def get_final_states(self) -> RnnStateStorage:
return self._states
class _MisoSeq2SeqWrapper(_Seq2SeqWrapper):
"""
It inherits ``_Seq2SeqWrapper'' from AllenNLP such that we have more flexibility
of defining our own seq2seq encoders.
"""
def __init__(self, *args, **kwargs) -> None:
super().__init__(*args, **kwargs)
def __call__(self, **kwargs) -> PytorchSeq2SeqWrapper:
return self.from_params(Params(kwargs))
# Logic requires custom from_params
def from_params(self, params: Params) -> _PytorchSeq2SeqWrapper:
if not params.pop_bool('batch_first', True):
raise ConfigurationError(
"Our encoder semantics assumes batch is always first!")
if self._module_class in self.PYTORCH_MODELS:
params['batch_first'] = True
stateful = params.pop_bool('stateful', False)
module = self._module_class(**params.as_dict(infer_type_and_cast=True))
return _PytorchSeq2SeqWrapper(module, stateful=stateful)
# pylint: disable=protected-access
Seq2SeqEncoder.register("miso_stacked_bilstm")(
_MisoSeq2SeqWrapper(MisoStackedBidirectionalLstm))
if weight_dropout > 0.0:
module = weight_drop_factory(self._module_class)(
module_args=params.as_dict(infer_type_and_cast=True),
weights=all_recurrent_weights,
wdrop=weight_dropout,
variational=variational,
)
else:
module = self._module_class(**params.as_dict(
infer_type_and_cast=True))
return PytorchSeq2SeqWrapper(module, stateful=stateful)
Seq2SeqEncoder.register("weightdrop_gru")(_WeightDropSeq2SeqWrapper(
torch.nn.GRU))
Seq2SeqEncoder.register("weightdrop_lstm")(_WeightDropSeq2SeqWrapper(
torch.nn.LSTM))
if __name__ == '__main__':
import torch
# Input is (seq, batch, input)
x = torch.autograd.Variable(torch.randn(2, 1, 10)).cuda()
h0 = None
###
print('Testing WeightDrop')
print('=-=-=-=-=-=-=-=-=-=')
from typing import Optional, Tuple
import torch
from torch.nn.utils.rnn import PackedSequence, pack_padded_sequence, pad_packed_sequence
from allennlp.modules.stacked_bidirectional_lstm import StackedBidirectionalLstm
from allennlp.common.checks import ConfigurationError
from allennlp.modules.seq2seq_encoders import PytorchSeq2SeqWrapper, Seq2SeqEncoder
Seq2SeqEncoder.register("miso_stacked_bilstm")
class MisoStackedBidirectionalLstmSeq2SeqEncoder(PytorchSeq2SeqWrapper):
"""
Registered as a `Seq2SeqEncoder` with name "stacked_bidirectional_lstm".
"""
def __init__(
self,
input_size: int,
hidden_size: int,
num_layers: int,
recurrent_dropout_probability: float = 0.0,
layer_dropout_probability: float = 0.0,
use_highway: bool = True,
stateful: bool = False,
) -> None:
module = MisoStackedBidirectionalLstm(
input_size=input_size,
hidden_size=hidden_size,
num_layers=num_layers,
recurrent_dropout_probability=recurrent_dropout_probability,
layer_dropout_probability=layer_dropout_probability,
