CODEWITHSUNDEEP
pythonmachine-learningkerasneural-networklstm
user218030
user218030

Reputation: 117

How to produce 1 output with multiple input variables LSTM

I'm trying to predict the opening price for the next day. I'm able to get the formatting correct to feed in the input i.e. ('Open','High' columns per day for n time). However when I format into a 3D array my shape is as follows:

(1200, 60, 2)

The X_train has 1200 samples, with 60 timestep (previous 60 days of historical data) and 2 features (open and high)

However, My issue arises when its reaches the keras coding part when implementing layers. This is my code I am using:

regressor.add(LSTM(units = 50, return_sequences = True, input_shape = (X_train.shape[1], 2)))
regressor.add(Dropout(0.2))

regressor.add(LSTM(units = 50, return_sequences = True))
regressor.add(Dropout(0.2))

regressor.add(LSTM(units = 50, return_sequences = True))
regressor.add(Dropout(0.2))

regressor.add(LSTM(units = 50))
regressor.add(Dropout(0.2))

regressor.add(Dense(units = 1))

The problem arises with the last line. I want the output to be only 1. So essentially I want the Open and High values of the input series to be used to work out the final singular output of just the Open price. However, by setting the Dense(units = 1), creates this error:

ValueError: Error when checking target: expected dense_1 to have shape (1,) but got array with shape (2,)

To fix this I have tried to change it to 2 Dense(units=2), however the final output produces 2 lines on the graph one for open and one for high which is not what I want. That's 2 outputs where I want 1. I'm not sure what to do with this scenario.

regressor.summary()

Model: "sequential_1"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
lstm_1 (LSTM)                (None, 60, 50)            10600     
_________________________________________________________________
dropout_1 (Dropout)          (None, 60, 50)            0         
_________________________________________________________________
lstm_2 (LSTM)                (None, 60, 50)            20200     
_________________________________________________________________
dropout_2 (Dropout)          (None, 60, 50)            0         
_________________________________________________________________
lstm_3 (LSTM)                (None, 60, 50)            20200     
_________________________________________________________________
dropout_3 (Dropout)          (None, 60, 50)            0         
_________________________________________________________________
lstm_4 (LSTM)                (None, 50)                20200     
_________________________________________________________________
dropout_4 (Dropout)          (None, 50)                0         
_________________________________________________________________
dense_1 (Dense)              (None, 2)                 102       
=================================================================
Total params: 71,302
Trainable params: 71,302
Non-trainable params: 0

Upvotes: 1

Views: 1578

Answers (1)

Kenan
Kenan

Reputation: 14094

When the dense layer is having a shape error it might be your label tensor that is not matching. Check that y_train has shape [1200, 1] so that you can use a dense of 1.

Upvotes: 1

Related Questions