# Training:
# ---------
optimizer = optim.Adam(model.parameters(),lr=meta_lr, betas=(0.9,0.999), amsgrad=False)
loss_fn = nn.CrossEntropyLoss()
for iteration in range(nepoch):
optimizer.zero_grad()
meta_train_error = 0.0
meta_valid_error = 0.0
meta_train_acc = 0.0
meta_valid_acc = 0.0
for task in range(batch_size):
dataset.change_cluster()
# Compute meta-training loss
learner = maml.clone()
batch_idc = dataset.sample()
batch = (h_concat_scaled[batch_idc,:],label[batch_idc])
evaluation_error, evaluation_acc = fast_adapt(batch,
learner,
loss_fn,
update_step,
shots)
evaluation_error.backward()
meta_train_error += evaluation_error.item()
meta_train_acc += evaluation_acc.item()
# Compute meta-validation loss
learner = maml.clone()
batch_idc = dataset.sample()
batch = (h_concat_scaled[batch_idc,:],label[batch_idc])
evaluation_error, evaluation_acc = fast_adapt(batch,
model = AnalogBeamformer(n_antenna = num_antenna, n_beam = N)
maml = MAML(model, lr=fast_lr, first_order=True)
# Training:
# ---------
optimizer = optim.Adam(model.parameters(),lr=meta_lr, betas=(0.9,0.999), amsgrad=False)
loss_fn = bf_gain_loss
for iteration in range(nepoch):
optimizer.zero_grad()
meta_train_error = 0.0
meta_valid_error = 0.0
for task in range(batch_size):
dataset.change_cluster()
# Compute meta-training loss
learner = maml.clone()
batch_idc = dataset.sample()
batch = (h_concat_scaled[batch_idc,:],egc_gain_scaled[batch_idc])
evaluation_error = fast_adapt_est_h(batch,
learner,
loss_fn,
update_step,
shots,
h_est_force_z)
evaluation_error.backward()
meta_train_error += evaluation_error.item()
# Compute meta-validation loss
learner = maml.clone()
batch_idc = dataset.sample()
batch = (h_concat_scaled[batch_idc,:],egc_gain_scaled[batch_idc])
evaluation_error = fast_adapt_est_h(batch,
nepoch = 50
#-------------------------------------------#
# Here should be the data_preparing function
# It is expected to return:
# train_inp, train_out, val_inp, and val_out
#-------------------------------------------#
h_real = np.load('D://Github Repositories/mmWave Beam Management/H_Matrices FineGrid/MISO_Static_FineGrid_Hmatrices_real.npy')[:,antenna_sel]
h_imag = np.load('D://Github Repositories/mmWave Beam Management/H_Matrices FineGrid/MISO_Static_FineGrid_Hmatrices_imag.npy')[:,antenna_sel]
loc = np.load('D://Github Repositories/mmWave Beam Management/H_Matrices FineGrid/MISO_Static_FineGrid_UE_location.npy')
# h_real = np.load('/Users/yh9277/Dropbox/ML Beam Alignment/Data/H_Matrices FineGrid/MISO_Static_FineGrid_Hmatrices_real.npy')
# h_imag = np.load('/Users/yh9277/Dropbox/ML Beam Alignment/Data/H_Matrices FineGrid/MISO_Static_FineGrid_Hmatrices_imag.npy')
BS_loc = [641,435,10]
num_samples = h_real.shape[0]
gc = GaussianCenters(n_clusters=8, arrival_rate = 1000, cluster_variance = 5, seed=r_seed)
sel_samples = gc.sample()
# gc.plot_sample(sel_samples)
# sel_samples = np.arange(10000)
h_real = h_real[sel_samples,:]
h_imag = h_imag[sel_samples,:]
loc = loc[sel_samples,:]
plt.figure(figsize=(8,6))
plt.scatter(loc[:,0], loc[:,1], s=1, label='UE')
plt.scatter(BS_loc[0], BS_loc[1], s=10, marker='s', label='BS')
plt.legend(loc='lower left')
plt.xlabel('x (meter)')
plt.ylabel('y (meter)')
plt.title('UE Distribution')
plt.show()