Python Model.set_updates方法代码示例

# 需要导入模块: from cle.cle.models import Model [as 别名]
# 或者: from cle.cle.models.Model import set_updates [as 别名]
ts, _, _ = y.shape
post_scan_shape = ((ts*batch_size, -1))
h1_in = h1_temp.reshape(post_scan_shape)
h2_in = h2_temp.reshape(post_scan_shape)
h3_in = h3_temp.reshape(post_scan_shape)
y_hat_in = output.fprop([h1_in, h2_in, h3_in])

cost = NllMulInd(y.flatten(), y_hat_in)
cost = cost.mean()
cost.name = 'cost'

model.inputs = [x, y]
model._params = params
model.nodes = nodes
model.set_updates(update_list)

optimizer = Adam(
    lr=0.001
)

extension = [
    GradientClipping(batch_size=batch_size),
    EpochCount(100),
    Monitoring(freq=100,
               ddout=[cost]),
    Picklize(freq=100, path=save_path)
]

mainloop = Training(
    name='toy_enwiki_gflstm',

# 需要导入模块: from cle.cle.models import Model [as 别名]
# 或者: from cle.cle.models.Model import set_updates [as 别名]
mean_theta_sig.name = 'mean_theta_sig'
min_theta_sig.name = 'min_theta_sig'

coeff_max = m_coeff_in.max()
coeff_min = m_coeff_in.min()
coeff_mean_max = m_coeff_in.mean(axis=0).max()
coeff_mean_min = m_coeff_in.mean(axis=0).min()
coeff_max.name = 'coeff_max'
coeff_min.name = 'coeff_min'
coeff_mean_max.name = 'coeff_mean_max'
coeff_mean_min.name = 'coeff_mean_min'

model.inputs = [x]
model.params = params
model.nodes = nodes
model.set_updates(shared_updates)

optimizer = Adam(
    lr=lr
)

monitor_fn = theano.function(inputs=[m_x],
                             outputs=[m_recon_term,
                                      max_theta_sig, mean_theta_sig, min_theta_sig,
                                      max_x, mean_x, min_x,
                                      max_theta_mu, mean_theta_mu, min_theta_mu,
                                      coeff_max, coeff_min, coeff_mean_max, coeff_mean_min],
                             on_unused_input='ignore')

extension = [
    GradientClipping(batch_size=batch_size, check_nan=1),

# 需要导入模块: from cle.cle.models import Model [as 别名]
# 或者: from cle.cle.models.Model import set_updates [as 别名]

#.........这里部分代码省略.........
    m_theta_sig_temp = theta_sig.fprop([m_theta_4_temp], params)
    m_coeff_temp = coeff.fprop([m_theta_4_temp], params)

    m_kl_temp = KLGaussianGaussian(m_phi_mu_temp, m_phi_sig_temp, m_prior_mu_temp, m_prior_sig_temp)

    m_x_shape = m_x.shape
    m_x_in = m_x.reshape((m_x_shape[0]*m_x_shape[1], -1))
    m_theta_mu_in = m_theta_mu_temp.reshape((m_x_shape[0]*m_x_shape[1], -1))
    m_theta_sig_in = m_theta_sig_temp.reshape((m_x_shape[0]*m_x_shape[1], -1))
    m_coeff_in = m_coeff_temp.reshape((m_x_shape[0]*m_x_shape[1], -1))

    m_recon = GMM(m_x_in, m_theta_mu_in, m_theta_sig_in, m_coeff_in)
    m_recon_term = m_recon.mean()
    m_kl_term = m_kl_temp.mean()
    m_nll_upper_bound = m_recon_term + m_kl_term
    m_nll_upper_bound.name = 'nll_upper_bound'
    m_recon_term.name = 'recon_term'
    m_kl_term.name = 'kl_term'

    max_x = m_x.max()
    mean_x = m_x.mean()
    min_x = m_x.min()
    max_x.name = 'max_x'
    mean_x.name = 'mean_x'
    min_x.name = 'min_x'

    max_theta_mu = m_theta_mu_in.max()
    mean_theta_mu = m_theta_mu_in.mean()
    min_theta_mu = m_theta_mu_in.min()
    max_theta_mu.name = 'max_theta_mu'
    mean_theta_mu.name = 'mean_theta_mu'
    min_theta_mu.name = 'min_theta_mu'

    max_theta_sig = m_theta_sig_in.max()
    mean_theta_sig = m_theta_sig_in.mean()
    min_theta_sig = m_theta_sig_in.min()
    max_theta_sig.name = 'max_theta_sig'
    mean_theta_sig.name = 'mean_theta_sig'
    min_theta_sig.name = 'min_theta_sig'

    max_phi_sig = m_phi_sig_temp.max()
    mean_phi_sig = m_phi_sig_temp.mean()
    min_phi_sig = m_phi_sig_temp.min()
    max_phi_sig.name = 'max_phi_sig'
    mean_phi_sig.name = 'mean_phi_sig'
    min_phi_sig.name = 'min_phi_sig'

