Why Do Two Different Ways of Adding a Drop Out Layer in Tensor Flow Function Differently?

Question

I was trying to add a hidden layer with dropout to my neural network. I found that the following code:

    model.add(
        tf.keras.layers.Dense(256, activation = "relu")
                )

    model.add(
        tf.keras.layers.Dropout(0.5)
        )

resulted in terrible accuracy (~65%) while the following code:

    model.add(
        tf.keras.layers.Dense(256, activation = "relu"),
        tf.keras.layers.Dropout(0.5)
        )

resulted in adequate accuracy (~95%). Can someone explain why these lines of code resulted in vastly different accuracies and which one is the proper way to add a hidden layer with dropout?

More generally, I was wondering if adding a neural network's layers one at a time is different than instantiating the neural network with one line, and if so why?

    model = tf.keras.models.Sequential()
    model.add(...) # Add one layer at a time
    ...

VS

    model = tf.keras.models.Sequential([... (complete neural network) ...])

Answer 1

According to https://www.tensorflow.org/api_docs/python/tf/keras/Sequential#add, model.add only takes one argument for layer, so your code

model.add(
        tf.keras.layers.Dense(256, activation = "relu"),
        tf.keras.layers.Dropout(0.5)
        )

effectively only add the dense layer and pass in a second argument True which is the default. If you compare it with

model.add(
        tf.keras.layers.Dense(256, activation = "relu"),
        True
        )

The results should be the same.