Python models.Model类代码示例

)
from cle.cle.train.opt import Adam
from cle.cle.utils import unpack, OrderedDict
from cle.datasets.bouncing_balls import BouncingBalls


#data_path = '/data/lisatmp3/chungjun/bouncing_balls/bouncing_ball_2balls_16wh_20len_50000cases.npy'
#save_path = '/u/chungjun/repos/cle/saved/'
data_path = '/home/junyoung/data/bouncing_balls/bouncing_ball_2balls_16wh_20len_50000cases.npy'
save_path = '/home/junyoung/repos/cle/saved/'

batch_size = 128
res = 256
debug = 0

model = Model()
trdata = BouncingBalls(name='train',
                       path=data_path)

init_W = InitCell('randn')
init_U = InitCell('ortho')
init_b = InitCell('zeros')

model.inputs = trdata.theano_vars()
x, y = model.inputs
if debug:
    x.tag.test_value = np.zeros((10, batch_size, res), dtype=np.float32)
    y.tag.test_value = np.zeros((10, batch_size, res), dtype=np.float32)

inputs = [x, y]
inputs_dim = {'x':256, 'y':256}

#data_path = '/home/junyoung/data/timit/readable/'
#save_path = '/home/junyoung/repos/sk/cle/models/nips2015/timit/pkl/'
data_path = '/data/lisa/data/timit/readable/'
save_path = '/data/lisatmp/chungjun/nips2015/timit/pkl/'

batch_size = 64
frame_size = 200
main_lstm_dim = 2000
p_x_dim = 450
x2s_dim = 450
k = 20
target_size = frame_size * k
lr = 1e-3
debug = 0

model = Model()
train_data = TIMIT(name='train',
                   path=data_path,
                   frame_size=frame_size,
                   shuffle=0,
                   use_n_gram=1)

X_mean = train_data.X_mean
X_std = train_data.X_std

valid_data = TIMIT(name='valid',
                   path=data_path,
                   frame_size=frame_size,
                   shuffle=0,
                   use_n_gram=1,
                   X_mean=X_mean,

    GradientClipping,
    Monitoring,
    Picklize
)
from cle.cle.train.opt import Adam
from cle.cle.utils import flatten, sharedX, unpack, OrderedDict
from cle.datasets.enwiki import EnWiki

data_path = '/home/junyoung/data/wikipedia-text/enwiki_char_and_word.npz'
save_path = '/home/junyoung/src/cle/saved/'

batch_size = 100
reset_freq = 100
debug = 0

model = Model()
train_data = EnWiki(name='train',
                    path=data_path)

test_data = EnWiki(name='test',
                   path=data_path)

init_W = InitCell('rand')
init_U = InitCell('ortho')
init_b = InitCell('zeros')

x, y = train_data.theano_vars()
if debug:
    x.tag.test_value = np.zeros((10, batch_size, 1), dtype=np.float32)
    y.tag.test_value = np.zeros((10, batch_size, 1), dtype=np.float32)

    Monitoring,
    Picklize
)
from cle.cle.train.opt import Adam
from cle.cle.utils import init_tparams, OrderedDict
from cle.datasets.bouncing_balls import BouncingBalls


data_path = '/data/lisatmp3/chungjun/bouncing_balls/bouncing_ball_2balls_16wh_20len_50000cases.npy'
save_path = '/u/chungjun/repos/cle/saved/'

batch_size = 128
frame_size = 256
debug = 0

model = Model()
train_data = BouncingBalls(name='train',
                           path=data_path)

valid_data = BouncingBalls(name='valid',
                           path=data_path)

x, y = train_data.theano_vars()

if debug:
    x.tag.test_value = np.zeros((10, batch_size, frame_size), dtype=np.float32)
    y.tag.test_value = np.zeros((10, batch_size, frame_size), dtype=np.float32)

init_W = InitCell('randn')
init_U = InitCell('ortho')
init_b = InitCell('zeros')

def main(args):

    trial = int(args['trial'])
    pkl_name = 'rnn_gauss_%d' % trial
    channel_name = 'valid_nll'

    data_path = args['data_path']
    save_path = args['save_path']

    monitoring_freq = int(args['monitoring_freq'])
    epoch = int(args['epoch'])
    batch_size = int(args['batch_size'])
    x_dim = int(args['x_dim'])
    z_dim = int(args['z_dim'])
    rnn_dim = int(args['rnn_dim'])
    lr = float(args['lr'])
    debug = int(args['debug'])

    print "trial no. %d" % trial
    print "batch size %d" % batch_size
    print "learning rate %f" % lr
    print "saving pkl file '%s'" % pkl_name
    print "to the save path '%s'" % save_path

    x2s_dim = 340
    s2x_dim = 340
    target_dim = x_dim - 1

    model = Model()
    train_data = IAMOnDB(name='train',
                         prep='normalize',
                         cond=False,
                         path=data_path)

