def get_elbow(components,partial_mask):
masked_eigenimgs = reshape_unmasked_values_to_shots( components.astype(np.float64),
partial_mask)
qs = np.linspace(0,1,1)
dc = DiffCorr(masked_eigenimgs,q_values=qs,k=0)
eigenimg_ac = dc.autocorr()[:,:,1]
if np.abs(eigenimg_ac[0])<0.5:
cutoff=1
else:
for ii,aa in enumerate(eigenimg_ac[:,0]):
if np.abs(aa)>=0.5:
continue
else:
cutoff=ii
break
return cutoff
def get_elbow(components, partial_mask):
masked_eigenimgs = reshape_unmasked_values_to_shots(
components.astype(np.float64), partial_mask)
qs = np.linspace(0, 1, 1)
dc = DiffCorr(masked_eigenimgs, q_values=qs, k=0)
eigenimg_ac = dc.autocorr()[:, :, 1]
if np.abs(eigenimg_ac[0]) < 0.5:
cutoff = 1
else:
for ii, aa in enumerate(eigenimg_ac[:, 0]):
if np.abs(aa) >= 0.5:
continue
else:
cutoff = ii
break
return cutoff
def pair_diff_PI(norm_shots, mask_corr, phi_offset=0, pair_method='int'):
if pair_method == 'corr':
print("doing corr pairing...")
#dummy qs
num_phi = norm_shots.shape[-1]
qs = np.array([1.0])
dc = DiffCorr(norm_shots, qs, 0, pre_dif=True)
corr = dc.autocorr()
corr /= mask_corr
corr = corr[:, :, phi_offset:num_phi / 2 - phi_offset]
eps = distance.cdist(corr[:, 0], corr[:, 0], metric='euclidean')
if pair_method == 'int':
print "doing intensity pair..."
eps = distance.cdist(norm_shots[:, 0],
norm_shots[:, 0],
metric='euclidean')
# do this so the diagonals are not the minimum, i.e. don't pair shot with itself
epsI = 1.1 * eps.max(1) * np.identity(eps.shape[0])
eps += epsI
shot_preference = np.roll(eps.argsort(1), 1, axis=1)
pref_dict = {str(E[0]): list(E[1:]) for E in shot_preference.astype(str)}
print("stable roommate pair....")
pairs_dict = stable.stableroomate(prefs=pref_dict)
pairing = np.array(MakeTagPairs._remove_duplicate_pairs(pairs_dict))
print("computing difference intensities...")
diff_norm = np.zeros(
(norm_shots.shape[0] / 2, norm_shots.shape[1], norm_shots.shape[-1]),
dtype=np.float64)
for index, pp in enumerate(pairing):
diff_norm[index] = norm_shots[pp[0]] - norm_shots[pp[1]]
return diff_norm, pairing
# get back the masked images and components
masked_mean_train =reshape_unmasked_values_to_shots(Train,partial_mask).mean(0)
# denoise
Train_noise = new_Train[:,:num_pca].dot(components[:num_pca])
denoise_Train= reshape_unmasked_values_to_shots(Train-Train_noise-Train.mean(0)[None,:]
, partial_mask)
if denoise_Train.shape[0]%2>0:
denoise_Train=denoise_Train[:-1]
denoise_Train=denoise_Train[:-1][::2]-denoise_Train[1:][::2]
dc=DiffCorr(denoise_Train,qvalues,0,pre_dif=True)
Train_difcor= dc.autocorr().mean(0)[0]
all_corrs.append(Train_difcor)
all_nums.append(Train.shape[0])
f_out.create_dataset('q%d/dif_cor%d'%(qidx, n_chunk)
,data=Train_difcor)
all_corrs = np.array(all_corrs)
all_nums = np.array(all_nums)
# print all_corrs.shape
# print all_nums
ave_corr = np.sum(all_corrs*(all_nums/float(all_nums.sum()))[:,None],axis=0 )
# print ave_corr.shape
for idx,ss in enumerate(shots):
norm_shots[idx]=normalize_shot(ss,this_mask)
