keras RNN model (TimeDistributed wrapper)

kerasRNN_model.py

#build model many to one
embedding_vector_length = 32
model = Sequential()
model.add(Embedding(input_dim=top_words, 
                    output_dim=embedding_vector_length, 
                    input_length=max_review_length))
model.add(LSTM(100, return_sequences=True))
model.add(Dropout(0.2))
model.add(LSTM(100, return_sequences=True))
model.add(Dropout(0.2))
model.add(TimeDistributed(Dense(1,activation='sigmoid')))
model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])
print(model.summary())

'''
____________________________________________________________________________________________________
Layer (type)                     Output Shape          Param #     Connected to                     
====================================================================================================
embedding_15 (Embedding)         (None, 500, 32)       160000      embedding_input_15[0][0]         
____________________________________________________________________________________________________
lstm_28 (LSTM)                   (None, 500, 100)      53200       embedding_15[0][0]               
____________________________________________________________________________________________________
dropout_28 (Dropout)             (None, 500, 100)      0           lstm_28[0][0]                    
____________________________________________________________________________________________________
lstm_29 (LSTM)                   (None, 500, 100)      80400       dropout_28[0][0]                 
____________________________________________________________________________________________________
dropout_29 (Dropout)             (None, 500, 100)      0           lstm_29[0][0]                    
____________________________________________________________________________________________________
timedistributed_7 (TimeDistribut (None, 500, 1)        101         dropout_29[0][0]                 
====================================================================================================
Total params: 293701
____________________________________________________________________________________________________
None

RNN outputs not flatten (timedistributed_7 layer has (Nonex500x1)

Why keep each timestep values separate? Because:

you're only want to interacting the values between its own timestep
you don't want to have a random interaction between different timesteps and channels.

*R.Luthfianto. Prediksi Struktur Sekunder Protein menggunakan Konvolusi dan Bidirectional Gated Recurrent Unit. Undergraduate thesis, Universitas Gadjah Mada, Yogyakarta, 2016.
'''