Updated at: 25 Mar 2019 12:32:22

The core data structure of Keras is a model, which is used organise layers.

General steps to training: 1. Create model (by adding layers) 2. Compile model to configure learning process .compile 3. Fit model by iterating on training data in batches, or manually train 4. Evaluate performance, or generate predictions .evaluate, predict

Model Types

Sequential Model Linear stack of layers.

  1. Layers

Created by passing into constructor or by adding layers using the .add method.

  1. Specifying the Input Shape

The model needs to know the input shape that it should expect.

  1. Compilation

Receives three arguments: an optimizer, a loss- function, list of metrics



All Keras layers have a number of common methods

  1. .get_weights: return the weights of the layer as a list of Numpy arrays
  2. .set_weights: sets the weights of the layer from a list of Numpy arrays
  3. .get_config: returns a dictionary containing the configuration of the layer

input and output tensors can also be retrieved easily.

Batch Normalization

Batch normalization

General steps

  1. Preprocessing


Static == 0(test mode) Static == 1(train mode)