search

Element-Research / rnn

Recurrent Neural Network library for Torch7's nn
BSD 3-Clause "New" or "Revised" License
938 stars 314 forks source link

rnn/Sequencer.lua:90: attempt to index field 'tableoutput' (a nil value) #420

Open tim9510019 opened 7 years ago

tim9510019 commented 7 years ago

I tried to evaluate the LSTM model, but the error message "nn/Sequencer.lua:90: attempt to index field 'tableoutput' (a nil value) " always shows up.

[net] loading model ../uni_image_np_50.t7
nn.Sequencer @ nn.Recursor @ nn.MaskZero @ nn.Sequential { [input -> (1) -> (2) -> (3) -> (4) -> (5) -> (6) -> (7) -> (8) -> (9) -> (10) -> (11) -> (12) -> (13) -> (14) -> (15) -> (16) -> (17) -> (18) -> (19) -> (20) -> (21) -> output] (1): cudnn.SpatialConvolution(4 -> 32, 3x3, 2,2) (2): nn.SpatialBatchNormalization (4D) (32) (3): cudnn.ReLU (4): cudnn.SpatialConvolution(32 -> 64, 3x3, 2,2) (5): nn.SpatialBatchNormalization (4D) (64) (6): cudnn.ReLU (7): nn.SpatialDropout(0.400000) (8): cudnn.SpatialConvolution(64 -> 128, 3x3, 2,2) (9): nn.SpatialBatchNormalization (4D) (128) (10): cudnn.ReLU (11): nn.SpatialDropout(0.400000) (12): nn.Reshape(1152) (13): nn.Linear(1152 -> 512) (14): nn.BatchNormalization (2D) (512) (15): cudnn.ReLU (16): nn.Dropout(0.5, busy) (17): nn.Linear(512 -> 512) (18): nn.LSTM(512 -> 512) (19): nn.Dropout(0.5, busy) (20): nn.Linear(512 -> 13) (21): cudnn.LogSoftMax } /home/timchen/torch/install/bin/luajit: /home/timchen/torch/install/share/lua/5.1/rnn/Sequencer.lua:90: attempt to index field 'tableoutput' (a nil value) stack traceback: /home/timchen/torch/install/share/lua/5.1/rnn/Sequencer.lua:90: in function 'forward' ./net/rnntrain.lua:28: in function 'batchEval' ./net/train.lua:24: in function 'epochEval' ./net/train.lua:46: in function 'train' net/main.lua:45: in main chunk [C]: in function 'dofile' ...chen/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:150: in main chunk [C]: at 0x00405d50

tastyminerals commented 7 years ago

tableoutput gets filled when Sequencer:updateOutput method is called, during forward pass. If that is not happening then I would check my model input first and see if I am actually feeding data into it.

Acciwu commented 2 years ago

I am reproducing the code of [deep-soli radar], the same problem you encountered, if rnn uses version [torch/rnn] and appears [attempt to call field 'Recurrent' (a nil value)], replace it with [ Element-Research/rnn] has the same problem as you [rnn/Sequencer.lua:90: attempt to index field 'tableoutput' (a nil value) ]

Did you solve this problem please?