l = data_np.shape[0]
tr = int(0.8*l)
valid_np = data_np[tr:]
data_np = data_np[:tr]
print("-----------------------------")
print(data_np.shape)
print(valid_np.shape)
train_ls = []
for i in range(len(data_np)):
img = data_np[i]
l = img[-1]
img = img[:-1]
reconstructed_img = torch.FloatTensor(np.array(img))
_,reconstructed_img = rbm.reconstruct(reconstructed_img)
reconstructed_img = reconstructed_img.squeeze()
reconstructed_img = list(reconstructed_img)
reconstructed_img.append(l)
train_ls.append(reconstructed_img)
valid_ls = []
for i in range(len(valid_np)):
img = valid_np[i]
l = img[-1]
img = img[:-1]
reconstructed_img = torch.FloatTensor(np.array(img))
_,reconstructed_img = rbm.reconstruct(reconstructed_img)
reconstructed_img = reconstructed_img.squeeze()
reconstructed_img = list(reconstructed_img)
reconstructed_img.append(l)
[torch.Tensor(number) for i in range(len(particular_mnist))])
dbn_mnist = DBN(visible_units=28 * 28,
hidden_units=[23 * 23, 18 * 18],
k=5,
learning_rate=0.01,
learning_rate_decay=True,
xavier_init=True,
increase_to_cd_k=False,
use_gpu=False)
Epoch = [10, 15, 20, 25, 30, 35]
for epoch in Epoch:
print('-----------------------{}------------'.format(epoch))
dbn_mnist.train_static(train_data, train_label, epoch, batch_size)
idx = 3
img = train_dataset.train_data[idx]
reconstructed_img = img.view(1, -1).type(torch.FloatTensor)
_, reconstructed_img = dbn_mnist.reconstruct(reconstructed_img)
reconstructed_img = reconstructed_img.view((28, 28))
print("The original number: {}".format(train_dataset.train_labels[idx]))
plt.imshow(img, cmap='gray')
plt.show()
print("The reconstructed image")
plt.title(epoch)
plt.imshow(reconstructed_img, cmap='gray')
plt.show()