    max_prior_sig = m_prior_sig_temp.max()
    mean_prior_sig = m_prior_sig_temp.mean()
    min_prior_sig = m_prior_sig_temp.min()
    max_prior_sig.name = 'max_prior_sig'
    mean_prior_sig.name = 'mean_prior_sig'
    min_prior_sig.name = 'min_prior_sig'

    model.inputs = [x]
    model.params = params
    model.nodes = nodes
    model.set_updates(shared_updates)

    optimizer = Adam(
        lr=lr
    )

    monitor_fn = theano.function(inputs=[m_x],
                                 outputs=[m_nll_upper_bound, m_recon_term, m_kl_term,
                                          max_phi_sig, mean_phi_sig, min_phi_sig,
                                          max_prior_sig, mean_prior_sig, min_prior_sig,
                                          max_theta_sig, mean_theta_sig, min_theta_sig,
                                          max_x, mean_x, min_x,
                                          max_theta_mu, mean_theta_mu, min_theta_mu],
                                 on_unused_input='ignore')

    extension = [
        GradientClipping(batch_size=batch_size, check_nan=1),
        EpochCount(epoch),
        Monitoring(freq=monitoring_freq,
                   monitor_fn=monitor_fn,
                   ddout=[m_nll_upper_bound, m_recon_term, m_kl_term,
                          max_phi_sig, mean_phi_sig, min_phi_sig,
                          max_prior_sig, mean_prior_sig, min_prior_sig,
                          max_theta_sig, mean_theta_sig, min_theta_sig,
                          max_x, mean_x, min_x,
                          max_theta_mu, mean_theta_mu, min_theta_mu],
                   data=[Iterator(train_data, m_batch_size, start=0, end=112640),
                         Iterator(valid_data, m_batch_size, start=2040064, end=2152704)]),
        Picklize(freq=monitoring_freq, force_save_freq=force_saving_freq, path=save_path),
        EarlyStopping(freq=monitoring_freq, force_save_freq=force_saving_freq, path=save_path, channel=channel_name),
        WeightNorm()
    ]

    mainloop = Training(
        name=pkl_name,
        data=Iterator(train_data, batch_size, start=0, end=2040064),
        model=model,
        optimizer=optimizer,
        cost=nll_upper_bound,
        outputs=[nll_upper_bound],
        extension=extension
    )
    mainloop.run()

# 需要导入模块: from cle.cle.models import Model [as 别名]
# 或者: from cle.cle.models.Model import set_updates [as 别名]

#.........这里部分代码省略.........
    min_x.name = "min_x"

    max_theta_mu = m_theta_mu_temp.max()
    mean_theta_mu = m_theta_mu_temp.mean()
    min_theta_mu = m_theta_mu_temp.min()
    max_theta_mu.name = "max_theta_mu"
    mean_theta_mu.name = "mean_theta_mu"
    min_theta_mu.name = "min_theta_mu"

    max_theta_sig = m_theta_sig_temp.max()
    mean_theta_sig = m_theta_sig_temp.mean()
    min_theta_sig = m_theta_sig_temp.min()
    max_theta_sig.name = "max_theta_sig"
    mean_theta_sig.name = "mean_theta_sig"
    min_theta_sig.name = "min_theta_sig"

    max_phi_sig = m_phi_sig_temp.max()
    mean_phi_sig = m_phi_sig_temp.mean()
    min_phi_sig = m_phi_sig_temp.min()
    max_phi_sig.name = "max_phi_sig"
    mean_phi_sig.name = "mean_phi_sig"
    min_phi_sig.name = "min_phi_sig"

    max_prior_sig = m_prior_sig_temp.max()
    mean_prior_sig = m_prior_sig_temp.mean()
    min_prior_sig = m_prior_sig_temp.min()
    max_prior_sig.name = "max_prior_sig"
    mean_prior_sig.name = "mean_prior_sig"
    min_prior_sig.name = "min_prior_sig"

    model.inputs = [x]
    model.params = params
    model.nodes = nodes
    model.set_updates(shared_updates)

    optimizer = Adam(lr=lr)

    monitor_fn = theano.function(
        inputs=[m_x],
        outputs=[
            m_nll_upper_bound,
            m_recon_term,
            m_kl_term,
            max_phi_sig,
            mean_phi_sig,
            min_phi_sig,
            max_prior_sig,
            mean_prior_sig,
            min_prior_sig,
            max_theta_sig,
            mean_theta_sig,
            min_theta_sig,
            max_x,
            mean_x,
            min_x,
            max_theta_mu,
            mean_theta_mu,
            min_theta_mu,
        ],
        on_unused_input="ignore",
    )

    extension = [
        GradientClipping(batch_size=batch_size, check_nan=1),
        EpochCount(epoch),
        Monitoring(