This is an unofficial PyTorch implementation by Ignacio Oguiza - oguiza@gmail.com based on:

class RNN_FCN[source]

RNN_FCN(c_in, c_out, seq_len=None, hidden_size=100, rnn_layers=1, bias=True, cell_dropout=0, rnn_dropout=0.8, bidirectional=False, shuffle=True, fc_dropout=0.0, conv_layers=[128, 256, 128], kss=[7, 5, 3], se=0) :: _RNN_FCN_Base

Same as nn.Module, but no need for subclasses to call super().__init__

class LSTM_FCN[source]

LSTM_FCN(c_in, c_out, seq_len=None, hidden_size=100, rnn_layers=1, bias=True, cell_dropout=0, rnn_dropout=0.8, bidirectional=False, shuffle=True, fc_dropout=0.0, conv_layers=[128, 256, 128], kss=[7, 5, 3], se=0) :: _RNN_FCN_Base

Same as nn.Module, but no need for subclasses to call super().__init__

class GRU_FCN[source]

GRU_FCN(c_in, c_out, seq_len=None, hidden_size=100, rnn_layers=1, bias=True, cell_dropout=0, rnn_dropout=0.8, bidirectional=False, shuffle=True, fc_dropout=0.0, conv_layers=[128, 256, 128], kss=[7, 5, 3], se=0) :: _RNN_FCN_Base

Same as nn.Module, but no need for subclasses to call super().__init__

class MRNN_FCN[source]

MRNN_FCN(*args, se=16, **kwargs) :: _RNN_FCN_Base

Same as nn.Module, but no need for subclasses to call super().__init__

class MLSTM_FCN[source]

MLSTM_FCN(*args, se=16, **kwargs) :: _RNN_FCN_Base

Same as nn.Module, but no need for subclasses to call super().__init__

class MGRU_FCN[source]

MGRU_FCN(*args, se=16, **kwargs) :: _RNN_FCN_Base

Same as nn.Module, but no need for subclasses to call super().__init__

bs = 16
n_vars = 3
seq_len = 12
c_out = 2
xb = torch.rand(bs, n_vars, seq_len)
test_eq(RNN_FCN(n_vars, c_out, seq_len)(xb).shape, [bs, c_out])
test_eq(LSTM_FCN(n_vars, c_out, seq_len)(xb).shape, [bs, c_out])
test_eq(MLSTM_FCN(n_vars, c_out, seq_len)(xb).shape, [bs, c_out])
test_eq(GRU_FCN(n_vars, c_out, shuffle=False)(xb).shape, [bs, c_out])
test_eq(GRU_FCN(n_vars, c_out, seq_len, shuffle=False)(xb).shape, [bs, c_out])

LSTM_FCN(n_vars, seq_len, c_out, se=8)

LSTM_FCN(
  (rnn): LSTM(2, 100, batch_first=True)
  (rnn_dropout): Dropout(p=0.8, inplace=False)
  (convblock1): ConvBlock(
    (0): Conv1d(3, 128, kernel_size=(7,), stride=(1,), padding=(3,), bias=False)
    (1): BatchNorm1d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (2): ReLU()
  )
  (se1): SqueezeExciteBlock(
    (avg_pool): GAP1d(
      (gap): AdaptiveAvgPool1d(output_size=1)
      (flatten): Flatten(full=False)
    )
    (fc): Sequential(
      (0): Linear(in_features=128, out_features=16, bias=False)
      (1): ReLU()
      (2): Linear(in_features=16, out_features=128, bias=False)
      (3): Sigmoid()
    )
  )
  (convblock2): ConvBlock(
    (0): Conv1d(128, 256, kernel_size=(5,), stride=(1,), padding=(2,), bias=False)
    (1): BatchNorm1d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (2): ReLU()
  )
  (se2): SqueezeExciteBlock(
    (avg_pool): GAP1d(
      (gap): AdaptiveAvgPool1d(output_size=1)
      (flatten): Flatten(full=False)
    )
    (fc): Sequential(
      (0): Linear(in_features=256, out_features=32, bias=False)
      (1): ReLU()
      (2): Linear(in_features=32, out_features=256, bias=False)
      (3): Sigmoid()
    )
  )
  (convblock3): ConvBlock(
    (0): Conv1d(256, 128, kernel_size=(3,), stride=(1,), padding=(1,), bias=False)
    (1): BatchNorm1d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (2): ReLU()
  )
  (gap): GAP1d(
    (gap): AdaptiveAvgPool1d(output_size=1)
    (flatten): Flatten(full=False)
  )
  (concat): Concat(1)
  (fc): Linear(in_features=228, out_features=12, bias=True)
)