    X_mean = train_data.X_mean
    X_std = train_data.X_std

    valid_data = IAMOnDB(name='valid',
                         prep='normalize',
                         cond=False,
                         path=data_path,
                         X_mean=X_mean,
                         X_std=X_std)

    init_W = InitCell('rand')
    init_U = InitCell('ortho')
    init_b = InitCell('zeros')
    init_b_sig = InitCell('const', mean=0.6)

    x, mask = train_data.theano_vars()

    if debug:
        x.tag.test_value = np.zeros((15, batch_size, x_dim), dtype=np.float32)
        temp = np.ones((15, batch_size), dtype=np.float32)
        temp[:, -2:] = 0.
        mask.tag.test_value = temp


    x_1 = FullyConnectedLayer(name='x_1',
                              parent=['x_t'],
                              parent_dim=[x_dim],
                              nout=x2s_dim,
                              unit='relu',
                              init_W=init_W,
                              init_b=init_b)

    rnn = LSTM(name='rnn',
               parent=['x_1'],
               parent_dim=[x2s_dim],
               nout=rnn_dim,
               unit='tanh',
               init_W=init_W,
               init_U=init_U,
               init_b=init_b)

    theta_1 = FullyConnectedLayer(name='theta_1',
                                  parent=['s_tm1'],
                                  parent_dim=[rnn_dim],
                                  nout=s2x_dim,
                                  unit='relu',
                                  init_W=init_W,
                                  init_b=init_b)

    theta_mu = FullyConnectedLayer(name='theta_mu',
                                   parent=['theta_1'],
                                   parent_dim=[s2x_dim],
                                   nout=target_dim,
                                   unit='linear',
                                   init_W=init_W,
                                   init_b=init_b)

    theta_sig = FullyConnectedLayer(name='theta_sig',
                                    parent=['theta_1'],
                                    parent_dim=[s2x_dim],
                                    nout=target_dim,
                                    unit='softplus',
                                    cons=1e-4,
                                    init_W=init_W,
                                    init_b=init_b_sig)

#.........这里部分代码省略.........

def main(args):

    trial = int(args['trial'])
    pkl_name = 'vrnn_gmm_%d' % trial
    channel_name = 'valid_nll_upper_bound'

    data_path = args['data_path']
    save_path = args['save_path']

    monitoring_freq = int(args['monitoring_freq'])
    force_saving_freq = int(args['force_saving_freq'])
    reset_freq = int(args['reset_freq'])
    epoch = int(args['epoch'])
    batch_size = int(args['batch_size'])
    m_batch_size = int(args['m_batch_size'])
    x_dim = int(args['x_dim'])
    z_dim = int(args['z_dim'])
    rnn_dim = int(args['rnn_dim'])
    k = int(args['num_k'])
    lr = float(args['lr'])
    debug = int(args['debug'])

    print "trial no. %d" % trial
    print "batch size %d" % batch_size
    print "learning rate %f" % lr
    print "saving pkl file '%s'" % pkl_name
    print "to the save path '%s'" % save_path

    q_z_dim = 500
    p_z_dim = 500
    p_x_dim = 500
    x2s_dim = 500
    z2s_dim = 500
    target_dim = x_dim * k

    file_name = 'blizzard_unseg_tbptt'
    normal_params = np.load(data_path + file_name + '_normal.npz')
    X_mean = normal_params['X_mean']
    X_std = normal_params['X_std']

    model = Model()
    train_data = Blizzard_tbptt(name='train',
                                path=data_path,
                                frame_size=x_dim,
                                file_name=file_name,
                                X_mean=X_mean,
                                X_std=X_std)

    valid_data = Blizzard_tbptt(name='valid',
                                path=data_path,
                                frame_size=x_dim,
                                file_name=file_name,
                                X_mean=X_mean,
                                X_std=X_std)

