nessai.flows.nets
Neural networks for use in flows.
Module Contents
Classes
A standard multi-layer perceptron. |
- class nessai.flows.nets.MLP(in_shape, out_shape, hidden_sizes, activation=F.relu, activate_output=False, dropout_probability=0.0)
Bases:
torch.nn.ModuleA standard multi-layer perceptron.
Based on the implementation in nflows and modified to include dropout and conditional inputs.
- Parameters:
- in_shapetuple
Input shape.
- out_shapetuple
Output shape.
- hidden_sizesList[int]
Number of neurons in the hidden layers.
- activationCallable
Activation function
- activate_outputUnion[bool, Callable]
Whether to activate the output layer. If a bool is specified the same activation function is used. If a callable inputs is specified, it will be used for the activation.
- dropout_probabilityfloat
Amount of dropout to apply after the hidden layers.
- forward(inputs, context=None)
Forward method that allows for kwargs such as context.
- Parameters:
- inputs
torch.tensor Inputs to the MLP
- contextNone
Conditional inputs, must be None. Only implemented to the function is compatible with other methods.
- inputs
- Raises:
- ValueError
If the context is not None.
- ValueError
If the input shape is incorrect.