How to properly stack LSTM and sklearn ML models - keras

Question

I am trying to stack Kears and ML models of sklean library for a regression problem. I am able to stack simple dense layers of Keras with any ML models.
My input type is similar to Boston house price (multi-variete). This code snippet shows how I tackle the problem:
#Scikit-learn Models
rnd_reg =RandomForestRegressor(n_estimators=100, random_state=42)
#Keras Model
def build_nn():
model = Sequential(
[Dense(512, activation='relu', input_shape=[X.shape[1]]),
Dense(256, activation='relu'),
Dropout(0.4),
Dense(128, activation='relu'),
Dense(64, activation='relu'),
Dropout(0.2),
Dense(1, activation='linear')
])
model.compile(optimizer='adam',
loss='mean_squared_error',
metrics=['MeanSquaredError', 'MeanAbsolutePercentageError'])
return model
keras_reg = tf.keras.wrappers.scikit_learn.KerasRegressor(build_nn,
epochs=20,
batch_size=1,
verbose=True)
keras_reg._estimator_type = "regressor"
st_reg = StackingRegressor(
estimators=[('rf', rnd_reg),
('Dense',keras_reg)],
final_estimator=XGBRegressor(random_state=42)
)
Everything works jus fine with this part. The issue is when I want to use RNN (LSTM for example) and stack it with sklearn models. I am not sure if this is due to the wrong input shape that I use or LSTM cannot be stacked with ML models of sklearn.
Below is my LSTM structure:
def build_nn1():
model = Sequential(
#[LSTM(50, activation='relu', batch_input_shape=(None, X.shape[1], 1)),
[LSTM(50, activation='relu', input_shape=[X.shape[1], 1]),
BatchNormalization(),
#Dropout(0.2),
Dense(20, activation='relu'),
Dense(1, activation='relu')
])
model.compile(optimizer=Adam(lr=1e-5),
loss='mean_squared_error',
metrics=['MeanSquaredError', 'MeanAbsolutePercentageError'])
return model
enter code here