# do we want to normalize by the entire range of intensity?
# divide into Train and test
num_shots = norm_shots.shape[0]
cutoff = int(num_pro_shots*0.1) # use 10% of the protein shots as testing set
partial_mask = this_mask.copy()
Train = norm_shots[cutoff:, partial_mask==1]
Test = norm_shots[:cutoff, partial_mask==1]
print ("%d test shots"%(Test.shape[0]))
print ("%d train shots"%(Train.shape[0]))
qvalues = np.linspace(0,1,partial_mask.shape[0])
mask_dc = DiffCorr(partial_mask,qvalues,0, pre_dif=True)
mask_cor = mask_dc.autocorr()
if args.num_pca is None:
num_pca = int(num_pca_components[qidx])
max_pca = num_pca+5
else:
num_pca = args.num_pca+1
max_pca = args.num_pca+1
print('denoisng with PCA critical num_pca_components = %d...'%num_pca)
if 'pca_components' not in f_out[q_group].keys():
# if there is no pca component saved, then run it and save the components
pca=PCA(n_components=50, whiten = False)
new_Train=pca.fit_transform(Train)
new_Test = pca.transform(Test)
if 'explained_variance_ratio' not in f_out[q_group].keys():
continue
if 'run%d' % run in f_out.keys():
print("already seen this run, skip!")
continue
##### load the mask used for this run
f_mask = h5py.File(os.path.join(args.mask_dir, 'run%d.tbl' % run), 'r')
mask = f_mask['polar_mask_binned'].value
mask = (mask == mask.max())
mask.shape
# do the mask cor
qs = np.linspace(0, 1, mask.shape[0])
dc = DiffCorr(mask[None, :, :], qs, 0, pre_dif=True)
mask_cor = dc.autocorr().mean(0)
f_mask.close()
f_out.create_group('run%d' % run)
all_ave_cors = []
all_nums = []
for qidx in range(35):
print('run%d q%d' % (run, qidx))
if 'num_pca_cutoff2' in f['q%d' % qidx].keys():
pca_num = f['q%d' % qidx]['num_pca_cutoff2'].value
else:
pca_num = f['q%d' % qidx]['num_pca_cutoff'].value
for idx, ss in enumerate(shots):
norm_shots[idx] = normalize_shot(ss, this_mask)
# do we want to normalize by the entire range of intensity?
# divide into Train and test
num_shots = norm_shots.shape[0]
cutoff = int(num_shots * 0.1) # use 10% of the shots as testing set
partial_mask = this_mask.copy()
Train = norm_shots[cutoff:, partial_mask == 1]
Test = norm_shots[:cutoff, partial_mask == 1]
print("%d test shots" % (Test.shape[0]))
print("%d train shots" % (Train.shape[0]))
qvalues = np.linspace(0, 1, partial_mask.shape[0])
mask_dc = DiffCorr(partial_mask, qvalues, 0, pre_dif=True)
mask_cor = mask_dc.autocorr()
max_pca_components = []
max_pca_components.append(int(num_pca_components[qidx]))
if 'num_pca_cutoff' in f_out[q_group].keys():
max_pca_components.append(f_out[q_group]['num_pca_cutoff'].value)
max_pca_components = list(set(max_pca_components))
if 'pca_components' not in f_out[q_group].keys():
