reagent.models package¶
Submodules¶
reagent.models.actor module¶
reagent.models.base module¶
-
class
reagent.models.base.ModelBase(*args, **kwargs)¶ Bases:
torch.nn.ModuleA base class to support exporting through ONNX
-
cpu_model()¶ Override this in DistributedDataParallel models
-
feature_config() → Optional[reagent.types.ModelFeatureConfig]¶ If the model needs additional preprocessing, e.g., using sequence features, returns the config here.
-
get_distributed_data_parallel_model()¶ Return DistributedDataParallel version of this model
This needs to be implemented explicitly because: 1) Model with EmbeddingBag module is not compatible with vanilla DistributedDataParallel 2) Exporting logic needs structured data. DistributedDataParallel doesn’t work with structured data.
-
get_target_network()¶ Return a copy of this network to be used as target network
Subclass should override this if the target network should share parameters with the network to be trained.
-
input_prototype() → Any¶ This function provides the input for ONNX graph tracing.
The return value should be what expected by forward().
-
reagent.models.bcq module¶
reagent.models.categorical_dqn module¶
reagent.models.cem_planner module¶
reagent.models.containers module¶
reagent.models.convolutional_network module¶
reagent.models.critic module¶
reagent.models.dqn module¶
reagent.models.dueling_q_network module¶
reagent.models.embedding_bag_concat module¶
reagent.models.fully_connected_network module¶
-
class
reagent.models.fully_connected_network.FullyConnectedNetwork(layers, activations, *, use_batch_norm=False, min_std=0.0, dropout_ratio=0.0, use_layer_norm=False, normalize_output=False)¶ Bases:
reagent.models.base.ModelBase-
forward(input: torch.Tensor) → torch.Tensor¶ Forward pass for generic feed-forward DNNs. Assumes activation names are valid pytorch activation names. :param input tensor
-
input_prototype()¶ This function provides the input for ONNX graph tracing.
The return value should be what expected by forward().
-
-
reagent.models.fully_connected_network.gaussian_fill_w_gain(tensor, activation, dim_in, min_std=0.0) → None¶ Gaussian initialization with gain.