Exemplo n.º 1
0
 def get_default_config() -> ConfigDict:
     c = ConfigDict(
         num_components=1,
         train_epochs=1000,
         # Component
         components_learning_rate=1e-3,
         components_batch_size=1000,
         components_num_epochs=10,
         components_net_reg_loss_fact=0.,
         components_net_drop_prob=0.0,
         components_net_hidden_layers=[50, 50],
         # Gating
         gating_learning_rate=1e-3,
         gating_batch_size=1000,
         gating_num_epochs=10,
         gating_net_reg_loss_fact=0.,
         gating_net_drop_prob=0.0,
         gating_net_hidden_layers=[50, 50],
         # Density Ratio Estimation
         dre_reg_loss_fact=
         0.0,  # Scaling Factor for L2 regularization of density ratio estimator
         dre_early_stopping=
         True,  # Use early stopping for density ratio estimator training
         dre_drop_prob=
         0.0,  # If smaller than 1 dropout with keep prob = 'keep_prob' is used
         dre_num_iters=
         1000,  # Number of density ratio estimator steps each iteration (i.e. max number if early stopping)
         dre_batch_size=
         1000,  # Batch size for density ratio estimator training
         dre_hidden_layers=[
             30, 30
         ]  # width of density ratio estimator  hidden layers
     )
     c.finalize_adding()
     return c
Exemplo n.º 2
0
 def get_default_config():
     c = ConfigDict(
         num_components=1,
         samples_per_component=500,
         train_epochs=1000,
         initialization="random",
         # Component Updates
         component_kl_bound=0.01,
         # Mixture Updates
         weight_kl_bound=0.01,
         # Density Ratio Estimation
         dre_reg_loss_fact=
         0.0,  # Scaling Factor for L2 regularization of density ratio estimator
         dre_early_stopping=
         True,  # Use early stopping for density ratio estimator training
         dre_drop_prob=
         0.0,  # If smaller than 1 dropout with keep prob = 'keep_prob' is used
         dre_num_iters=
         1000,  # Number of density ratio estimator steps each iteration (i.e. max number if early stopping)
         dre_batch_size=
         1000,  # Batch size for density ratio estimator training
         dre_hidden_layers=[
             30, 30
         ]  # width of density ratio estimator  hidden layers
     )
     c.finalize_adding()
     return c
Exemplo n.º 3
0
 def get_default_config() -> ConfigDict:
     config = ConfigDict(num_basis=15,
                         bandwidth=3,
                         trans_net_hidden_units=[64, 64],
                         trans_net_hidden_activation="Tanh",
                         learn_trans_covar=True,
                         trans_covar=0.1,
                         learn_initial_state_covar=False,
                         initial_state_covar=10,
                         learning_rate=1e-3,
                         enc_out_norm='pre',
                         clip_gradients=True,
                         never_invalid=True)
     config.finalize_adding()
     return config
Exemplo n.º 4
0
 def get_default_config() -> ConfigDict:
     config = ConfigDict(num_basis=15,
                         bandwidth=3,
                         trans_net_hidden_units=[],
                         control_net_hidden_units=[60],
                         trans_net_hidden_activation="Tanh",
                         control_net_hidden_activation='ReLU',
                         learn_trans_covar=True,
                         trans_covar=1,
                         learn_initial_state_covar=True,
                         initial_state_covar=1,
                         learning_rate=7e-3,
                         enc_out_norm='post',
                         clip_gradients=True,
                         never_invalid=True)
     config.finalize_adding()
     return config