package caisar-ir

You can search for identifiers within the package.

in-package search v0.2.0

caisar-ir
- Ir
  - Nier_cfg
    
    Edge
    
    MakeVertex
    
    I
    
    NierCFG
    
    E
    
    I
    
    V
    
    NierCFGFloat
    
    E
    
    V
    
    Node
    
    Tensor
    
    VInput

Legend:
Library
Module
Module type
Parameter
Class
Class type

include Graph.Sig.I
  with type V.t = (Base.float, Stdlib.Bigarray.float64_elt) Node.t
   and type V.label = (Base.float, Stdlib.Bigarray.float64_elt) Node.t
   and type E.t =
         (Base.float, Stdlib.Bigarray.float64_elt) Node.t
         * Edge.t
         * (Base.float, Stdlib.Bigarray.float64_elt) Node.t
   and type E.label = Edge.t

An imperative graph is a graph.

include Graph.Sig.G
  with type V.t = (Base.float, Stdlib.Bigarray.float64_elt) Node.t
  with type V.label = (Base.float, Stdlib.Bigarray.float64_elt) Node.t
  with type E.t =
         (Base.float, Stdlib.Bigarray.float64_elt) Node.t
         * Edge.t
         * (Base.float, Stdlib.Bigarray.float64_elt) Node.t
  with type E.label = Edge.t

Graph structure

type t

Abstract type of graphs

module V : 
  Graph.Sig.VERTEX
    with type t = (Base.float, Stdlib.Bigarray.float64_elt) Node.t
    with type label = (Base.float, Stdlib.Bigarray.float64_elt) Node.t

Vertices have type V.t and are labeled with type V.label (note that an implementation may identify the vertex with its label)

type vertex = V.t

module E : 
  Graph.Sig.EDGE
    with type vertex = vertex
    with type t =
           (Base.float, Stdlib.Bigarray.float64_elt) Node.t
           * Edge.t
           * (Base.float, Stdlib.Bigarray.float64_elt) Node.t
    with type label = Edge.t

Edges have type E.t and are labeled with type E.label. src (resp. dst) returns the origin (resp. the destination) of a given edge.

type edge = E.t

val is_directed : bool

Is this an implementation of directed graphs?

Size functions

val is_empty : t -> bool

val nb_vertex : t -> int

val nb_edges : t -> int

Degree of a vertex

val out_degree : t -> vertex -> int

out_degree g v returns the out-degree of v in g.

raises Invalid_argument
if v is not in g.

val in_degree : t -> vertex -> int

in_degree g v returns the in-degree of v in g.

raises Invalid_argument
if v is not in g.

Membership functions

val mem_vertex : t -> vertex -> bool

val mem_edge : t -> vertex -> vertex -> bool

val mem_edge_e : t -> edge -> bool

val find_edge : t -> vertex -> vertex -> edge

find_edge g v1 v2 returns the edge from v1 to v2 if it exists. Unspecified behaviour if g has several edges from v1 to v2.

raises Not_found
if no such edge exists.

val find_all_edges : t -> vertex -> vertex -> edge list

find_all_edges g v1 v2 returns all the edges from v1 to v2.

since ocamlgraph 1.8

Successors and predecessors

You should better use iterators on successors/predecessors (see Section "Vertex iterators").

val succ : t -> vertex -> vertex list

succ g v returns the successors of v in g.

raises Invalid_argument
if v is not in g.

val pred : t -> vertex -> vertex list

pred g v returns the predecessors of v in g.

raises Invalid_argument
if v is not in g.

Labeled edges going from/to a vertex

val succ_e : t -> vertex -> edge list

succ_e g v returns the edges going from v in g.

raises Invalid_argument
if v is not in g.

val pred_e : t -> vertex -> edge list

pred_e g v returns the edges going to v in g.

raises Invalid_argument
if v is not in g.

Graph iterators

val iter_vertex : (vertex -> unit) -> t -> unit

Iter on all vertices of a graph.

val fold_vertex : (vertex -> 'a -> 'a) -> t -> 'a -> 'a

Fold on all vertices of a graph.

val iter_edges : (vertex -> vertex -> unit) -> t -> unit

Iter on all edges of a graph. Edge label is ignored.

val fold_edges : (vertex -> vertex -> 'a -> 'a) -> t -> 'a -> 'a

Fold on all edges of a graph. Edge label is ignored.

val iter_edges_e : (edge -> unit) -> t -> unit

Iter on all edges of a graph.

val fold_edges_e : (edge -> 'a -> 'a) -> t -> 'a -> 'a

Fold on all edges of a graph.

val map_vertex : (vertex -> vertex) -> t -> t

Map on all vertices of a graph.

