Common Layers

USAGE

All functions can be imported using layers module

fully_connected(inputs, features_out, name, weights_initializer=None, bias_initializer=None, act_fn=None, weight_decay=None)[source]

Fully Connected Layer (or Dense Layer). It automatically reshapes a tensor to get a 2D input preserving batch size on first dimension [Batch x Features].

Parameters:
  • inputs – Input tensor of any size.
  • features_out – Number of output features, or number of classes if used as output layer.
  • name – Operation name.
  • weights_initializer – Weights initializer function. Default None uses XAVIER.
  • bias_initializer – Bias initializer function. Default None uses XAVIER.
  • act_fn – Activation function handler.
  • weight_decay – L2 decay lambda value.

Returns: Tensor of shape [Batch x Features_Out]