Closed Arsenal591 closed 6 years ago
aselle
commented
6 years ago You can generate a LSTM cell indidivually and manually loop back state (i.e. don't use static_rnn or dynamic_rnn). It will convert the lstm as individual operators. For more help, please provide a simple example that you are trying to run.
tensorflowbutler
commented
6 years ago It has been 14 days with no activity and the awaiting response label was assigned. Is this still an issue?
tofulawrence
commented
6 years ago Automatically closing due to lack of recent activity. Please update the issue when new information becomes available, and we will reopen the issue. Thanks!
robbiemu
commented
5 years ago @aselle I have a simple example to volunteer :) I wish you would take mine as an example! Full source is https://colab.research.google.com/drive/1sB8vv3TM5Sp45TicJ8PxEvpUAX42XqPY and I can gladly provide sample data as well.
Here's the part specific to consrtructing the model:
from keras.models import Sequential
from keras.layers import Dense, LSTM, Dropout
regressor = Sequential()
regressor.add(LSTM(units=nh, return_sequences=True,
input_shape=(timesteps, dim)))
layers = (len(endog) + 1) if len(endog) > 1 else 2 # 4
for i in range(1, layers):
if i < layers - 1:
cell = LSTM(units=nh, return_sequences=True)
else:
cell = LSTM(units=nh)
regressor.add(cell)
regressor.add(Dropout(dropout)) # 0.2
regressor.add(Dense(units=dim))
regressor.compile(optimizer='adam', loss='mean_squared_error')this is a pretty small/simple LSTM for use with categorical and continuous time series data, with, for my data and prediction range, about 4 layers of about 12 nodes + dropout and output.
System information
Source code / logs
See below.
Describe the problem
I'm trying to convert a
.pbgraph into tflite format, which containes a bi-directional LSTM. However, whenever I run the command:It failed with following error messages:
And I assume that LSTM is not currently supported in tflite. So is that the case? And if it's true, is there any way that I can work around this problem?