MissingInputError under theano, but runs fine under tensorflow

[billshoo]

The following Keras model runs smoothly under tensorflow, but gives "theano.gof.fg.MissingInputError" on the last line under theano (X and y can be any dummy data)

    X = np.random.rand(100000, 320)
    y = np.random.rand(100000, 1)
    model = Sequential()
    model.add(TimeDistributed(Dropout(0.5), batch_input_shape=(1, 100000, 320)))
    model.add(TimeDistributed(Dense(1000)))
    model.add(TimeDistributed((Dropout(0.5))))
    model.add(LSTM(1000, batch_input_shape=(1, 100000, 1000), 
                   return_sequences=True, stateful=True))
    model.add(TimeDistributed((Dropout(0.5))))
    model.add(TimeDistributed(Dense(1)))
    opti = RMSprop(lr=0.001)
    model.compile(optimizer=opti, loss='mse')
    model.train_on_batch(X[np.newaxis, 0:100000, :], y[np.newaxis, 0:100000])