class RecurrentAttention(nn.Module):
def __init__(self,
input_size,
context_size,
hidden_size,
num_layers=1,
bias=True,
batch_first=False,
dropout=0):
super(RecurrentAttention, self).__init__()
self.rnn = nn.LSTM(input_size, hidden_size, num_layers, bias,
batch_first)
self.attn = BahdanauAttention(hidden_size,
context_size,
context_size,
normalize=True,
batch_first=batch_first)
self.dropout = nn.Dropout(dropout)
def forward(self, inputs, hidden, context, context_len):
# set attention mask, sequences have different lengths, this mask
# allows to include only valid elements of context in attention's
# softmax
self.attn.set_mask(context_len, context)
rnn_outputs, hidden = self.rnn(inputs, hidden)
attn_outputs, scores = self.attn(rnn_outputs, context)
rnn_outputs = self.dropout(rnn_outputs)
return rnn_outputs, hidden, attn_outputs, scores
class RecurrentAttention(nn.Module):
"""
LSTM with an attention module.
"""
def __init__(self,
input_size,
context_size,
hidden_size,
num_layers=1,
bias=True,
batch_first=False,
dropout=0):
"""
Constructor for the RecurrentAttention.
:param input_size: number of features in input tensor
:param context_size: number of features in output from encoder
:param hidden_size: internal hidden size
:param num_layers: number of layers in LSTM
:param bias: enables bias in LSTM layers
:param batch_first: if True the model uses (batch,seq,feature) tensors,
if false the model uses (seq, batch, feature)
:param dropout: probability of dropout
"""
super(RecurrentAttention, self).__init__()
self.rnn = nn.LSTM(input_size, hidden_size, num_layers, bias,
batch_first)
self.attn = BahdanauAttention(hidden_size,
context_size,
context_size,
normalize=True,
batch_first=batch_first)
self.dropout = nn.Dropout(dropout)
def forward(self, inputs, hidden, context, context_len):
"""
Execute RecurrentAttention.
:param inputs: tensor with inputs
:param hidden: hidden state for LSTM layer
:param context: context tensor from encoder
:param context_len: vector of encoder sequence lengths
:returns (rnn_outputs, hidden, attn_output, attn_scores)
"""
# set attention mask, sequences have different lengths, this mask
# allows to include only valid elements of context in attention's
# softmax
self.attn.set_mask(context_len, context)
rnn_outputs, hidden = self.rnn(inputs, hidden)
attn_outputs, scores = self.attn(rnn_outputs, context)
rnn_outputs = self.dropout(rnn_outputs)
return rnn_outputs, hidden, attn_outputs, scores
class RecurrentAttention(nn.Module):
def __init__(self,
input_size,
context_size,
hidden_size,
num_layers=1,
bias=True,
batch_first=False,
dropout=0):
super(RecurrentAttention, self).__init__()
self.rnn = nn.LSTM(input_size, hidden_size, num_layers, bias,
batch_first)
# SSY 2.1.1 seq2seq/models/attention.py nn.Linear torch.bmm
self.attn = BahdanauAttention(hidden_size,
context_size,
context_size,
normalize=True,
batch_first=batch_first)
self.dropout = nn.Dropout(dropout)
def forward(self, inputs, hidden, context, context_len):
# set attention mask, sequences have different lengths, this mask
# allows to include only valid elements of context in attention's
# softmax
self.attn.set_mask(context_len, context)
rnn_outputs, hidden = self.rnn(bf16cutfp.apply(inputs), hidden)
# SSY it seems to be None
#bf16cutfp.apply(hidden))
rnn_outputs = bf16cutbp.apply(rnn_outputs)
hidden = bf16cutbp.apply(hidden)
attn_outputs, scores = self.attn(rnn_outputs, context)
rnn_outputs = self.dropout(rnn_outputs)
return rnn_outputs, hidden, attn_outputs, scores