The current implementation requires the supplied function to be injective. Said otherwise, map_vertex cannot be used to contract a graph by mapping several vertices to the same vertex. To contract a graph, use instead create, add_vertex, and add_edge.

Vertex iterators

Each iterator iterator f v g iters f to the successors/predecessors of v in the graph g and raises Invalid_argument if v is not in g. It is the same for functions fold_* which use an additional accumulator.

<b>Time complexity for ocamlgraph implementations:</b> operations on successors are in O(1) amortized for imperative graphs and in O(ln(|V|)) for persistent graphs while operations on predecessors are in O(max(|V|,|E|)) for imperative graphs and in O(max(|V|,|E|)*ln|V|) for persistent graphs.

iter/fold on all successors/predecessors of a vertex.

val iter_succ : (vertex -> unit) -> t -> vertex -> unit

val iter_pred : (vertex -> unit) -> t -> vertex -> unit

val fold_succ : (vertex -> 'a -> 'a) -> t -> vertex -> 'a -> 'a

val fold_pred : (vertex -> 'a -> 'a) -> t -> vertex -> 'a -> 'a

iter/fold on all edges going from/to a vertex.

val iter_succ_e : (edge -> unit) -> t -> vertex -> unit

val fold_succ_e : (edge -> 'a -> 'a) -> t -> vertex -> 'a -> 'a

val iter_pred_e : (edge -> unit) -> t -> vertex -> unit

val fold_pred_e : (edge -> 'a -> 'a) -> t -> vertex -> 'a -> 'a

val create : ?size:int -> unit -> t

create () returns an empty graph. Optionally, a size can be given, which should be on the order of the expected number of vertices that will be in the graph (for hash tables-based implementations). The graph grows as needed, so size is just an initial guess.

val clear : t -> unit

Remove all vertices and edges from the given graph.

since ocamlgraph 1.4

val copy : t -> t

copy g returns a copy of g. Vertices and edges (and eventually marks, see module Mark) are duplicated.

val add_vertex : t -> vertex -> unit

add_vertex g v adds the vertex v to the graph g. Do nothing if v is already in g.

val remove_vertex : t -> vertex -> unit

remove g v removes the vertex v from the graph g (and all the edges going from v in g). Do nothing if v is not in g.

<b>Time complexity for ocamlgraph implementations:</b> O(|V|*ln(D)) for unlabeled graphs and O(|V|*D) for labeled graphs. D is the maximal degree of the graph.

val add_edge : t -> vertex -> vertex -> unit

add_edge g v1 v2 adds an edge from the vertex v1 to the vertex v2 in the graph g. Add also v1 (resp. v2) in g if v1 (resp. v2) is not in g. Do nothing if this edge is already in g.

val add_edge_e : t -> edge -> unit

add_edge_e g e adds the edge e in the graph g. Add also E.src e (resp. E.dst e) in g if E.src e (resp. E.dst e) is not in g. Do nothing if e is already in g.

val remove_edge : t -> vertex -> vertex -> unit

remove_edge g v1 v2 removes the edge going from v1 to v2 from the graph g. If the graph is labelled, all the edges going from v1 to v2 are removed from g. Do nothing if this edge is not in g.

raises Invalid_argument
if v1 or v2 are not in g.

val remove_edge_e : t -> edge -> unit

remove_edge_e g e removes the edge e from the graph g. Do nothing if e is not in g.

raises Invalid_argument
if E.src e or E.dst e are not in g.

val init_cfg : t

val vertex_list : t -> vertex Base.list

val preds : t -> vertex -> vertex Base.list

preds_names g v returns a list of names of predecessors nodes

val preds_names : t -> vertex -> Base.string Base.list

val succs : t -> vertex -> vertex Base.list

succs_names g v returns a list of names of predecessors nodes

val succs_names : t -> vertex -> Base.string Base.list

input_node g returns the nodes considered as describing the inputs of the neural network.

val input_nodes : t -> vertex Base.list

val find_vertices : t -> (vertex -> Base.bool) -> vertex Base.list

data_node_of n returns one node containing a tensor * data among the predecessors of n

val data_node_of : vertex -> t -> vertex Base.option

infer_shape g n sh o_b returns the inferred shape of the output of node n in NIER g with input shape sh. Shape inference is made using the node operator and its predecessors shapes. o_b is true when performing backward propagation, to choose which matrix size to consider.

val infer_shape : 
  t ->
  vertex ->
  Node.shape ->
  on_backward:Base.bool ->
  Node.shape