LSTM autoencoder with Keras data shape issue

Question

I am trying to make a model using Keras with LSTM autoencoder. Here what I have tried

data = df.values
timesteps = 10
dim = data.shape[1]
samples = data.shape[0]
data.shape = (int(samples/timesteps),timesteps,dim)

and then

model = Sequential()
model.add(LSTM(50,input_shape=(timesteps,dim),return_sequences=True))
model.add(LSTM(50,input_shape=(timesteps,dim),return_sequences=True))
model.add(LSTM(50,input_shape=(timesteps,dim),return_sequences=True))
model.add(LSTM(50,input_shape=(timesteps,dim),return_sequences=True))
model.add(LSTM(50,input_shape=(timesteps,dim),return_sequences=True))
model.add(LSTM(50,input_shape=(timesteps,dim),return_sequences=True))
model.add(LSTM(50,input_shape=(timesteps,dim),return_sequences=True))
model.add(LSTM(50,input_shape=(timesteps,dim),return_sequences=True))
model.add(LSTM(50,input_shape=(timesteps,dim),return_sequences=True))
model.add(LSTM(50,input_shape=(timesteps,dim),return_sequences=True))
model.add(LSTM(50,input_shape=(timesteps,dim),return_sequences=True))
model.compile(loss='mae', optimizer='adam')

this is my model fit

model.fit(data, data, epochs=50, batch_size=72, validation_data=(data, data), verbose=0, shuffle=False)

This is the error message I am getting

ValueError: Error when checking target: expected lstm_33 to have shape (None, 10, 50) but got array with shape (711, 10, 1)

How can I fix this ?

I have only I data set

Update

input data shape I have = (7110, 1)

This is an Univariate time series data

On which layer you fail? the first? also post your data.shape, you define dim = data.shape[1] but then use it as data.shape[2]? — Dinari
– Dinari, Commented Dec 18, 2018 at 19:46
This isn't specific to your question, but I noticed your error says lstm_33, so that means it is keeping all of your layers from previous runs. If you run from keras import backend then backend.clear_session() between model builds, then you'll start with a fresh graph and it would start from lstm, lstm_1, etc. — Chris Farr
– Chris Farr, Commented Dec 18, 2018 at 21:37

Kurtis Streutker · Accepted Answer · 2018-12-18 20:23:47Z

3

The error is caused by specifying input_shape=(timesteps,dim) for all the layers. You only need to do this for the first layer and the rest will be inferred by the previous layer. What is happening is you are overriding the input shape which is causing the error.

answered Dec 18, 2018 at 20:23

Kurtis Streutker

1,35413 silver badges15 bronze badges

Sign up to request clarification or add additional context in comments.

1 Comment

TheTechGuy Over a year ago

Thanks, I just fixed it

Collectives™ on Stack Overflow

LSTM autoencoder with Keras data shape issue

Update

1 Answer 1

1 Comment

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

Update

1 Answer 1

1 Comment

Your Answer

Sign up or log in

Post as a guest

Related