# if there is no pca component saved, then run it and save the components
pca = PCA(n_components=50, whiten=False)
new_Train = pca.fit_transform(Train)
new_Test = pca.transform(Test)
if 'explained_variance_ratio' not in f_out[q_group].keys():
f_out.create_group('q%d' % qidx)
shots = PI[:, qidx, :][:, None, :]
this_mask = mask[qidx][None, :]
norm_shots = np.zeros_like(shots)
for idx, ss in enumerate(shots):
norm_shots[idx] = normalize_shot(ss, this_mask)
print("computing single shot correlations")
phi_offset = 10
num_phi = norm_shots.shape[-1]
qs = np.array([1.0])
dc = DiffCorr(this_mask[None, :, :], qs, 0, pre_dif=True)
mask_corr = dc.autocorr()
dc = DiffCorr(norm_shots, qs, 0, pre_dif=True)
corr = dc.autocorr()
corr /= mask_corr
corr = corr[:, :, phi_offset:num_phi / 2 - phi_offset]
pca = PCA(n_components=args.num_pca)
new_corr = pca.fit_transform(corr[:, 0, :])
kmeans = KMeans(n_clusters=args.num_clusters)
kmeans.fit(new_corr)
f_out.create_dataset('q%d/cluster_labels' % qidx, data=kmeans.labels_)
f_out.create_dataset('q%d/explained_variance_ratio' % qidx,
cluster_set_keys = f_cluster.keys()
out_file = run_file.replace('.tbl', '_cor.h5')
f_out = h5py.File(os.path.join(save_dir, out_file), 'w')
if 'polar_mask_binned' in f.keys():
mask = np.array(
f['polar_mask_binned'].value == f['polar_mask_binned'].value.max(),
dtype=int)
else:
mask = np.load(
'/reg/d/psdm/cxi/cxilp6715/scratch/water_data/binned_pmask_basic.npy')
qs = np.linspace(0.2, 0.88, mask.shape[0])
dc = DiffCorr(mask[None, :, :], qs, 0, pre_dif=True)
mask_ac = dc.autocorr()
PI = f['polar_imgs']
shot_tags = np.arange(0, PI.shape[0])
for set_key in cluster_set_keys:
print("computing diff cor for %s..." % set_key)
qidx = int(set_key.split('q')[1])
labels = f_cluster[set_key]['cluster_labels'].value.astype(int)
f_out.create_group(set_key)
unique_labels = np.unique(labels)
cluster_corrs = []
cluster_num_shots = []
for ll in unique_labels:
sys.exit()
out_file2 = run_file.replace('.tbl', '_chunks_intershot_uncertainty.h5')
f_out2 = h5py.File(os.path.join(save_dir, out_file2), 'w')
if 'polar_mask_binned' in f.keys():
mask = np.array(
f['polar_mask_binned'].value == f['polar_mask_binned'].value.max(),
dtype=int)
else:
print("there is no mask stored with the shots")
sys.exit()
# mask = np.load('/reg/d/psdm/cxi/cxilp6715/results/shared_files/binned_pmask_basic.npy')
# do the mask cor
qs = np.linspace(0, 1, mask.shape[0])
dc = DiffCorr(mask[None, :, :], qs, 0, pre_dif=True)
mask_cor = dc.autocorr().mean(0)
PI = f['polar_imgs']
# filter by photon energy. If the photon energy of the shot if not within 100 EV of the average, do not use
photon_energy = np.nan_to_num(f['ebeam']['photon_energy'].value)
mean_E = photon_energy.mean()
E_sigma = 100.