    x = train_data.theano_vars()
    m_x = valid_data.theano_vars()

    if debug:
        x.tag.test_value = np.zeros((15, batch_size, x_dim), dtype=theano.config.floatX)
        m_x.tag.test_value = np.zeros((15, m_batch_size, x_dim), dtype=theano.config.floatX)

    init_W = InitCell('rand')
    init_U = InitCell('ortho')
    init_b = InitCell('zeros')
    init_b_sig = InitCell('const', mean=0.6)

    x_1 = FullyConnectedLayer(name='x_1',
                              parent=['x_t'],
                              parent_dim=[x_dim],
                              nout=x2s_dim,
                              unit='relu',
                              init_W=init_W,
                              init_b=init_b)

    x_2 = FullyConnectedLayer(name='x_2',
                              parent=['x_1'],
                              parent_dim=[x2s_dim],
                              nout=x2s_dim,
                              unit='relu',
                              init_W=init_W,
                              init_b=init_b)

    x_3 = FullyConnectedLayer(name='x_3',
                              parent=['x_2'],
                              parent_dim=[x2s_dim],
                              nout=x2s_dim,
                              unit='relu',
                              init_W=init_W,
                              init_b=init_b)

    x_4 = FullyConnectedLayer(name='x_4',
                              parent=['x_3'],
                              parent_dim=[x2s_dim],
                              nout=x2s_dim,
                              unit='relu',
                              init_W=init_W,
                              init_b=init_b)

    z_1 = FullyConnectedLayer(name='z_1',
#.........这里部分代码省略.........

    Monitoring,
    Picklize
)
from cle.cle.train.opt import RMSProp
from cle.cle.utils import init_tparams, sharedX
from cle.cle.utils.compat import OrderedDict
from cle.datasets.music import Music

data_path = '/home/junyoung/data/music/MuseData.pickle'
save_path = '/home/junyoung/repos/cle/saved/'

batch_size = 10
nlabel = 105
debug = 1

model = Model()
train_data = Music(name='train',
                   path=data_path,
                   nlabel=nlabel)

valid_data = Music(name='valid',
                   path=data_path,
                   nlabel=nlabel)

# Choose the random initialization method
init_W = InitCell('randn')
init_U = InitCell('ortho')
init_b = InitCell('zeros')

x, y, mask = train_data.theano_vars()
# You must use THEANO_FLAGS="compute_test_value=raise" python -m ipdb

def main(args):

    trial = int(args["trial"])
    pkl_name = "vrnn_gauss_%d" % trial
    channel_name = "valid_nll_upper_bound"

    data_path = args["data_path"]
    save_path = args["save_path"]

    monitoring_freq = int(args["monitoring_freq"])
    epoch = int(args["epoch"])
    batch_size = int(args["batch_size"])
    x_dim = int(args["x_dim"])
    z_dim = int(args["z_dim"])
    rnn_dim = int(args["rnn_dim"])
    lr = float(args["lr"])
    debug = int(args["debug"])

    print "trial no. %d" % trial
    print "batch size %d" % batch_size
    print "learning rate %f" % lr
    print "saving pkl file '%s'" % pkl_name
    print "to the save path '%s'" % save_path

    q_z_dim = 150
    p_z_dim = 150
    p_x_dim = 250
    x2s_dim = 250
    z2s_dim = 150
    target_dim = x_dim - 1

    model = Model()
    train_data = IAMOnDB(name="train", prep="normalize", cond=False, path=data_path)

    X_mean = train_data.X_mean
    X_std = train_data.X_std

    valid_data = IAMOnDB(name="valid", prep="normalize", cond=False, path=data_path, X_mean=X_mean, X_std=X_std)

    init_W = InitCell("rand")
    init_U = InitCell("ortho")
    init_b = InitCell("zeros")
    init_b_sig = InitCell("const", mean=0.6)

    x, mask = train_data.theano_vars()

    if debug:
        x.tag.test_value = np.zeros((15, batch_size, x_dim), dtype=np.float32)
        temp = np.ones((15, batch_size), dtype=np.float32)
        temp[:, -2:] = 0.0
        mask.tag.test_value = temp

    x_1 = FullyConnectedLayer(
        name="x_1", parent=["x_t"], parent_dim=[x_dim], nout=x2s_dim, unit="relu", init_W=init_W, init_b=init_b
    )

