def Modify(logits_student, target):
if torch.is_tensor(logits_student):
logits_student = logits_student.detach().cpu().numpy()
if torch.is_tensor(target):
target = target.cpu().numpy()
num = logits_student.shape[0]
assert (num == target.shape[0])
pred = []
truth = []
for i in range(num):
ptemp = istft(logits_student[i][0], logits_student[i][1])
ttemp = istft(target[i][0], target[i][1])
pred.append(ptemp)
truth.append(ttemp)
# pdb.set_trace()
return np.array(pred, dtype='float32'), np.array(truth, dtype='float32')
def Modify(logits_student, xgg, target, ygg, refill, images):
if torch.is_tensor(images):
images = images.detach().cpu().numpy()
if torch.is_tensor(logits_student):
logits_student = logits_student.detach().cpu().numpy()
if torch.is_tensor(target):
target = target.detach().cpu().numpy()
if torch.is_tensor(xgg):
xgg = xgg.detach().cpu().numpy()
if torch.is_tensor(ygg):
ygg = ygg.detach().cpu().numpy()
num = logits_student.shape[0]
assert (num == target.shape[0])
pred = []
truth = []
bsflag, noise_ori, batch_snr = refill
bsflag = bsflag.cpu().numpy()
noise_ori = noise_ori.cpu().numpy()
batch_snr = batch_snr.cpu().numpy()
j = 0
for i in range(num):
ptemp = istft(logits_student[i], xgg[i])
ttemp = istft(target[i], ygg[i])
pred.append(ptemp)
truth.append(ttemp)
if batch_snr[i] == ROOT_SNRS[j]:
j = j + 1
if j == len(ROOT_SNRS):
j = 0
if np.sum(bsflag[i]) == 0:
tag = 'NBS_' + str(batch_snr[i])
else:
tag = 'BS_' + str(batch_snr[i])
plt.subplot(311)
# pdb.set_trace()
ct = plt.pcolormesh(xgg[i])
plt.colorbar(ct, label="power (dB)")
plt.subplot(312)
ct = plt.pcolormesh(ygg[i])
plt.colorbar(ct, label="power (dB)")
plt.subplot(313)
ct = plt.pcolormesh(images[i])
plt.colorbar(ct, label="power (dB)")
plt.suptitle(tag, fontsize=14)
plt.show()
# pdb.set_trace()
return np.array(pred, dtype='float32'), np.array(truth, dtype='float32')
def Modify(logits_student0, logits_student1, target0, target1):
if torch.is_tensor(logits_student0):
logits_student0 = logits_student0.detach().cpu().numpy()
if torch.is_tensor(logits_student1):
logits_student1 = logits_student1.detach().cpu().numpy()
if torch.is_tensor(target0):
target0 = target0.cpu().numpy()
if torch.is_tensor(target1):
target1 = target1.cpu().numpy()
num = logits_student0.shape[0]
assert (num == target0.shape[0])
assert (num == target1.shape[0])
pred = []
truth = []
for i in range(num):
# pdb.set_trace()
ptemp = istft(logits_student0[i], logits_student1[i])
ttemp = istft(target0[i], target1[i])
pred.append(ptemp)
truth.append(ttemp)
# pdb.set_trace()
return np.array(pred, dtype='float32'), np.array(truth, dtype='float32')
def unfold(whole_package):
new_package = []
for geti in whole_package:
new_geti = []
for tile in geti:
if torch.is_tensor(tile):
tile = tile.cpu().numpy()
new_geti.append(tile)
# pdb.set_trace()
assert (len(new_geti) == 7)
images, xgg, pred, truth, mm, nn, kk = new_geti
assert (pred.shape[0] == truth.shape[0])
assert (truth.shape[0] == mm.shape[0])
assert (mm.shape[0] == nn.shape[0])
assert (nn.shape[0] == kk.shape[0])
temp = []
for i in range(pred.shape[0]):
ori_wav = istft(images[i], xgg[i])
temp.append((ori_wav, pred[i], truth[i], mm[i], nn[i], kk[i]))
new_package = new_package + temp
return new_package
if __name__ == '__main__':
a = data_generator()
B = BSdataset_BR_recovery(a.trains)
okay = B.data
i = 0
for datum in okay:
if datum[2][2][1] == ROOT_SNRS[i]:
# pdb.set_trace()
# ampli = datum[0][0][0]
# phas = datum[0][0][1]
ampli = datum[1][0]
phas = datum[1][1]
original = datum[1][2]
synt = istft(ampli, phas)
if np.sum(datum[2][0]) == 0:
label = 'NBS'
else:
label = 'BS'
tstr = label + '_' + datum[2][2][0] + '_' + str(datum[2][2][1])
# add = ori
plt.subplot(411)
plt.plot(synt)
plt.subplot(412)
plt.plot(original)
plt.subplot(413)
coutour = plt.pcolormesh(ampli)
plt.colorbar(coutour, label="power (dB)")
# stft = GetFrequencyFeatures(he, 4000)