shot_tage_to_keep=np.where( (photon_energy> (mean_E-E_sigma))\
+(photon_energy< (mean_E-E_sigma)) )[0]
print('Num of shots to be used: %d' % (shot_tage_to_keep.size))
# figure which qs are used for pairing
qmin = args.qmin
qmax = args.qmax
if qmax is None:
masked_mean_train =reshape_unmasked_values_to_shots(Train,partial_mask).mean(0)
#### this is just for saving to get error bars
if save_cors:
grp=f_out['q%d'%qidx]
nn=grp['num_pca_cutoff'].value
if 'all_difcors' in grp['pca%d'%nn].keys():
print('already save dif cors for this cutoff (%d) at q%d'%(nn,qidx))
else:
Train_noise = new_Train[:,:nn].dot(components[:nn])
denoise_Train= reshape_unmasked_values_to_shots(norm_shots-Train_noise-Train.mean(0)[None,:]
, partial_mask)
dc=DiffCorr(denoise_Train,qvalues,0,pre_dif=False)
Train_difcor= dc.autocorr()
f_out.create_dataset('%s/pca%d/all_train_difcors'%(cluster_group,nn)
,data=Train_difcor)
del norm_shots
continue
# denoise
for nn in range(1, max_pca):
pca_group = '%s/pca%d'%(cluster_group,nn)
if 'pca%d'%nn in f_out[cluster_group].keys():
print("pca denoise at pca n_components = %d is already done. Skip!"%nn)
continue
if nn>0:
for idx,ss in enumerate(shots):
norm_shots[idx]=normalize_shot(ss,this_mask)
# do we want to normalize by the entire range of intensity?
# divide into Train and test
num_shots = norm_shots.shape[0]
cutoff = int(num_shots*0.1) # use 10% of the shots as testing set
partial_mask = this_mask.copy()
Train = norm_shots[cutoff:, partial_mask==1]
Test = norm_shots[:cutoff, partial_mask==1]
print ("%d test shots"%(Test.shape[0]))
print ("%d train shots"%(Train.shape[0]))
qvalues = np.linspace(0,1,partial_mask.shape[0])
mask_dc = DiffCorr(partial_mask,qvalues,0, pre_dif=True)
mask_cor = mask_dc.autocorr()
num_pca = int(num_pca_components[qidx])
if num_pca >0:
# do PCA stuff
print('denoisng with PCA num_pca_components = %d...'%num_pca)
pca=PCA(n_components=num_pca, whiten = False)
new_Train=pca.fit_transform(Train)
new_Test = pca.transform(Test)
# get back the masked images and components
components=pca.components_
masked_mean_train =reshape_unmasked_values_to_shots(Train,partial_mask).mean(0)
masked_mean_test =reshape_unmasked_values_to_shots(Test,partial_mask).mean(0)
# denoise
grp = f_out['q%d' % qidx]
nn = grp['num_pca_cutoff'].value
if 'all_test_difcors' in grp['pca%d' % nn].keys():
print('already save dif cors for this cutoff (%d) at q%d' %
(nn, qidx))
else:
Test_noise = new_Test[:, :nn].dot(components[:nn])
denoise_Test = reshape_unmasked_values_to_shots(
Test - Test_noise - Test.mean(0)[None, :], partial_mask)
Train_noise = new_Train[:, :nn].dot(components[:nn])
denoise_Train = reshape_unmasked_values_to_shots(
Train - Train_noise - Train.mean(0)[None, :], partial_mask)
dc = DiffCorr(denoise_Train, qvalues, 0, pre_dif=False)
Train_difcor = dc.autocorr()
dc = DiffCorr(denoise_Test, qvalues, 0, pre_dif=False)
Test_difcor = dc.autocorr()
f_out.create_dataset('q%d/pca%d/all_test_difcors' % (qidx, nn),
data=Test_difcor)
f_out.create_dataset('q%d/pca%d/all_train_difcors' % (qidx, nn),
data=Train_difcor)
del shots
del norm_shots
continue
