def main():
logging.basicConfig(level=logging.INFO)
(x_train, _), (x_test, _) = tf.keras.datasets.mnist.load_data()
x_train = x_train / 255
x_test = x_test / 255
dataset = tf.data.Dataset.from_tensor_slices(x_train.reshape(-1, 28 * 28).astype(np.float32))
dataset = dataset.shuffle(1024, reshuffle_each_iteration=True)
rbm = BBRBM(n_visible=28 * 28, n_hidden=64)
rbm.fit(dataset, epoches=100, batch_size=10)
for i in np.random.choice(np.arange(x_test.shape[0]), 5, replace=False):
x = x_test[i]
x_tensor = tf.convert_to_tensor(x.reshape(1, 28 * 28), dtype=tf.float32)
x_reconstructed_tensor = rbm.reconstruct(x_tensor)
x_reconstructed = x_reconstructed_tensor.numpy().reshape(28, 28)
Image.fromarray((x * 255).astype(np.uint8)).save(f'{i}_original.png')
Image.fromarray((x_reconstructed * 255).astype(np.uint8)).save(f'{i}_reconstructed.png')
#show_digit(image.reshape(28, 28), "Original image")
#print("image label",mnist.test.labels[j])
a = image.reshape(28, 28)
c = image.reshape(28, 28)
#show_digit(image.reshape(28,28))
# img = a[0:16,0:28] #crop the image
img = a * mask_b
img = np.concatenate((t1, img.flatten()), axis=0)
img_org = img
#print("shape of of org img", img_org.shape)
#imga = random_image#imga = img
#show_digit(img_org[10:794].reshape(28,28),"croped input")
#reconstruct image for N-MC
for i in range(400):
image_rec1 = bbrbm.reconstruct(img.reshape(1, -1), 0)
#print("shape of of rec1",image_rec1.shape)
image_rec1 = image_rec1.reshape(794, )
if (i > 400 - num_avg - 1):
store_recon_vu[i - (400 - num_avg)] = image_rec1
#print("stored labels : ", store_labels)
#print("index i : ", i)
#print("new shape of of rec1", image_rec1.shape)
t1 = image_rec1[0:10]
rec_backup = image_rec1
image_rec1 = image_rec1[10:794].reshape(28, 28)
#print("size ofa", a.size)
img = img_org + np.concatenate(
(t1, (image_rec1 * mask_c).flatten()), axis=0)
#show_digit(image_rec1.reshape(30, 30), "returned image")
#crop dimentions x = 6 y = 6 a = image.reshape(28, 28) img = a[0:16, 0:28] #crop the image #img = cropND(a,(x,y)) #show cropped image #rint(img) plt.imshow(a) plt.show() plt.imshow(img) plt.show() #pad the image to make it 780 before feeding it to the BM imge = np.pad(img, [(6, ), (0, )], mode='constant') #print(imge) plt.imshow(imge) plt.show() #reconstruct image_rec = bbrbm.reconstruct(imge.reshape(1, -1)) #show_digit(image) #show_digit(image_rec) #plot reconstructed image plt.imshow(image_rec.reshape(28, 28)) plt.show() ############################ #save the weights filename = 'weights' name = 'bbrbm' bbrbm.save_weights(filename, name)
digits_total_counts = np.ones(10, dtype=np.int32) * 10
for idx in range(mnist_test_images.shape[0]):
image = mnist_test_images[idx, ]
label = mnist_test_labels[idx, ]
for digit in range(10):
digit_label = np.zeros(10)
digit_label[digit] = 1
if (label == digit_label).all() and digits_current_counts[digit] < 10:
nrow = digits_current_counts[digit]
sample_idx = nrow * 10 + digit
mnist_test_images_samples[sample_idx, ] = image
mnist_test_images_samples_rec[sample_idx, ] = \
bbrbm.reconstruct(image.reshape([1, -1]))
mnist_test_images_samples_plt[sample_idx * 2, ] = \
mnist_test_images_samples[sample_idx, ]
mnist_test_images_samples_plt[sample_idx * 2 + 1, ] = \
mnist_test_images_samples_rec[sample_idx, ]
digits_current_counts[digit] += 1
if (digits_current_counts == digits_total_counts).all():
break
# %%
# funtion of plotting mnist data
def plot_mnist(mnist_images, nrows, ncols, cmap='gray'):
fig = plt.figure(figsize=(ncols, nrows))
gs = gridspec.GridSpec(nrows, ncols)
#names in case saving reconstruction is desired
fname = ["1-3","2-3","3-3","4-3","5-3","6-3","7-3","8-3","9-3","10-3","11-3","12-3","13-3","14-3","15-3","16-3","17-3","18-3","19-3","20-3"]
#802936312
for j in range(1):
#reconstruct image for N-MC
mnist_labeled = mnist_labeled_backup*mask_1a
show_digit(mask_1a[0][0:784].reshape(28, 28), "mask_1a")
show_digit(mask_1a[0][784:794].reshape(1, -1), "mask_1a labels")
show_digit(mask_1b[0][0:784].reshape(28, 28), "mask_1b")
show_digit(mnist_labeled[0][0:784].reshape(28, 28), "croped input image")
show_digit(mask_1b[0][784:794].reshape(1, -1), "mask_1b labels")
show_digit(mnist_labeled[0][784:794].reshape(1, -1), "copped labels")
for i in range(100):
image_rec1 = bbrbm.reconstruct(mnist_labeled)
#print("shape of of rec1",image_rec1.shape)
#mnist_labeled = image_rec1 # image_rec1.reshape(794,)
mnist_labeled = mnist_labeled_backup*mask_1a + image_rec1*mask_1b
#show_digit(image_rec1[0][78:7094].reshape(1, -1), "returned labels")
#show_digit(image_rec1[0][0:784].reshape(28, 28), "recreated image")
#show_digit(mnist_labeled[0][0:784].reshape(28, 28), "to feed back image")
#show_digit(mnist_labeled[0][784:794].reshape(1, -1), "to feed back labels")
#plot the reconstruction results
for n in range(20):
fig = plt.figure(figsize=(10, 10))
# setting values to rows and column variables for the figure
def run(first_hidden, second_hidden, learning_rate, epochs, batch_size):
x_train, x_test, y_train, y_test = get_datasets()
# ------------------FIRST HIDDEN-------------------
weight_matrix = generate_weight_matrix(x_train.shape[1], hidden)
visible_biases = np.zeros(shape=(784))
hidden_biases = np.zeros(shape=(first_hidden))
bbrbm = BBRBM(
n_visible=784,
n_hidden=first_hidden,
learning_rate=learning_rate,
momentum=0,
# momentum=0.95,
use_tqdm=True)
bbrbm.set_weights(weight_matrix, visible_biases, hidden_biases)
errs = bbrbm.fit(x_train, n_epoches=epochs, batch_size=batch_size)
first_weight_matrix = np.asarray(bbrbm.get_weight_matrix())
first_visible_biases = bbrbm.get_visible_biases()
first_hidden_biases = bbrbm.get_hidden_biases()
first_output = np.dot(x_test, weight_matrix)
for i in range(first_output.shape[0]):
first_output[i, :] += first_hidden_biases
for j in range(first_output.shape[1]):
# sigmoid(relu(x))
first_output[i, j] = sigmoid(np.max(0, first_output[i, j]))
# ------------------SECOND HIDDEN-------------------
bbrbm_second = BBRBM(
n_visible=first_hidden,
n_hidden=784,
learning_rate=learning_rate,
momentum=0,
# momentum=0.95,
use_tqdm=True)
trained_outputs = np.ndarray(shape=(x_train.shape[0], x_train.shape[1]))
for i in range(trained_outputs.shape[0]):
trained_outputs[i, :] = bbrbm.reconstruct(x_train[i].reshape(1, -1))
sgd = optimizers.SGD(lr=0.3, decay=0, momentum=0, nesterov=True)
model = Sequential()
model.add(
Dense(10,
input_dim=784,
activation="softmax",
kernel_initializer=keras.initializers.RandomNormal(mean=0.0,
stddev=0.01),
bias_initializer='zeros'))
model.compile(optimizer=sgd,
loss='mean_squared_error',
metrics=['accuracy'],
loss_weights=None,
sample_weight_mode=None,
weighted_metrics=None,
target_tensors=None)
earlyStopping = keras.callbacks.EarlyStopping(monitor='val_loss',
min_delta=0,
patience=0,
verbose=0,
mode='auto')
model.fit(trained_outputs,
np.asarray(y_train),
batch_size=batch_size,
epochs=epochs,
verbose=1,
shuffle=True,
callbacks=[earlyStopping],
validation_data=(x_test, y_test))
score = model.evaluate(x_test, np.asarray(y_test), verbose=1)
print('Test loss:', score[0])
print('Test accuracy:', score[1])
# vis(x_train, trained_outputs, 20)
trained_outputs = np.dot(x_train, weight_matrix)
for i in range(trained_outputs.shape[0]):
trained_outputs[i, :] += hidden_biases
debug = 0
# test = np.dot(x_test, weight_matrix)
#
# preds = []
# for i in range(test.shape[0]):
#
# test[i,:]+=hidden_biases
# output_weight_matrix = generate_output_weight_matrix(test.shape[1])
# sgd = optimizers.SGD(lr=0.01, decay=0, momentum=0, nesterov=True)
# model = Model(inputs=test, outputs=y_test)
model.compile(optimizer=sgd,
loss='mean_squared_error',
metrics=['accuracy'],
loss_weights=None,
sample_weight_mode=None,
weighted_metrics=None,
target_tensors=None)
debug = 0