ugcd.Reset()
ugcd.CopyAllOff()
for i in range(ugcd.GetNumberOfArrays()):
ugcd.RemoveArray(0)
# import POD mode
for i in range(M):
ugcd.AddArray(dsa.numpyTovtkDataArray(modes[i],prefix+'_POD_mode_{}_{}'.format(fieldname,i)))
# add average field
ugcd.AddArray(dsa.numpyTovtkDataArray(field_avg,'field_{}_avg'.format(fieldname)))
ugw = vtk.vtkXMLUnstructuredGridWriter()
ugw.SetInputDataObject(geom)
ugw.SetFileName(POD_sm_filename)
ugw.Write()
if doReconstruction:
print "!!! do Reconstrution with {} POD modes at time {}".format(MR,ReconTime)
#get an empty mesh
ugcd = geom.GetCellData()
ugcd.Reset()
ugcd.CopyAllOff()
for i in range(ugcd.GetNumberOfArrays()):
ugcd.RemoveArray(0)
# reconstruct from first MR POD modes
#
ReconN = np.searchsorted(times,ReconTime)
print "!!!actually, reconstruction is done at time {} rather than time {}".format(times[ReconN],ReconTime)
recon_field = np.einsum("i...,i,i",modes[:MR],eigen_vals[:MR]**0.5,np.asarray(eigen_vecs)[ReconN,:MR])+field_avg;
ugcd.AddArray(dsa.numpyTovtkDataArray(recon_field,prefix+'_POD_{}_Reconstructed_{}_{}'.format(MR,fieldname,ReconTime)))
ugw = vtk.vtkXMLUnstructuredGridWriter()
ugw.SetInputDataObject(geom)
ugw.SetFileName(POD_reconstruction_filename)
ugw.Write()
if do_DMD:
print "!!!Begin to calculate DMD modes"
# if field is a vector, reshape the fields and corresponding volument weight
if field_is_vector:
shp_vec = fields.shape
shp_flat = (fields.shape[0],fields.shape[1]*fields.shape[2])
fields = fields.reshape(shp_flat)
V = np.tile(V,shp_vec[2])
# DMD, I do not know which mode is important, so I have to discard modes_
modes_, ritz_vals, mode_norms, build_coeffs = mr.compute_DMD_matrices_snaps_method(fields.T,[],inner_product_weights=V,return_all=True)
# if field is a vector, reshape the fields, V and output matrix
if field_is_vector:
fields = fields.reshape(shp_vec)
V = V[:shp_vec[1]]
# sorting
eorder = np.argsort(mode_norms)[::-1]
# re-order the outputs
ritz_vals = ritz_vals[eorder]
mode_norms = mode_norms[eorder]
build_coeffs = build_coeffs[:,eorder]
#build the DMD_modes
DMD_modes = np.einsum('ijk,il->ljk', fields,build_coeffs[:,:M_DMD])
if output_DMD_info:
print "!!!output DMD info to :",DMD_info_filename
# output modes info
header_str = 'DMD modes info\n'
header_str += 'ritz_vals.real,ritz_vals.imag,growth_rate, frequency, mode_norms\n'
header_str += r'AU,AU,1/s, Hz, AU'
dt = np.average(times[1:]-times[:-1]) #time step
np.savetxt(DMD_info_filename, \
np.c_[ np.real(ritz_vals), \
np.imag(ritz_vals), \
np.log(np.abs(ritz_vals))/dt, \
np.angle(ritz_vals)/dt, \
mode_norms], \
delimiter = ',', \
header = header_str)
if output_DMD_build_coeffs:
print "!!!output DMD build coeffs. to :",DMD_build_coeffs_filename
np.savez(DMD_build_coeffs_filename, build_coeffs)
if output_DMD_spatial_modes:
print "!!!output DMD info to :",DMD_sm_filename
#output to xml vtk unstructured grid file
ugcd = geom.GetCellData()
ugcd.Reset()
ugcd.CopyAllOff()
for i in range(ugcd.GetNumberOfArrays()):
ugcd.RemoveArray(0)
#import pi
from numpy import pi
for i in range(M_DMD):
ugcd.AddArray(dsa.numpyTovtkDataArray(np.abs(DMD_modes[i]),prefix+'_DMD_mode_abs_{}_{}'.format(fieldname,i)))
ugcd.AddArray(dsa.numpyTovtkDataArray(np.angle(DMD_modes[i])*180/pi,prefix+'_DMD_mode_angle_{}_{}'.format(fieldname,i)))
ugw = vtk.vtkXMLUnstructuredGridWriter()
ugw.SetInputDataObject(geom)
ugw.SetFileName(DMD_sm_filename)
ugw.Write()
from future.builtins import range
import numpy as np
import modred as mr
num_vecs = 30
# Arbitrary data
vecs = np.random.random((100, num_vecs))
num_modes = 5
modes, eig_vals = mr.compute_POD_matrices_snaps_method(vecs,
list(range(num_modes)))
fb2_rig = RIGID_DISPLACEMENTS (fb2)
ib1_rig = RIGID_DISPLACEMENTS (ib1)
ib2_rig = RIGID_DISPLACEMENTS (ib2)
pod_input = [(fb1_rig, fb1_defo, 'FB1', fbmod),
(fb2_rig, fb2_defo, 'FB2', fbmod),
(ib1_rig, ib1_defo, 'IB1', ibmod),
(ib2_rig, ib2_defo, 'IB2', ibmod)]
for (rig, defo, label, num_modes) in pod_input:
vecs = numpy.transpose(numpy.array(rig+defo))
svec = vecs.shape[0]
nvec = vecs.shape[1]
print '%s:' % label, 'calculating', num_modes, 'POD modes from', nvec, 'input vectors of size', svec, '...'
try:
modes, vals = modred.compute_POD_matrices_snaps_method(vecs, list(range(num_modes)))
except:
print 'POD generation has failed --> it is possible that you tried to extract too many modes'
if label[0:2] == 'FB':
print ' try using [-fbmod a_smaller_number] and re-run'
else:
print ' try using [-ibmod a_smaller_number] and re-run'
sys.exit ()
mod = numpy.transpose(modes).tolist()
val = vals.tolist()
basevec = [x for vec in mod for x in vec]
podbase = (val[0:len(mod)], basevec)
path = afile.replace ('.inp','_' + label + '_base.pickle.gz')
print 'Saving: %s' % path
pickle.dump(podbase, gzip.open(path,'wb'))