# denoise
masked_mean_train = reshape_unmasked_values_to_shots(
Train, partial_mask).mean(0)
# denoise
Train_noise = new_Train[:, :num_pca].dot(components[:num_pca])
denoise_Train = reshape_unmasked_values_to_shots(
Train - Train_noise - Train.mean(0)[None, :], partial_mask)
if denoise_Train.shape[0] % 2 > 0:
denoise_Train = denoise_Train[:-1]
denoise_Train_diff = denoise_Train[:-1][::2] - denoise_Train[1:][::2]
dc = DiffCorr(denoise_Train_diff, qvalues, 0, pre_dif=True)
Train_difcor = dc.autocorr().mean(0)[0]
all_corrs.append(Train_difcor)
all_nums.append(Train.shape[0])
f_out.create_dataset('q%d/dif_cor%d' % (qidx, n_chunk),
data=Train_difcor)
#########do clustering with corr PCA clustering#########
print("computing single shot correlations")
phi_offset = 10
num_phi = denoise_Train.shape[-1]
mask_dc = DiffCorr(partial_mask, qvalues, 0, pre_dif=True)
mask_cor = mask_dc.autocorr()
# get back the masked images and components
masked_mean_train = reshape_unmasked_values_to_shots(
Train, partial_mask).mean(0)
# denoise
Train_noise = new_Train[:, :num_pca].dot(components[:num_pca])
denoise_Train = reshape_unmasked_values_to_shots(
Train - Train_noise - Train.mean(0)[None, :], partial_mask)
if denoise_Train.shape[0] % 2 > 0:
denoise_Train = denoise_Train[:-1]
denoise_Train = denoise_Train[:-1][::2] - denoise_Train[1:][::2]
dc = DiffCorr(denoise_Train, qvalues, 0, pre_dif=True)
Train_difcor = dc.autocorr().mean(0)[0]
all_corrs.append(Train_difcor)
all_nums.append(Train.shape[0])
f_out.create_dataset('q%d/dif_cor%d' % (qidx, n_chunk),
data=Train_difcor)
all_corrs = np.array(all_corrs)
all_nums = np.array(all_nums)
# print all_corrs.shape
# print all_nums
ave_corr = np.sum(all_corrs * (all_nums / float(all_nums.sum()))[:, None],
axis=0)
# print ave_corr.shape
for idx, ss in enumerate(shots):
norm_shots[idx] = normalize_shot(ss, this_mask)
# do we want to normalize by the entire range of intensity?
# divide into Train and test
num_shots = norm_shots.shape[0]
cutoff = int(num_shots * 0.1) # use 10% of the shots as testing set
partial_mask = this_mask.copy()
Train = norm_shots[cutoff:, partial_mask == 1]
Test = norm_shots[:cutoff, partial_mask == 1]
print("%d test shots" % (Test.shape[0]))
print("%d train shots" % (Train.shape[0]))
qvalues = np.linspace(0, 1, partial_mask.shape[0])
mask_dc = DiffCorr(partial_mask, qvalues, 0, pre_dif=True)
mask_cor = mask_dc.autocorr()
if args.num_pca > 0:
# do PCA stuff
print('denoisng with PCA...')
pca = PCA(n_components=args.num_pca, whiten=False)
new_Train = pca.fit_transform(Train)
new_Test = pca.transform(Test)
# get back the masked images and components
components = pca.components_
masked_mean_train = reshape_unmasked_values_to_shots(
Train, partial_mask).mean(0)
masked_mean_test = reshape_unmasked_values_to_shots(
Test, partial_mask).mean(0)
# denoise
cluster_file = run_file.replace('.tbl','_PCA-cluster.h5')
f_cluster = h5py.File(os.path.join(cluster_dir, cluster_file),'r')
cluster_set_keys = f_cluster.keys()
out_file = run_file.replace('.tbl','_cor.h5')
f_out = h5py.File(os.path.join(save_dir, out_file),'w')
if 'polar_mask_binned' in f.keys():
