Example Models

PyTorch models used in example scripts.

Submodules

syllabus.examples.models.minigrid_model module

class syllabus.examples.models.minigrid_model.Categorical(num_inputs, num_outputs)

Bases: Module

Categorical distribution (NN module)

forward(x)

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class syllabus.examples.models.minigrid_model.FixedCategorical(probs=None, logits=None, validate_args=None)

Bases: Categorical

Categorical distribution object

log_probs(actions)

mode()

Returns the mode of the distribution.

sample()

Generates a sample_shape shaped sample or sample_shape shaped batch of samples if the distribution parameters are batched.

class syllabus.examples.models.minigrid_model.MinigridAgent(obs_shape, num_actions, arch='small', base_kwargs=None)

Bases: MinigridPolicy

get_action_and_value(x, action=None, full_log_probs=False)

get_value(x)

class syllabus.examples.models.minigrid_model.MinigridPolicy(obs_shape, num_actions, arch='small', base_kwargs=None)

Bases: Module

Actor-Critic module

act(inputs, deterministic=False)

evaluate_actions(inputs, action)

forward()

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

get_value(inputs)

property is_recurrent

property recurrent_hidden_state_size

Size of rnn_hx.

syllabus.examples.models.minigrid_model.apply_init_(modules)

Initialize NN modules

syllabus.examples.models.minigrid_model.init(module, weight_init, bias_init, gain=1)

syllabus.examples.models.minigrid_model.init_(m)

syllabus.examples.models.minigrid_model.init_relu_(m)

syllabus.examples.models.minigrid_model.init_tanh_(m)

syllabus.examples.models.procgen_model module

class syllabus.examples.models.procgen_model.BasicBlock(n_channels, stride=1)

Bases: Module

Residual Network Block

forward(x)

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class syllabus.examples.models.procgen_model.Categorical(num_inputs, num_outputs)

Bases: Module

Categorical distribution (NN module)

forward(x)

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class syllabus.examples.models.procgen_model.Conv2d_tf(*args, **kwargs)

Bases: Conv2d

Conv2d with the padding behavior from TF

forward(input)

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class syllabus.examples.models.procgen_model.FixedCategorical(probs=None, logits=None, validate_args=None)

Bases: Categorical

Categorical distribution object

log_probs(actions)

mode()

Returns the mode of the distribution.

sample()

Generates a sample_shape shaped sample or sample_shape shaped batch of samples if the distribution parameters are batched.

class syllabus.examples.models.procgen_model.Flatten(*args, **kwargs)

Bases: Module

Flatten a tensor

forward(x)

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class syllabus.examples.models.procgen_model.NNBase(hidden_size)

Bases: Module

Actor-Critic network (base class)

property output_size

class syllabus.examples.models.procgen_model.ProcgenAgent(obs_shape, num_actions, base_kwargs=None, detach_critic=False)

Bases: Module

get_action(inputs)

get_action_and_value(inputs, action=None, deterministic=False)

get_pi(inputs)

get_pi_value_and_aux_value(inputs)

get_value(inputs)

class syllabus.examples.models.procgen_model.ProcgenLSTMAgent(obs_shape, num_actions, base_kwargs=None, detach_critic=False)

Bases: Module

get_action(inputs, lstm_state, done)

get_action_and_value(inputs, lstm_state, done, action=None, deterministic=False)

get_pi(inputs, lstm_state, done)

get_pi_value_and_aux_value(inputs, lstm_state, done)

get_states(inputs, lstm_state, done)

get_value(inputs, lstm_state, done)

class syllabus.examples.models.procgen_model.ResNetBase(num_inputs, hidden_size=256, channels=[16, 32, 32])

Bases: NNBase

Residual Network

forward(inputs)

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

syllabus.examples.models.procgen_model.apply_init_(modules)

Initialize NN modules

syllabus.examples.models.procgen_model.init(module, weight_init, bias_init, gain=1)

syllabus.examples.models.procgen_model.init_(m)

syllabus.examples.models.procgen_model.init_relu_(m)

syllabus.examples.models.procgen_model.init_tanh_(m)

syllabus.examples.models.procgen_model.layer_init(layer, std=1.4142135623730951, bias_const=0.0)