Help: RNN and LSTM multiple features with and without timesteps

Hello everyone,

I got the following problem. I got input data with different shapes 600 samples of 4 features with just 1 value per feature (600, 4) and 600 samples of 1 feature with each value of the shape (289, 4) so (600, 289, 4) lets call it feature5. I want to train a model with all 5 features. The values in feature5 are weather data over the last 24 hours each 5 minutes (=289). That’s the reason why I thought I might have to use an LSTM layer. The problem is the other 4 features don’t have time related data. So I thought I need a model for two input datasets that’s why I use the code from the tensorflow webpage for RNN model with nested input output.

class NestedCell(keras.layers.Layer):
    def __init__(self, unit_1, unit_2, unit_3, **kwargs):
        self.unit_1 = unit_1
        self.unit_2 = unit_2
        self.unit_3 = unit_3
        self.state_size = [tf.TensorShape([unit_1]), tf.TensorShape([unit_2, unit_3])]
        self.output_size = [tf.TensorShape([unit_1]), tf.TensorShape([unit_2, unit_3])]
        super(NestedCell, self).__init__(**kwargs)

    def build(self, input_shapes):
        # expect input_shape to contain 2 items, [(batch, i1), (batch, i2, i3)]
        i1 = input_shapes[0][1]
        i2 = input_shapes[1][1]
        i3 = input_shapes[1][2]

        self.kernel_1 = self.add_weight(
            shape=(i1, self.unit_1), initializer="uniform", name="kernel_1"
        self.kernel_2_3 = self.add_weight(
            shape=(i2, i3, self.unit_2, self.unit_3),

    def call(self, inputs, states):
        # inputs should be in [(batch, input_1), (batch, input_2, input_3)]
        # state should be in shape [(batch, unit_1), (batch, unit_2, unit_3)]
        input_1, input_2 = tf.nest.flatten(inputs)
        s1, s2 = states

        output_1 = tf.matmul(input_1, self.kernel_1)
        output_2_3 = tf.einsum("bij,ijkl->bkl", input_2, self.kernel_2_3)
        state_1 = s1 + output_1
        state_2_3 = s2 + output_2_3

        output = (output_1, output_2_3)
        new_states = (state_1, state_2_3)

        return output, new_states

    def get_config(self):
        return {"unit_1": self.unit_1, "unit_2": unit_2, "unit_3": self.unit_3}

unit_1 = 10
unit_2 = 20
unit_3 = 30

i1 = 32
i2 = 64
i3 = 32
batch_size = 64
num_batches = 10
timestep = 50

cell = NestedCell(unit_1, unit_2, unit_3)
rnn = keras.layers.RNN(cell)

input_1 = keras.Input((None, i1))
input_2 = keras.Input((None, i2, i3))

outputs = rnn((input_1, input_2))

model = keras.models.Model([input_1, input_2], outputs)

model.compile(optimizer="adam", loss="mse", metrics=["accuracy"])

input_1_data = np.random.random((batch_size * num_batches, timestep, i1))
input_2_data = np.random.random((batch_size * num_batches, timestep, i2, i3))
target_1_data = np.random.random((batch_size * num_batches, unit_1))
target_2_data = np.random.random((batch_size * num_batches, unit_2, unit_3))
input_data = [input_1_data, input_2_data]
target_data = [target_1_data, target_2_data], target_data, batch_size=batch_size)

Can I use RNN where some features don’t have any time related data? How can I make the code above work for my problem? Or is there maybe a better solution for my problem?

All help will be much appreciated.

Cheers, Mo

Hi @Moritz One would typically use RNN with sequence data. Still, one can always creates a metal model that is the combination of different models, each sub-model type being more “natural” for your different types of data.