    z_1 = FullyConnectedLayer(
        name="z_1", parent=["z_t"], parent_dim=[z_dim], nout=z2s_dim, unit="relu", init_W=init_W, init_b=init_b
    )

    rnn = LSTM(
        name="rnn",
        parent=["x_1", "z_1"],
        parent_dim=[x2s_dim, z2s_dim],
        nout=rnn_dim,
        unit="tanh",
        init_W=init_W,
        init_U=init_U,
        init_b=init_b,
    )

    phi_1 = FullyConnectedLayer(
        name="phi_1",
        parent=["x_1", "s_tm1"],
        parent_dim=[x2s_dim, rnn_dim],
        nout=q_z_dim,
        unit="relu",
        init_W=init_W,
        init_b=init_b,
    )

    phi_mu = FullyConnectedLayer(
        name="phi_mu", parent=["phi_1"], parent_dim=[q_z_dim], nout=z_dim, unit="linear", init_W=init_W, init_b=init_b
    )

    phi_sig = FullyConnectedLayer(
        name="phi_sig",
        parent=["phi_1"],
        parent_dim=[q_z_dim],
        nout=z_dim,
        unit="softplus",
        cons=1e-4,
        init_W=init_W,
        init_b=init_b_sig,
    )

    prior_1 = FullyConnectedLayer(
        name="prior_1", parent=["s_tm1"], parent_dim=[rnn_dim], nout=p_z_dim, unit="relu", init_W=init_W, init_b=init_b
    )

#.........这里部分代码省略.........

from cle.cle.train.opt import Adam
from cle.cle.utils import flatten
from cle.cle.utils.compat import OrderedDict
from cle.datasets.mnist import MNIST


datapath = '/home/junyoung/data/mnist/mnist_binarized_salakhutdinov.pkl'
savepath = '/home/junyoung/repos/cle/saved/'

batch_size = 100
inpsz = 784
latsz = 100
n_steps = 64
debug = 0

model = Model()
data = MNIST(name='train',
             unsupervised=1,
             path=datapath)

init_W = InitCell('rand')
init_U = InitCell('ortho')
init_b = InitCell('zeros')
init_b_sig = InitCell('const', mean=0.6)

x, _ = data.theano_vars()
if debug:
    x.tag.test_value = np.zeros((batch_size, 784), dtype=np.float32)

error = ErrorLayer(name='error',
                   parent=['x'],

frame_size = 200
latent_size = 200
rnn_dim = 4000
p_x_dim = 700
x2s_dim = 700
k = 20
target_size = frame_size * k
lr = 3e-4
debug = 0

file_name = 'blizzard_unseg_tbptt'
normal_params = np.load(data_path + file_name + '_normal.npz')
X_mean = normal_params['X_mean']
X_std = normal_params['X_std']

model = Model()
model = Model()
train_data = Blizzard_tbptt(name='train',
                            path=data_path,
                            frame_size=frame_size,
                            file_name=file_name,
                            X_mean=X_mean,
                            X_std=X_std)

valid_data = Blizzard_tbptt(name='valid',
                            path=data_path,
                            frame_size=frame_size,
                            file_name=file_name,
                            X_mean=X_mean,
                            X_std=X_std)

def main(args):

    trial = int(args["trial"])
    pkl_name = "vrnn_gauss_%d" % trial
    channel_name = "valid_nll_upper_bound"

    data_path = args["data_path"]
    save_path = args["save_path"]
    data_path = os.path.expanduser(args["data_path"])
    save_path = os.path.expanduser(args["save_path"])
    monitoring_freq = int(args["monitoring_freq"])
    force_saving_freq = int(args["force_saving_freq"])
    reset_freq = int(args["reset_freq"])
    epoch = int(args["epoch"])
    batch_size = int(args["batch_size"])
    m_batch_size = int(args["m_batch_size"])
    x_dim = int(args["x_dim"])
    z_dim = int(args["z_dim"])
    rnn_dim = int(args["rnn_dim"])
    lr = float(args["lr"])
    debug = int(args["debug"])

    print "trial no. %d" % trial
    print "batch size %d" % batch_size
    print "learning rate %f" % lr
    print "saving pkl file '%s'" % pkl_name
    print "to the save path '%s'" % save_path

    q_z_dim = 500
    p_z_dim = 500
    p_x_dim = 600
    x2s_dim = 600
    z2s_dim = 500
    target_dim = x_dim

