LSTM on EEG data using Keras

[rakiba...@gmail.com]

I am working with EEG data. I want to use the window approach on row signal. My train data size is (1000, 50, 100, 32). 

Where,

1000= sample size/ trial number 

50= number of the window in each trial 

100= sample of time series signal  

32= number of channel.

I want to run LSTM on that signal. Please see the sample code

    from sklearn.model_selection import train_test_split

    import numpy as np

    seed = 7

    np.random.seed(seed)

    X=np.random.rand(1000, 50, 100, 32)

    y=np.random.rand(1000)

    X_train, X_test, y_train, y_test = train_test_split(X, 

    y, test_size=0.33, random_state=seed)

    

    from keras.models import Sequential

    from keras.layers import Dense

    from keras.layers import LSTM

    from keras.layers import Dropout

    model = Sequential()

    model.add(LSTM(50, return_sequences=False, 

    activation='relu',batch_size=5, input_shape=(5,10, 32)))

    model.add(Dropout(0.2))

    model.add(Dense(1, activation='sigmoid'))

    

    model.compile(loss='mae', optimizer='adam')

    print(model.summary())

    history=model.fit(X_train, y_train, validation_data 

    (X_test,y_test), epochs=3, batch_size=10)

I can not run the code. Can you please help me?

[Ted Yu]

With the following change (note the change to input_shape), 

model.add(LSTM(50, return_sequences=True,

  activation='relu',batch_size=5, input_shape=(10, 32)))

I got:

_________________________________________________________________

Layer (type)                 Output Shape              Param #

=================================================================

lstm_1 (LSTM)                (5, 10, 50)               16600

_________________________________________________________________

dropout_1 (Dropout)          (5, 10, 50)               0

_________________________________________________________________

dense_1 (Dense)              (5, 10, 1)                51

=================================================================

Total params: 16,651

Trainable params: 16,651

Non-trainable params: 0

--
You received this message because you are subscribed to the Google Groups "Keras-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to keras-users...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/keras-users/d4967f00-4eb2-4c0d-b376-bde84ce37657%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[rakiba...@gmail.com]

but, if I want to run it. it gives me value error: "Error when checking input: expected lstm_57_input to have 3 dimensions, but got array with shape (670, 50, 100, 32) "

Can you help me to fix the model.fit part

Thanks

[Sergey O.]

The expected input is

(samples,time,channels)

Your descriptions are very vague:

1000= sample size/ trial number 

50= number of the window in each trial 

100= sample of time series signal  

32= number of channel.

Do you mean?

1000 = samples

50 = I'm guessing you are sliding a window of 50 across a time series???

100 = number of windows?

32 = channels

--
You received this message because you are subscribed to the Google Groups "Keras-users" group.

To unsubscribe from this group and stop receiving emails from it, send an email to keras-users+unsubscribe@googlegroups.com.

To view this discussion on the web, visit https://groups.google.com/d/msgid/keras-users/371e9423-a327-4c95-bbc8-d850c384c122%40googlegroups.com.