mask = np.array(f['polar_mask_binned'].value==f['polar_mask_binned'].value.max(), dtype = int)
else:
mask = np.load('/reg/d/psdm/cxi/cxilp6715/scratch/water_data/binned_pmask_basic.npy')
qs=np.linspace(0.2,0.88,mask.shape[0])
dc=DiffCorr(mask[None,:,:],qs,0,pre_dif=True)
mask_ac=dc.autocorr()
PI = f['polar_imgs']
shot_tags = np.arange(0,PI.shape[0])
for set_key in cluster_set_keys:
print("computing diff cor for %s..."%set_key)
qidx = int( set_key.split('q')[1] )
labels = f_cluster[set_key]['cluster_labels'].value.astype(int)
f_out.create_group(set_key)
unique_labels=np.unique(labels)
cluster_corrs=[]
cluster_num_shots=[]
if args.save:
grp=f_out['q%d'%qidx]
nn=grp['num_pca_cutoff'].value
if 'all_test_difcors' in grp['pca%d'%nn].keys():
print('already save dif cors for this cutoff (%d) at q%d'%(nn,qidx))
else:
Test_noise = new_Test[:,:nn].dot(components[:nn])
denoise_Test = reshape_unmasked_values_to_shots(Test-Test_noise-Test.mean(0)[None,:],
partial_mask)
Train_noise = new_Train[:,:nn].dot(components[:nn])
denoise_Train= reshape_unmasked_values_to_shots(Train-Train_noise-Train.mean(0)[None,:]
, partial_mask)
dc=DiffCorr(denoise_Train,qvalues,0,pre_dif=False)
Train_difcor= dc.autocorr()
dc=DiffCorr(denoise_Test,qvalues,0,pre_dif=False)
Test_difcor= dc.autocorr()
f_out.create_dataset('q%d/pca%d/all_test_difcors'%(qidx,nn)
,data=Test_difcor)
f_out.create_dataset('q%d/pca%d/all_train_difcors'%(qidx,nn)
,data=Train_difcor)
del shots
del norm_shots
continue
nn=grp['num_pca_cutoff2'].value
else:
nn=grp['num_pca_cutoff'].value
if 'all_test_difcors' in grp['pca%d'%nn].keys():
print('already save dif cors for this cutoff (%d) at q%d'%(nn,qidx))
else:
Test_noise = new_Test[:,:nn].dot(components[:nn])
denoise_Test = reshape_unmasked_values_to_shots(Test-Test_noise-Test.mean(0)[None,:],
partial_mask)
Train_noise = new_Train[:,:nn].dot(components[:nn])
denoise_Train= reshape_unmasked_values_to_shots(Train-Train_noise-Train.mean(0)[None,:]
, partial_mask)
dc=DiffCorr(denoise_Train,qvalues,0,pre_dif=False)
Train_difcor= dc.autocorr()
dc=DiffCorr(denoise_Test,qvalues,0,pre_dif=False)
Test_difcor= dc.autocorr()
f_out.create_dataset('q%d/pca%d/all_test_difcors'%(qidx,nn)
,data=Test_difcor)
f_out.create_dataset('q%d/pca%d/all_train_difcors'%(qidx,nn)
,data=Train_difcor)
del norm_shots
else:
# denoise
# output file to save data
out_file = run_file.replace('.tbl', '_pca0.h5')
f_out = h5py.File(os.path.join(save_dir, out_file), 'w')
if 'polar_mask_binned' in f.keys():
mask = np.array(
f['polar_mask_binned'].value == f['polar_mask_binned'].value.max(),
dtype=int)
else:
print("there is no mask stored with the shots")
sys.exit()
# mask = np.load('/reg/d/psdm/cxi/cxilp6715/results/shared_files/binned_pmask_basic.npy')
qs = np.linspace(0, 1, mask.shape[0])
dc = DiffCorr(mask[None, :, :], qs, 0, pre_dif=True)
mask_cor = dc.autocorr().mean(0)
PI = f['polar_imgs']
# filter by photon energy. If the photon energy of the shot if not within 100 EV of the average, do not use
photon_energy = np.nan_to_num(f['ebeam']['photon_energy'].value)
mean_E = photon_energy.mean()
E_sigma = 100.
