Input is the Keras constructor that declares what a model will receive before real data is available. It creates a symbolic tensor: an object carrying input shape and dtype information, but no actual values. Layers can consume this object to build a model graph, making Input especially useful when a model has multiple inputs, multiple outputs, or a nonlinear topology. section 7.2.2
A two-layer Functional model
inputs = keras.Input(shape=(3,), name="my_input") features = layers.Dense(64, activation="relu")(inputs) outputs = layers.Dense(10, activation="softmax")(features) model = keras.Model(inputs=inputs, outputs=outputs, name="my_functional_model") Here, shape=(3,) describes each input example, while the symbolic tensor carries a batch-inclusive shape of (None, 3) and a float32 dtype. Each layer returns another symbolic tensor with updated specifications, and Model connects the declared input to the final output. section 7.2.2
Use Input when you need to:
- Declare a model's input shape in advance so a Sequential model builds immediately and summary() can show layer shapes while layers are added. section 7.2.1
- Name input objects for clearer model graphs and for matching named input arrays when training multi-input models. section 7.2.2
- Declare several inputs, such as title, text body, and tags, then connect them through shared layers to multiple outputs. section 7.2.2
Listing 7.8 A simple Functional model with two Dense layers
inputs = keras.Input(shape=(3,), name="my_input") features = layers.Dense(64, activation="relu")(inputs) outputs = layers.Dense(10, activation="softmax")(features) model = keras.Model(inputs=inputs, outputs=outputs, name="my_functional_model")
Inspect the complete minimal example: Input declares the symbolic starting point, Dense layers transform it, and Model binds inputs to outputs. section 7.2.2
