Keras is user-friendly, easy to extend modularly.
Keras does not process low-level calculation. Instead, it uses another library.
Neural layers are independent modules that you can combine to create new models with cost functions, optimizers, launch schemes, activation functions and editing schemes. New modules and functions are easy to add. Models are defined in Python code, not separate model configuration files.
Why Use Keras
The biggest reasons for using Keras are primarily due to its principles of being user-friendly. Beyond the ease of learning and the ease of modeling, Keras supports a wide range of export options, wide-ranging adoption.
Keras is supported by Google, Microsoft, Amazon, Apple, Nvidia, Uber.
Keras does not conduct its own low-level operations such as suitable tensor products and convolutions; uses a backend engine for this.
model Core Keras is a data structure. Keras has two main models :
Sequential model and functional API.
Keras Functional API
The Keras Sequential model is simple, but the model topology is limited. Keras is useful for creating multiple input/multiple output models, oriented acelike graphics (DAGs) and complex models as models with common layers such as functional API.
The functional API uses the same layers as the Sequential model, but provides more flexibility in combining them. In the functional API, it first defines the layers, then creates, compiles, and shrinks the model.
Click here for more information