Module `Ocannl.Operation`Source

Computational primitives for neural networks, integrating Tensor with Assignments.

Sourcemodule Asgns = Arrayjit.Assignments

Sourcemodule Idx = Arrayjit.Indexing

Sourcemodule At : sig ... end

Sourcemodule Initial_NTDSL : sig ... end

Sourcemodule Initial_TDSL : sig ... end

val add : 
  ?label:Base.string list ->
  ?grad_spec:Tensor.grad_spec ->
  Tensor.t ->
  Tensor.t ->
  Tensor.t

Source

val sub : 
  ?label:Base.string list ->
  ?grad_spec:Tensor.grad_spec ->
  Tensor.t ->
  Tensor.t ->
  Tensor.t

Source

val mul : 
  Shape.compose_type ->
  op_asn:
    (v:Tensor.tn ->
      t1:Tensor.t ->
      t2:Tensor.t ->
      projections:Tensor.projections Base.Lazy.t ->
      Tensor.asgns) ->
  label:Base.string Base.list ->
  ?grad_spec:Tensor.grad_spec ->
  Tensor.t ->
  Tensor.t ->
  Tensor.t

Source

val pointmul : 
  ?label:Base.string list ->
  ?grad_spec:Tensor.grad_spec ->
  Tensor.t ->
  Tensor.t ->
  Tensor.t

Source

val matmul : 
  ?label:Base.string list ->
  ?grad_spec:Tensor.grad_spec ->
  Tensor.t ->
  Tensor.t ->
  Tensor.t

Source

val einsum : 
  ?label:Base.string list ->
  Base.string ->
  ?grad_spec:Tensor.grad_spec ->
  Tensor.t ->
  Tensor.t ->
  Tensor.t

Similar to the explicit mode of numpy.einsum, the binary variant. Can compute various forms of matrix multiplication, inner and outer products, etc.

Note that "a,b->c" from numpy is "a;b=>c" in OCANNL, since "->" is used to separate the input and the output axes.

Source

val outer_sum : 
  ?label:Base.string list ->
  Base.string ->
  ?grad_spec:Tensor.grad_spec ->
  Tensor.t ->
  Tensor.t ->
  Tensor.t

Like einsum, but adds instead than multiplying the resulting values.

Source

val einsum1 : 
  ?label:Base.string list ->
  Base.string ->
  ?grad_spec:Tensor.grad_spec ->
  Tensor.t ->
  Tensor.t

Similar to the explicit mode of numpy.einsum, the unary variant. Can permute axes, extract diagonals, compute traces etc.

Note that "a->c" from numpy is "a=>c" in OCANNL, since "->" is used to separate the input and the output axes.

Source

val relu : 
  ?label:Base.string list ->
  ?grad_spec:Tensor.grad_spec ->
  Tensor.t ->
  Tensor.t

Sourcemodule NDO_without_pow : sig ... end

Source

val pointpow : 
  ?label:Base.string Base.list ->
  grad_spec:Tensor.grad_spec ->
  Base.float ->
  Tensor.t ->
  Tensor.t

Sourcemodule NDO_without_div : sig ... end

Source

val pointdiv : 
  ?label:Base.string Base.list ->
  grad_spec:Tensor.grad_spec ->
  Tensor.t ->
  Tensor.t ->
  Tensor.t

Source

val range : 
  ?label:Base.string list ->
  ?grad_spec:Tensor.grad_spec ->
  ?axis_label:Base.string ->
  Base.Int.t ->
  Tensor.t

Source

val range_of_shape : 
  ?label:Base.string list ->
  ?grad_spec:Tensor.grad_spec ->
  ?batch_dims:Base.Int.t Base.List.t ->
  ?input_dims:Base.Int.t Base.List.t ->
  ?output_dims:Base.Int.t Base.List.t ->
  ?batch_axes:(Base.string * Base.Int.t) Base.List.t ->
  ?input_axes:(Base.string * Base.Int.t) Base.List.t ->
  ?output_axes:(Base.string * Base.Int.t) Base.List.t ->
  unit ->
  Tensor.t

Sourceval stop_gradient : ?label:Base.string list -> Tensor.t -> Tensor.t

A stop_gradient is an identity in the forward pass and a no-op in the backprop pass.

Source

val slice : 
  ?label:Base.string list ->
  grad_spec:Tensor.grad_spec ->
  Idx.static_symbol ->
  Tensor.t ->
  Tensor.t

Source

val embed_symbol : 
  ?label:Base.string list ->
  Arrayjit.Indexing.static_symbol ->
  Tensor.t

Sourcemodule DO : sig ... end

Sourcemodule NDO : sig ... end

Sourcemodule TDSL : sig ... end

Sourcemodule NTDSL : sig ... end

package neural_nets_lib

Module Ocannl.OperationSource

Module `Ocannl.Operation`Source