def forward(self, inputs, history=None, **kwargs) -> Distribution:
"""
Note that history is ignored.
"""
outputs = self.conditioner(inputs, **kwargs)
return parameterize_conditional(self.dist_type, outputs,
self.event_size)
def forward(self, inputs, **kwargs) -> Distribution:
"""
:param inputs: [..., input_size]
:return: a parameterized Distribution
"""
outputs = self.conditioner(inputs, **kwargs)
return parameterize_conditional(self.dist_type, outputs, self.event_size)
def forward(self, inputs, history, **kwargs) -> Distribution:
"""
Note that history is required.
"""
if inputs is None:
h = history
else:
h = torch.cat([history, inputs], -1)
outputs = self.conditioner(h, **kwargs)
return parameterize_conditional(self.dist_type, outputs,
self.event_size)