shot_tage_to_keep=np.where( (photon_energy> (mean_E-E_sigma))\
+(photon_energy< (mean_E-E_sigma)) )[0]
print('Num of shots to be used: %d' % (shot_tage_to_keep.size))
# figure which qs are used for pairing
qmin = args.qmin
qmax = args.qmax
if qmax is None:
f_out.create_group('q%d'%qidx)
shots=PI[:,qidx,:][:,None,:]
this_mask = mask[qidx][None,:]
norm_shots = np.zeros_like(shots)
for idx,ss in enumerate(shots):
norm_shots[idx]=normalize_shot(ss,this_mask)
print("computing single shot correlations")
phi_offset=10
num_phi=norm_shots.shape[-1]
qs = np.array([1.0])
dc=DiffCorr(this_mask[None,:,:],qs,0,pre_dif=True)
mask_corr=dc.autocorr()
dc = DiffCorr(norm_shots,
qs,0,pre_dif=True)
corr = dc.autocorr()
corr/=mask_corr
corr=corr[:,:,phi_offset:num_phi/2-phi_offset]
pca=PCA(n_components=args.num_pca)
new_corr=pca.fit_transform(corr[:,0,:])
kmeans=KMeans(n_clusters=args.num_clusters)
kmeans.fit(new_corr)
f_out.create_dataset('q%d/cluster_labels'%qidx,data=kmeans.labels_)
for idx,ss in enumerate(shots):
norm_shots[idx]=normalize_shot(ss,this_mask)
# do we want to normalize by the entire range of intensity?
# divide into Train and test
num_shots = norm_shots.shape[0]
cutoff = int(num_shots*0.1) # use 10% of the shots as testing set
partial_mask = this_mask.copy()
Train = norm_shots[cutoff:, partial_mask==1]
Test = norm_shots[:cutoff, partial_mask==1]
print ("%d test shots"%(Test.shape[0]))
print ("%d train shots"%(Train.shape[0]))
qvalues = np.linspace(0,1,partial_mask.shape[0])
mask_dc = DiffCorr(partial_mask,qvalues,0, pre_dif=True)
mask_cor = mask_dc.autocorr()
if args.num_pca is None:
num_pca = int(num_pca_components[qidx])
max_pca = num_pca+5
else:
num_pca = args.num_pca+1
max_pca = args.num_pca+1
print('denoisng with PCA critical num_pca_components = %d...'%num_pca)
if 'pca_components' not in f_out[q_group].keys():
# if there is no pca component saved, then run it and save the components
pca=PCA(n_components=50, whiten = False)
new_Train=pca.fit_transform(Train)
new_Test = pca.transform(Test)
if 'explained_variance_ratio' not in f_out[q_group].keys():
for idx, ss in enumerate(shots):
norm_shots[idx] = normalize_shot(ss, this_mask)
# do we want to normalize by the entire range of intensity?
# divide into Train and test
num_shots = norm_shots.shape[0]
cutoff = int(num_shots * 0.1) # use 10% of the shots as testing set
partial_mask = this_mask.copy()
Train = norm_shots[cutoff:, partial_mask == 1]
Test = norm_shots[:cutoff, partial_mask == 1]
print("%d test shots" % (Test.shape[0]))
print("%d train shots" % (Train.shape[0]))
qvalues = np.linspace(0, 1, partial_mask.shape[0])
mask_dc = DiffCorr(partial_mask, qvalues, 0, pre_dif=True)
mask_cor = mask_dc.autocorr()
if args.num_pca is None:
num_pca = int(num_pca_components[qidx])
max_pca = num_pca + 5
else:
num_pca = args.num_pca + 1
max_pca = args.num_pca + 1
print('denoisng with PCA critical num_pca_components = %d...' % num_pca)
if 'pca_components' not in f_out[q_group].keys():
# if there is no pca component saved, then run it and save the components
pca = PCA(n_components=50, whiten=False)
new_Train = pca.fit_transform(Train)
new_Test = pca.transform(Test)
if 'explained_variance_ratio' not in f_out[q_group].keys():