package tensorflow

  1. Overview
  2. Docs
module Batch_norm : sig ... end
module Update_ops_store : sig ... end
val batch_norm : ?decay:float -> [< `double | `float ] as 'a Node.t -> is_training:[ `bool ] Node.t -> update_ops_store:Update_ops_store.t -> 'a Node.t

batch_norm ?decay node ~is_training takes as input a node which last dimension is assumed to be the feature dimension on which batch norm is computed. When is_training is true the batch statistics from node are used. When false variables storing the running mean and variance are used instead.

type activation =
  1. | Relu
  2. | Softmax
  3. | Tanh
  4. | Leaky_relu of float
  5. | Sigmoid
module Linear : sig ... end
val linear : ?activation:activation -> ?use_bias:bool -> [< `double | `float ] as 'a Node.t -> output_dim:int -> 'a Node.t
type padding =
  1. | Same
  2. | Valid
val max_pool : ?padding:padding -> [< `double | `float ] as 'a Node.t -> ksize:(int * int) -> strides:(int * int) -> 'a Node.t
module Conv2D : sig ... end
val conv2d : ?padding:padding -> ?use_bias:bool -> [< `double | `float ] as 'a Node.t -> ksize:(int * int) -> strides:(int * int) -> output_dim:int -> 'a Node.t
module Conv2DTranspose : sig ... end
val conv2d_transpose : ?padding:padding -> ?use_bias:bool -> [< `double | `float ] as 'a Node.t -> ksize:(int * int) -> strides:(int * int) -> output_filters:int -> 'a Node.t
val flatten : 'a Node.t -> 'a Node.t

flatten preserves the first (batch) dimension.

val reshape : 'a Node.t -> shape:int list -> 'a Node.t