def __init__(self, dict_size: int, embedding_size: int, hidden_units: int):
self.layers = [layer.Embedding(dict_size, embedding_size, 'l0'),
layer.Linear(embedding_size, hidden_units, 'l1-0'),
layer.Linear(embedding_size, hidden_units, 'l1-1'),
layer.Linear(embedding_size, hidden_units, 'l1-2'),
layer.Linear(hidden_units, dict_size, 'l2'),
layer.Softmax(dict_size, 'l3')]
self.h: list = None
return
def __init__(self, dict_size: int, embedding_size: int, hidden_units: int):
self.layers = [layer.Embedding(dict_size, embedding_size, 'l0'), # layer 0
layer.Linear(embedding_size, hidden_units, 'l1-0'), # layer 1
layer.Linear(embedding_size, hidden_units, 'l1-1'), # layer 2
layer.Linear(embedding_size, hidden_units, 'l1-2'), # layer 3
layer.Tanh(hidden_units, 'l2'), # layer 4
layer.Linear(hidden_units, dict_size, 'l3'), # layer 5
layer.Softmax(dict_size, 'l4')] # layer 6
self.h: list = None
return
def __init__(self, dict_size: int, embedding_size: int, hidden_units: int):
self.layers = [layer.Embedding(dict_size, embedding_size, 'l0-Embedding'),
layer.Recursive(embedding_size, hidden_units, 'l1-R'),
layer.Tanh(hidden_units, 'l2-Tanh'),
layer.Linear(hidden_units, dict_size, 'l3-L'),
layer.Softmax(dict_size, 'l4-Softmax')]
self.h: list = None
return
utf.cross_train(mynn, data_store_train, data_store_valid, train_settings)
"""
#############################
print('Settings 3')
train_settings = TrainSettings(learning_rate=0.005,
batch_size=16,
momentum=0.0,
plot_callback=cb.plot_callback,
loss_callback=cb.loss_callback,
filename='script-1-1-3',
epoch=200,
prefix='e16')
layers = [
layer.Linear(784, 100),
#layer.BN(100, 100),
layer.Sigmoid(100),
layer.Linear(100, 10),
layer.Softmax(10)
]
mynn = network.MLP(layers)
utf.cross_train(mynn, data_store_train, data_store_valid, train_settings)
print('Settings 4')
train_settings = TrainSettings(learning_rate=0.01,
batch_size=16,
momentum=0.0,
plot_callback=cb.plot_callback,
print('start initializing...')
network.init_nn(random_seed=1099)
learning_rates = [0.001]
momentums = [0.9]
regularizers = [0.0001]
x_train, y_train = load_data.load_from_path(data_train_filepath)
x_valid, y_valid = load_data.load_from_path(data_valid_filepath)
withBN = True
if not withBN:
for i2 in range(len(regularizers)):
for i3 in range(len(momentums)):
for i4 in range(len(learning_rates)):
layers = [layer.Linear(784, 100),
layer.Sigmoid(100, 100),
layer.Linear(100, 25),
layer.Sigmoid(25, 25),
layer.Linear(25, 10),
layer.Softmax(10, 10)]
name = 'network2' + '-' + str(i2) + '-' + str(i3) + '-' + str(i4) + '.dump'
myNN = NN(layers, learning_rate=learning_rates[i4], regularizer=regularizers[i2], momentum=momentums[i3])
myNN.train(x_train, y_train, x_valid, y_valid, epoch=300, batch_size=32)
if withBN:
for i2 in range(len(regularizers)):
for i3 in range(len(momentums)):
for i4 in range(len(learning_rates)):
layers = [layer.Linear(784, 100),
layer.BN(100, 100),
myNN.train_dump(x_train, y_train, x_valid, y_valid, epoch=200, dump_file=os.path.join(path, '../temp/network-3.dump'))
'''
hidden_units = [20, 100, 200, 500]
regularizers = [0.0, 0.0001, 0.001]
momentums = [0.0, 0.5, 0.9]
learning_rates = [0.1, 0.01, 0.2, 0.5]
for i1 in range(len(hidden_units)):
for i2 in range(len(regularizers)):
for i3 in range(len(momentums)):
for i4 in range(len(learning_rates)):
layers = [
layer.Linear(784, 100),
layer.BN(100, 100),
layer.ReLU(100, 100),
layer.BN(100, 100),
layer.SoftmaxLayer(100, 10)
]
name = 'network-' + str(i1) + '-' + str(i2) + '-' + str(
i3) + '-' + str(i4) + '.dump'
myNN = NN(layers,
learning_rate=learning_rates[i4],
regularizer=regularizers[i2],
momentum=momentums[i3])
myNN.train(x_train,
y_train,
x_valid,
y_valid,