    file_name = "blizzard_tbptt"
    normal_params = np.load(data_path + file_name + "_normal.npz")
    X_mean = normal_params["X_mean"]
    X_std = normal_params["X_std"]

    model = Model()
    train_data = Blizzard_tbptt(
        name="train", path=data_path, frame_size=x_dim, file_name=file_name, X_mean=X_mean, X_std=X_std
    )

    valid_data = Blizzard_tbptt(
        name="valid", path=data_path, frame_size=x_dim, file_name=file_name, X_mean=X_mean, X_std=X_std
    )

    x = train_data.theano_vars()
    m_x = valid_data.theano_vars()

    if debug:
        x.tag.test_value = np.zeros((15, batch_size, x_dim), dtype=theano.config.floatX)
        m_x.tag.test_value = np.zeros((15, m_batch_size, x_dim), dtype=theano.config.floatX)

    init_W = InitCell("rand")
    init_U = InitCell("ortho")
    init_b = InitCell("zeros")
    init_b_sig = InitCell("const", mean=0.6)

    x_1 = FullyConnectedLayer(
        name="x_1", parent=["x_t"], parent_dim=[x_dim], nout=x2s_dim, unit="relu", init_W=init_W, init_b=init_b
    )

    x_2 = FullyConnectedLayer(
        name="x_2", parent=["x_1"], parent_dim=[x2s_dim], nout=x2s_dim, unit="relu", init_W=init_W, init_b=init_b
    )

    x_3 = FullyConnectedLayer(
        name="x_3", parent=["x_2"], parent_dim=[x2s_dim], nout=x2s_dim, unit="relu", init_W=init_W, init_b=init_b
    )

    x_4 = FullyConnectedLayer(
        name="x_4", parent=["x_3"], parent_dim=[x2s_dim], nout=x2s_dim, unit="relu", init_W=init_W, init_b=init_b
    )

    z_1 = FullyConnectedLayer(
        name="z_1", parent=["z_t"], parent_dim=[z_dim], nout=z2s_dim, unit="relu", init_W=init_W, init_b=init_b
    )

    z_2 = FullyConnectedLayer(
        name="z_2", parent=["z_1"], parent_dim=[z2s_dim], nout=z2s_dim, unit="relu", init_W=init_W, init_b=init_b
    )

    z_3 = FullyConnectedLayer(
        name="z_3", parent=["z_2"], parent_dim=[z2s_dim], nout=z2s_dim, unit="relu", init_W=init_W, init_b=init_b
    )

    z_4 = FullyConnectedLayer(
        name="z_4", parent=["z_3"], parent_dim=[z2s_dim], nout=z2s_dim, unit="relu", init_W=init_W, init_b=init_b
    )

    rnn = LSTM(
        name="rnn",
        parent=["x_4", "z_4"],
        parent_dim=[x2s_dim, z2s_dim],
        nout=rnn_dim,
        unit="tanh",
        init_W=init_W,
#.........这里部分代码省略.........

# Set your dataset
data_path = ['/data/lisa/data/cifar10/pylearn2_gcn_whitened/train.npy',
            '/u/chungjun/repos/cle/labels/trainy.npy']
test_data_path = ['/data/lisa/data/cifar10/pylearn2_gcn_whitened/test.npy',
                '/u/chungjun/repos/cle/labels/testy.npy']
save_path = '/u/chungjun/repos/cle/saved/'
#data_path = ['/home/junyoung/data/cifar10/pylearn2_gcn_whitened/train.npy',
#            '/home/junyoung/data/cifar10/pylearn2_gcn_whitened/trainy.npy']
#test_data_path = ['/home/junyoung/data/cifar10/pylearn2_gcn_whitened/test.npy',
#                '/home/junyoung/data/cifar10/pylearn2_gcn_whitened/testy.npy']
#save_path = '/home/junyoung/repos/cle/saved/'

batch_size = 128
debug = 0

model = Model()
train_data = CIFAR10(name='train',
                     path=data_path)

test_data = CIFAR10(name='test',
                    path=test_data_path)

# Choose the random initialization method
init_W = InitCell('rand')
init_b = InitCell('zeros')

# Define nodes: objects
model.inputs = train_data.theano_vars()
x, y = model.inputs
# You must use THEANO_FLAGS="compute_test_value=raise" python -m ipdb
if debug: