def __getitem__(self, idx):
if torch.is_tensor(idx):
idx = idx.tolist()
sample = self.features[idx]
return sample, sample
if __name__ == "__main__":
dataset = RepresentationDataset(nuclear_charge=1)
vae_structure = VAE(dataset.features.shape[-1], layer_size=128, n_layers=2, variant=3,
dimensions=2, activation=F.leaky_relu)
model = Model(dataset=dataset, model=vae_structure, epochs=50, learning_rate=3e-3, batch_size=100,
log_interval=100)
print(3e-2)
model.fit()
model.set_learning_rate(1e-2)
print(1e-2)
model.epochs = 30
model.fit()
model.set_learning_rate(3e-3)
print(3e-3)
model.epochs = 90
model.fit()
model.set_learning_rate(1e-3)
print(1e-3)
model.epochs = 30
model.fit()
from data import loading_data
from main import Model
def image(filename):
image = cv2.imread(filename, cv2.IMREAD_COLOR)
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
img = cv2.resize(gray, (28, 28), interpolation=cv2.INTER_AREA)
img = img.reshape(1, 28, 28, 1)
img = img / 255.0
return img
if __name__ == "__main__":
model = Model(input_shape=(28, 28, 1), classes=10)
train_ds, test_ds = loading_data()
model.compile(
learning_rate=0.01,
optimizer='sgd',
loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True))
model.fit(epochs=15, train_ds=train_ds, test_ds=test_ds)
#filename='images/9.jpg'
#img = image(filename)
#pred = nn.predict(img)
#final_pred = np.argmax(pred)
#print(pred)
#print(final_pred)
model.save('digit_model.h5')
